Typology Eva Van Lier

Parts of Speech and Dependent Clauses

A typological study
Published by

LOT phone +31 (0)30 253 6006

Janskerkhof 13 fax +31 (0)30 253 6406
3512 BL Utrecht e-mail [email protected]
The Netherlands http:/ / wwwlot.let.uu.nl
Cover design
Stien Eelsingh 1903-1964, Schaftende vrouwen op het land, 1955-1956,
55 x 70 cm. p.c., inv. SES E31.
Layout
Claartje Sweere
isbn 978-90-78328-95-7
nur 616
Copyright © 2009 by Eva van Lier. All rights reserved.

Parts of Speech and Dependent Clauses
A typological study
academisch proefschrift
ter verkrijging van de graad van doctor

aan de Universiteit van Amsterdam
op gezag van de Rector Magnificus
prof. dr. D.C. van den Boom
ten overstaan van een door het college van promoties ingestelde
commissie, in het openbaar te verdedigen in de Agnietenkapel
op vrijdag 25 september 2009, te 12.00 uur
door
Eva Helena van Lier

geboren te Den Haag
Promotiecommissie
Promotor
prof. dr. P.C. Hengeveld
Overige leden
prof. dr. J.B. den Besten
dr. S. Cristofaro
dr. J. Don
prof. dr. O.C.M. Fischer
prof. dr. H. de Hoop
dr. A. Malchukov
Faculteit der Geesteswetenschappen

Contents
Acknowledgments vii
List of Figures xi
List of Tables xiii
List of Abbreviations xix
1 Introduction 1
2 Parts of Speech 7
2.1 Introduction 7
2.2 Functional-typological attempts at defining parts of speech 8
2.3 Hengeveld’s syntactic-pragmatic approach 12
2.3.1 Slots and fillers 12
2.3.2 The cross-linguistic perspective 15
2.3.2.1 General introduction 15
2.3.2.2 The Parts of Speech Hierarchy (Hengeveld 1992) 17
2.3.2.3 Intermediate systems(Hengeveld, Rijkhoff & Siewierska 2004) 19
2.3.2.4 More diversity (Hengeveld & Van Lier 2008, forthcoming) 22
2.3.3 Summary 33
2.4 Critique on the syntactic approach 33
2.4.1 Introduction 33
2.4.2 Croft’s Universal-Typological Theory of Parts of Speech 34
2.4.3 Lexical categories within and across languages: 37
The generality and the subclass problems
2.4.4 Summary, outlook 41
2.5 Defining lexical flexibility 42
2.5.2 The criterion of compositionality: flexibility and semantic shift 42
2.5.2.1 Evans and Osada’s proposal and some alternatives 42
2.5.2.2 An integrative approach to semantics in flexible languages 47
2.5.3 The criterion of exhaustiveness: 59
How flexible should a flexible language be?
2.5.4 The criterion of equivalent combinatorics: Bi-directional flexibility 59
2.5.5 Lexical flexibility: A summary 62
2.6 Summary 62
3 Dependent Clauses 65
3.1 Introduction 65
3.2 Dependent clauses and propositional functions 67
3.2.1 Slots and fillers 67
3.2.2 Structural coding in dependent clauses: 69
Subordinating conjunctions and dependent verb forms
3.2.3 Rigid and flexible dependent clauses 73
3.2.4 Summary 78
3.3 Behavioural potential in dependent clauses: 79
De-categorization and re-categorization
3.3.2 Verbal and nominal feature hierarchies 80
3.3.2.1 Introduction 80
3.3.2.2 The verbal feature hierarchy 80
3.3.2.3 The nominal feature hierarchy 83
3.3.2.4 Summary, outlook 86
3.3.3 Feature mixing in DCs across languages: 86
Cross-linguistic generalizations and functional motivations
3.3.3.2 The typology of subordination (Cristofaro 2003) 88
3.3.3.3 The typology of nominalization (Malchukov 2004) 91
3.3.4 Summary, outlook 97
3.4 A two-step typology of dependent clauses 98
3.5 Summary 108
4 Sampling, Research Questions, Method 109
4.1 Introduction 109
4.2 The language sample 109
4.3 Research Questions 114
4.3.2 Global functional matching 115
4.3.3 Global matching including differentiation for structural DC type 115
4.3.4 Specific functional matching 118
4.3.5 Specific matching including differentiation for structural DC type 121
4.3.6 Summary 122
4.4 Method 123
4.5 Outlook 129
5 Parts of Speech in the Languages of the Sample 131

5.2 Overview of the language data 132
5.3 Results in relation to implicational map of Parts of Speech 137
5.3.2 Predicted and attested ‘pure’ systems 138
5.3.3 Predicted and attested ‘intermediate’ systems 141
5.3.4 Predicted but not attested systems 145
(and attested but not predicted systems)
5.3.5 Summary 150
5.4 The generality and the subclass problems in practice 150
5.4.1 The generality problem 150
5.4.2 The subclass problem 155
5.4.2.1 Fuzzy boundaries 155
5.4.2.2 Restricted classes 158
5.5 Non-verbal predication 162
5.6 Identifying flexibility 166
5.6.2 Languages with contentives 166
5.6.2.1 Semantic compositionality 166
5.6.2.2 Exhaustiveness 173
5.6.2.3 Equivalent combinatorics 175
5.6.3 Non-verbs, nominals, and modifiers 177
5.6.3.1 Non-verbs 178
5.6.3.2 Nominals 180
5.6.3.3 Modifiers 183
5.6.4 Summary 188
5.7 Conclusion, outlook 185
6 Dependent Clauses in the Languages of the Sample 187

6.2 DC constructions and propositional functions 187
6.2.2 Rigid and flexible DCs: Attested types 188
6.2.2.1 Preliminary considerations 188
6.2.2.2 Language data 196
6.2.3 Discussion 204
6.2.3.1 Attested and predicted types 204
6.2.3.2 Predicted but not attested types: 211
The problem of predicate clauses
6.2.4 Summary 216
6.3 DC constructions and behavioural potential 216
6.3.2 Basic Data 217
6.3.2.1 Formal parameters 217
6.3.3 A three-way typology of DCs according to behavioural potential 229
6.3.3.2 Preliminary considerations 230
6.3.4 Behavioural potential of different structural DC types 242
6.4 An integrated DC typology: Functional possibilities of three
structural DC types 251
6.5 Summary, outlook 259
7 Results 261
7.2 Global functional matches: 262
Flexibility versus rigidity in the PoS and DC domains
7.2.1 Global matching without differentiating for structural DC types 262
7.2.1.2 Global match for flexible constructions 262
7.2.1.3 Global match for rigid constructions 264
7.2.1.4 Summary 264
7.2.2 The parameter of structural DC type 264
7.2.2.2 Global match for flexible constructions, 265
differentiating for structural DC type
7.2.2.3 Global match for rigid constructions,
differentiating for structural DC type 269
7.2.2.4 Summary 277
7.3 Specific functional matches: 278
Types and amounts of flexibility / rigidity in the PoS and DC domains
7.3.2 Specific matches for flexible constructions 278
7.3.2.2 Flexible match 1: Lexical contentives / non-verbs and contentive /
multi-functional clauses 280
7.3.2.3 Flexible match 2: Nominals and nominal clauses 284
7.3.2.4 Flexible match 3: Modifiers and modifier clauses 287
7.3.2.5 Summary 288
7.3.3 Specific matches for rigid constructions 288
7.3.3.2 Rigid match 1: Nouns and complement clauses 289
7.3.3.3 Rigid match 2: Adjectives and relative clauses 291
7.3.3.4 Rigid match 3: Manner adverbs and adverbial manner clauses 295
7.3.3.5 Summary 297
7.4 Summary, Conclusion 297
7.5 Language data 298
8 Discussion – Flexibility and Functional Transparency 313

8.2 Theoretical background: 314
Trade-off effects between flexible and rigid constructions
8.3 Applying the complexity hypothesis to PoS and deranked DCs 320
8.3.1 Hypothesis 320
8.3.2 Results 322
8.3.2.1 Distribution of deranked DCs in languages with
flexible PoS systems 322
8.3.2.2 Distribution of deranked DCs in languages with
rigid PoS systems 331
8.3.3 Summary 334
8.4 Balanced constructions: Functional transparency and other factors 336
8.4.2 Functional transparency 336
8.4.2.1 Flexible balanced DCs in languages with flexible PoS systems 337
8.4.2.2 Flexible balanced DCs in languages with rigid PoS systems 340
8.4.3 Other factors: diachrony and cognition 344
8.4.4 Summary 345
8.5 Summary 345
9 Summary and Conclusion 349
References 355
References for the Sample Languages 369
Appendices
Appendix I: Language Sample 379
Appendix II: Excluded Parts of Speech systems 381
Appendix III: Dependent Clause Constructions. Key Examples 387
Summary in English 453

Samenvatting in het Nederlands (Summary in Dutch) 467
Curriculum Vitae 481
Acknowledgments
I like languages. At least that much was clear to me when I finished high
school. The problem was that I didn’t know yet why I liked them so much.
And since there didn’t seem to be any particular language that I liked better
than any other, I decided to start studying my own language: Dutch. During
my first classes in Dutch syntax I discovered, to my great excitement, that there
was a real academic discipline devoted to my gut feeling about languages; it
was called Linguistics. Moreover, I realized that the existence of Linguistics
implied the existence of Linguists. Very soon I knew I wanted to belong to
this group of people – that I wanted to be a linguist. I hope this book shows
that I have taken an important step on my way to becoming one.
One day in August 2004, when it was pouring with rain and I was in
a particularly bad mood, Kees Hengeveld called me to say that I got the
job: I was going to be a PhD candidate at the Department of Theoretical
Linguistics of the University of Amsterdam. Three days later my desk was
waiting for me at the Herengracht 338. Some years earlier, when I was still
a student, the huge, high-ceilinged, large-windowed room on the ground
floor of this beautiful building had been renovated. I remember clearly that
I peeped inside the room and thought: If only one of these desks could once
be mine…and here it was!
The other desks in the room belonged to three wonderful persons:
Rafael, Annerieke, and Gerry. They were the firsts of many more people
that have made my time as a PhD candidate so thoroughly enjoyable. I wish
VII
to express my deep gratitude to all of them.
First, I want to thank Kees, for the enthusiasm, dedication, and patience
with which he helped me to structure lines of thought, improve analyses,
and gain confidence as a researcher along the way. He did this not only
in his role of supervisor of my MA, MPhil, and PhD theses, but also as a
colleague in our various joint projects. I am very grateful for all these years
of collaboration, during which I learned innumerable cool things about
innumerable cool languages, as well as a few lessons about myself. Among
the latter, there is one thing Kees taught me that I will always remember:
that it is essential to invest energy in those things in (linguistic) life that
make one’s heart beat faster.
I also wish to thank Hans den Besten, Sonia Cristofaro, Jan Don, Olga
Fischer, Helen de Hoop, and Andrej Malchukov for accepting the invitation
to be a member of my committee. Special thanks go to Jan Don, who helped
me with editing the manuscript in its final stages, and who has more generally
been a wonderful colleague with whom I have collaborated intensively and
with great pleasure. Another thank you goes to Andrej Malchukov, who
took the time for interesting and helpful discussions during my stay at the
Max Planck Institute in Leipzig.
But there are many other people who have made valuable contributions
to the present work: Dik Bakker, whose extensive e-mails helped me to
overcome occasional fits of doubt and confusion; Elena Maslova, for her
generous and crystal clear comments on the interpretation of quantitative
typological data; Rob Schoonen, for his kind advice on statistical issues;
Michael Cysouw, Martin Haspelmath, and Jaklin Kornfilt for inspiring
discussions at the Max Planck in Leipzig; John Peterson and Nick Evans
for helping me out on the fascinating languages they know so much about. I
also wish to thank all the members of the parts-of-speech research group of
the Amsterdam Center for Language and Communication for their sharing
of expertise and spirit.
More generally, a great amount of people have spiced up my linguistic life
in the past years. First, I wish to thank my fellow students of the Linguistics
Research Master in Amsterdam for a very stimulating year: It’s good to be
among like-minded language-freaks. Also, many thanks go to all the people
who contribute to the warm bath called the Linguistics Department of the
University of Amsterdam. I’m going to miss you all. Finally, I feel grateful to
have met so many young fellow-typologists, with whom I have spent good
times in Leipzig and at conferences all-over Europe. All these people have
VIII | Parts of Speech and Dependent Clauses

confirmed, time and again, my feeling that linguists are inspiring, open-
minded, and great fun!
Some linguists are particularly dear to me. First, a special thank you
goes to Rafael – whose jokes are even better than his looks – for sharing his
Cofán with me, for asking difficult typological questions, and for helping
me to believe, notwithstanding gloomy moods, that our work matters after
all. Another big thank you goes to Esther and Margot, for lighting up the
Bungehuis in their very own lovely ways and for all the shared lunches,
dinners, ice-creams, and borrels (and their burning-up during runs along
the Amstel). I also want to thank Josje Verhagen and Suzanne Dikker, for
great collaboration projects during our student years, and for the many
drinks and dinners over which we shared our experiences in and outside of
linguistic careers, keeping each other’s spirits high. A very special thank you
is for Gerry Wanders, for reading the manuscript of my dissertation with so
much care.
Of course, there are also many non-linguists to thank, in fact too many to
list them all. Nonetheless, I want to mention Floor and Anne, for their patience
with all my struggles, as well as for celebrating the good times with me. And
Gijs, for never tiring of asking puzzling questions about my work. Finally, I
want thank Claartje, Sanne and Annet for making this book look so pretty.
Very particularly, I want to thank my paranimf, Lotte, for adopting me

after the Herengracht 388 was sold, and for being the best possible roomy,
colleague, and friend ever since. I am grateful for her sharing; not just of
working space and coffee, but also of classical music, intellectual frustrations
and successes, panic attacks, statistical know-how, sore muscles, broken
hearts, a great many laughs, and more time with me than anyone else. Lotte,
you’re a star!
I also want to thank my other paranimf. Joost, thank you for being my big
brother, and for giving me your unconditional love and support.
I want to thank my parents. Papa, thanks for being my father in a way

that makes me feel proud to be your daughter, and lucky to have your
courageous embrace of life as a guiding example. Mama, thanks for making
me remember you the way I do.
Finally, I want to thank Sander. My love, I doubt whether there is any other
Acknowledgments | IX
linguist’s partner who understands linguistics as well as you do. But that’s
just one of the countless reasons why you are – and have been for so many
years – an essential part of my life.
Amsterdam, June 2009
X | Parts of Speech and Dependent Clauses

List of Figures
Chapter 2 Parts of Speech

Figure 2.1 Syntactic slots as distinguished in Hengeveld (1992) 12
Figure 2.2 Syntactic slots and lexical categories 13
Figure 2.3 The functional potential of parts of speech classes 15
Figure 2.4 Possible parts of speech systems according to Hengeveld 1992 19
Figure 2.5 The flexible intermediate PoS system of Turkish 19
Figure 2.6 The rigid intermediate PoS system of Alamblak 21
Figure 2.7 Parts of speech systems, including intermediate types 22
Figure 2.8 The implicational map of parts of speech 25
Figure 2.9 The PoS system of Santali 32
Figure 2.10 The PoS system of Garo 32
Figure 2.11 Properties of semantic classes 34
Figure 2.12 Nouns, adjectives and verb as prototypical combinations of propositional
act functions and semantic classes 35
Figure 2.13 The noun-adjective distinction in Japanese 40
Figure 2.14 The conceptual structure of kiss. 47
Chapter 3 Dependent Clauses

Figure 3.1a The space of propositional functions 68
Figure 3.1b Propositional functions and dependent clause types 68
XI
Chapter 5 Parts of Speech in the Languages of the Sample
Figure 5.1 The PoS system of Warao 141
Figure 5.2 The PoS system of Kayardild 148
Figure 5.3 The noun-adjective distinction in Japanese (2) 156
XII | Parts of Speech and Dependent Clauses

List of Tables
Chapter 3 Dependent Clauses

Table 3.1 DC types and their internal formal properties 108
Chapter 4 Sampling, Research Questions, Method

Table 4.1 The language sample 113-114
Table 4.3 Functional connection between PoS and deranked DCs 117
Table 4.4 A 2x2 contingency table 124
Table 4.5a Maslova test no. 1: Change DC parameter to PoS=DC 125
Table 4.5b Maslova test no. 2: Change PoS parameter to PoS=DC 125
Table 4.6 A symmetrical dependency; original distribution 126
Table 4.7a Distribution for PoS and PoS = DC (Maslova-test no. 1) 126
Table 4.7b Distribution for DC and PoS = DC (Maslova-test no. 2) 126
Table 4.8 A one-sided asymmetrical dependency; original distribution 127
Table 4.10 A two-sided asymmetrical dependency; original distribution 128
Chapter 5 Parts of Speech in the Languages of the Sample

Table 5.1 PoS classes in the sample languages 132-137
XIII
Table 5.2 Verbs, nouns and nominals in Japanese 156
Table 5.3 Verbs and Adjectives in Mandarin Chinese 158
Chapter 6 Dependent Clauses in the Languages of the Sample

Table 6.1 DCs in the sample languages and their possibilities
concerning the expression of propositional functions 197-204
Table 6.2 Basic data on behavioural potential of all DC constructions 220-229
Table 6.4 Three-way classification of DCs 236-242
Table 6.5 Behavioural potential of type 1 balanced DCs 242-246
Table 6.6 Behavioural potential of type 2 D-SENT DCs 246-249
Table 6.7 Behavioural potential of type 3 D-ALT DCs 250-251
Table 6.8 Functional possibilities of type 1 balanced DCs 252-255
Table 6.9 Functional possibilities of type 2 D-SENT DCs 255-257
Table 6.10 Functional possibilities of type 3 D-ALT DCs 257-258
Chapter 7 Results
Table 7.1 Frequencies for languages with / without flexible PoS and DCs 263
Table 7.2 Frequencies for language with / without rigid PoS only
and rigid DCs only 264
Table 7.3 Frequencies for languages with / without flexible PoS
and flexible DCs of type 1 266
Table 7.4 Frequencies for languages with / without flexible PoS
and flexible DCs of type 2 / 3 266
Table 7.5a Maslova test no. 1: change DC parameter to PoS=DC 266
Table 7.6 Frequencies for languages with / without flex PoS
and flex DCs of type 2 268
Table 7.7 Frequencies for languages with / without flex PoS
and flex DCs of type 3 268
Table 7.2' Frequencies for language with / without rigid PoS only
and rigid DCs only 270
Table 7.9 Frequencies for languages with / without lexical contentives / non-verbs
and contentive clauses / multi-functional clauses 280
Table 7.10b Maslova test no. 2: change PoS parameter to PoS=DC 281
XIV | Parts of Speech and Dependent Clauses

Table 7.11 Frequencies for languages with / without contentives / non-verbs
and contentive clauses / multi-functional clauses of type 1 282
Table 7.12 Frequencies for languages with / without lexical contentives / non-verbs
and contentive clauses / multi-functional clauses of type 3 283
Table 7.13b Maslova test no. 2: change PoS parameter to PoS=DC 283
Table 7.14 Frequencies for languages with / without nominals and nominal clauses 284
Table 7.15 Frequencies for languages with / without nominals and nominal clauses
of type 1 285
of type 2 / 3 286
of type 2 286
of type 3 286
Table 7.19 Frequencies for languages with / without modifiers and modifier clauses 287
Table 7.20 Frequencies for languages with / without nouns and complement clauses 290
Table 7.21 Frequencies for language with / without nouns and complement clauses
of type 1 290
of type 2 / 3 291
of type 2 291
of type 3 291
Table 7.25 Frequencies for languages with / without adjectives and relative clauses 292
Table 7.26 frequencies for languages with / without adjectives and relative clauses
of type 1 293
Table 7.27 Frequencies for languages with / without adjectives and relative clauses
of type 2 / 3 293
of type 2 294
of type 3 294
Table 7.30 Frequencies for languages with / without (derived) manner adverbs
and adverbial manner clauses 295
and adverbial manner clauses of type 1 296
List of Tables | XV
and adverbial manner clauses of type 2 / 3 297
Table 7.33 Availability and type of flexible DCs in languages with one or
more flexible PoS classes 298-299
Table 7.34 Availability and type of flexible DCs in languages with rigid
PoS classes only 299-300
Table 7.35 Availability and type of contentive / multi-functional clauses
in languages with lexical contentives / non-verbs 300
Table 7.36 Availability and type of contentive / multi-functional clauses
in languages without lexical contentives / non-verbs 301-302
Table 7.37 Availability and type of nominal clauses in languages with
lexical nominals 302
Table 7.38 Availability and type of and nominal clauses in languages
without lexical nominals 302-303
Table 7.39 Availability of modifier clauses in languages with (derived)
lexical modifiers 303-304
Table 7.40 Languages without (derived) lexical modifiers, but with
modifier clauses 304
Table 7.41 Availability and type of complement clauses in languages
without nouns 304
Table 7.42 Availability and type of complement clauses in languages
with nouns 304-305
Table 7.43 Availability and type of relative clauses in languages
without adjectives, but with a flexible lexical strategy for
the function of modifier in a referential phrase 306
Table 7.44 Availability and type of relative clauses in languages with an open
class of simple adjectives 306-307
Table 7.45 Availability and type of relative clauses in languages without an open
class of simple adjectives, but with derived adjectives
(and a small class of simple adjectives) 307
class of simple / derived class adjectives, but with a small class of adjectives 308
Table 7.47 Availability and type of relative clauses in languages without a lexical
strategy for the function of modifier in a referential phrase 308
Table 7.48 Availability and type of adverbial manner clauses in languages without an
open class of simple / derived manner adverbs, but with a flexible lexical
strategy for the function of modifier in a predicate phrase
(and in some cases also a small class of rigid manner adverbs) 309
XVI | Parts of Speech and Dependent Clauses

Table 7.49 Availability and type of adverbial manner clauses in languages with an
open class of simple manner adverbs 310
Table 7.50 Availability and type of adverbial manner clauses in languages without an
open class of simple manner adverbs, but with derived manner adverbs
(and in some cases a small class of simple manner adverbs) 310
class of simple / derived class manner adverbs, but with a small class
of manner adverbs 311
Table 7.52 Availability and type of adverbial manner clauses in languages without
a lexical strategy for the function of modifier in a predicate phrase 311
Chapter 8 Discussion: Flexibility and Functional Transparency

Table 8.1 Languages with a flexible PoS class, and the distribution of deranked
DCs that appear in at least one of the same function(s) 323-324
Table 8.2 The principle of increasing complexity: Flexible PoS systems and
deranked DCs 327
Table 8.3 Rigid PoS classes in languages with flexible PoS, and the
distribution of deranked DCs that appear in the same function
as these rigid PoS classes. 332-333
Table 8.4 Rigid PoS classes in languages with rigid PoS only, and the
distribution of deranked DCs that appear in the same function 334-335
Table 8.5 Flexible balanced DCs in languages with flexible PoS systems 337
Table 8.6 Flexible balanced DCs in languages with rigid PoS systems 341
List of Tables | XVII

List of Abbreviations
ab = ablative
abs = absolutive
acc = accusative
act = active
adv = adverbial marker / subordinator
aff = affirmative
afr = aforementioned
ag = agent
all = allative
anaph = anaphoric proform
anim = animate
a.obl = associating oblique case
aor = aorist
appl = applicative
art = article
assoc = associative
attr = attributive
aux = auxiliary
av = actor voice
ben = beneficiary
ca = continuative aspect
caus = causative
XIX
ccm = conjunction class marker
cl = classifier
cm = challengeable marker
cntr = contrastive focus
c.obl = complementizing oblique case
com = completion
comp = complementizer
cond = conditional
conj = conjunction
conn = connector
cons = consequential
cont = contemporative mood
conv = converb
cop = copula
cr = correlative
dat = dative
decl = declarative marker
deic = deictic element
def = definite
defoc = de-focused
dem = demonstrative
dep = dependent form
dim = diminutive
dir = directional
dist = distal
distr = distributive
dnmlz = denominalizer
d.so = distal extension: point of view of source
dtr = de-transitivizer
du = dual
dur = durative
dyn = dynamic
emph = emphatic
emot = emotive
erg = ergative
ev.cop = evidential copula
excl = exclusive
fem = feminine
XX | Parts of Speech and Dependent Clauses

fin = finite
foc = focus
fut = future
gen = genitive
genr = general TAM particle
ger = gerund
hab = habitual
hum = human
ideoph = ideophone
immed = immediate
imp = imperative
inanim = inanimate
incep = inceptive
incl = inclusive
ind = indicative
indf = indefinite
iness = inessive
inf = infinitive
instr = instrumental
interj = interjection
intr = intransitive
i.obj = indirect object
iov = indirect object version
ipfr = imperfective
irr = irrealis
iter = iterative
lg = ligature
lk = linker
loc = locative
l.pron = logophoric pronoun
lv = locative voice
m = middle voice
m.abl = modal ablative case
masc = masculine
m.loc = modal locative case
mod = modality marker
modif = modifier
m.prop = modal proprietive case
List of Abbreviations | XXI

n = non-past low tone verbal prefix
nc = non-control
neg = negation
neut = neuter
nmlz = nominalizer
nom = nominative
n.fin = non-finite
n.pst = non-past
nv = neutral version
obj = object marker
obl = oblique
onom = onomatopoeic form
opt = optative
or = oriented
orig = origin
ov = object version
part = partitive
pass = passive
pat = patient
pfv = perfective
phtm = phrase terminal marker
pl = plural
pn = proper name
po = potential object (verbal extension)
poss = possessor
possd = possessed
pot = potential
pref = prefix
pret = preterite (simple past)
prev = preverb
progr = progressive
pron = pronoun
pros = prosecutive
prox = proximate
prs = present
prt = particle
pst = past
ptc = participle
XXII | Parts of Speech and Dependent Clauses

pv = patient voice
q = interrogative
quot = quotative
rdp = reduplication
real = realis
ref = referential
refl = reflexive
rel = relative marker
relc = relational case
rem = remote
res = resultative
rcp = reciprocal
rsp = resumptive pronoun
sbjv = subjunctive
seq.conv = sequential converb
semtr = semi-transitivizer
sf = stem forming formative
sg = singular
sim = adverbial marker of simultaneity
spec = specifier
ss = same subject
sr = source
stat = stative
sub = subordinator
subj = subject marker
suff = suffix
superess = superessive
superl = superlative
sv = subjective version
sw = switch reference
tent = tentative
ths = thematic suffix
top = topic
tr = transitivizer
trl = translative case
uv = undergoer voice
val = validator
ve = ventive prefix
List of Abbreviations | XXIII

Introduction
Languages are formal systems used for human communication. Three basic
communicative functions, or communicative acts, may be distinguished:
Reference, predication, and modification. The act of reference identifies a
referent, i.e. an entity that one wants to talk about, and establishes a cognitive
file for that referent. The act of predication ascribes something to a referent,
i.e. it reports that this referent is involved in some state of affairs. The act of
modification enriches either reference or predication, by expressing an additional
feature of either a referent or a state of affairs (Searle 1969, Croft 2001: 66).
The functions of reference, predication, and modification may be
fulfilled through linguistic forms of various types. The most basic way to do
this is probably by means of a phrase consisting of a single lexical element.
The lexical elements of any particular language can be divided into different
groups or classes, called parts of speech. Prototypically, the class of nouns is
used for communicative acts of reference (cf. car in example (1a)); verbs are
used for predication (cf. wash in (1b)); adjectives for the modification of
referential expressions (cf. red in (1c)); and adverbs for the modification of
predications (cf. quickly in (1d)).
(1) a. I wash the car.

b. I wash the car.
c. I wash the red car.
d. I wash the car quickly.
1
However, not all languages follow this particular pattern, in which every
part of speech class is specialized for the expression of a single function.
Alternatively, languages may display part of speech classifications in which
two or more functions can be expressed by members of the same class.
Hockett (1958: 225) compares such functionally different parts of speech
to “athletic squads, trained in different ways to play much the same game.”
Some languages train specialists, while others try to make an “all-‘round
player or triple-threat man of every member in the squad.” These two types of
techniques may also be combined within a single language system, “producing
some specialists but also good numbers of double-threat and triple-threat men.”
Adopting a terminology proposed by Hengeveld (1992), I will call parts
of speech that are functionally specialized rigid, and those that can express
multiple functions flexible.
Apart from simple lexical items, however, more complex linguistic
structures may also be used to express the functions of reference, predication,
and modification within an utterance. One possibility is the use of a clause-
like construction within a larger utterance. Traditionally, this linguistic
situation is referred to as subordination. Consider the example in (2), where
the subordinate or dependent clause ‘John will come home soon’ is used as a
referring expression that functions as the object of the predicate of the
main or matrix clause ‘Mary hopes’, to which it is linked by means of the
subordination marker that.
(2) Mary hopes [that John will come home soon].
Like parts of speech, certain types of dependent clause constructions are

rigid, i.e. can fulfil only a single type of function, while others are flexible, i.e.
may express multiple functions. For instance, the English that-construction,
illustrated in (2) above, is flexible because it can also be used for modification,
as is shown in (3), where the dependent clause ‘that we bought together
yesterday’ expresses an additional property of the referent of ‘book’, which is
the direct object within the matrix clause ‘I gave her the book’.
(3) I gave her the book [that we bought together yesterday].
Apart from their possibilities regarding the expression of communicative

function(s), dependent clauses – being complex constructions – can
2 | Parts of Speech and Dependent Clauses

also be studied from the perspective of their internal morpho-syntactic
properties. Usually, the morpho-syntactic properties of dependent clauses
in a particular language are described in comparison to the properties of
ordinary independent clauses in that language. Consider the difference
between examples (4a) and (4b): The dependent clause in (4a), marked with
the subordinator that, expresses future tense (will move) and the subject
Paul is encoded in the same way as it would be in an independent clause.
This type of dependent clause, which structurally resembles an independent
clause, is called balanced. In contrast, the dependent clause in (4b) is marked
by the special -ing form of the predicate, it cannot express tense, and it has a
subject that is encoded as a possessor (Paul’s). This second type of dependent
clause, the structure of which deviates to some extent from the structure of
an independent clause, is called deranked (Stassen 1985).
(4) a. John regrets [that Paul will move to Finland].

b. John regrets [Paul’s moving to Finland].
The aim of the present research is to investigate the relationship between

the functional flexibility or rigidity displayed by a particular language’s parts
of speech classes on the one hand, and by its balanced / deranked dependent
clauses on the other hand. More specifically, this study tries to discover to
what extent it is possible to predict, on the basis of a language’s parts of
speech system – that is, its major part of speech classes and their respective
potential to express particular functions – what the functional potential of
its dependent clause constructions will be.
This goal is approached by means of an investigation of the parts of
speech classes and dependent clause constructions of a genealogically and
geographically balanced sample of 50 languages. For every major part of speech
class and for every relevant dependent clause construction in each of the sample
languages it is determined which function(s) it may express. These typological
data pertaining to parts of speech and dependent clauses are then combined
in order to identify dependency relations between the functional behaviour
of lexical and clausal constructions in languages. The results of these analyses
are interpreted within the framework of functional linguistics, which assumes
that grammars, i.e. linguistic forms and structures, are ultimately explainable in
terms of their function: the encoding of human communication.
This book is organized as follows. The first part, to which Chapters 2,
3, and 4 belong, provides the theoretical and methodological background
Chapter 1 – Introduction | 3
of the research. Chapters 2 and 3 are devoted to the theory and typology of
this study’s two dramatis personae: parts of speech and dependent clauses,
respectively. First, in Chapter 2, I give an overview of possible approaches
to parts of speech classification as proposed in the functional-typological
literature. Two (relatively) recent approaches are discussed and compared
in detail: Hengeveld’s pragmatic-syntactic theory of parts of speech and
Croft’s universal-typological theory of parts of speech. The former serves
as a point of departure for the present study, because it defines parts of
speech exclusively in terms of the communicative function(s) that they can
express (Hengeveld 1992, Hengeveld et al. 2004, Hengeveld & Rijkhoff
2005, Hengeveld & Van Lier 2008, 2009). The mono-dimensionality of
this approach has been the target of its most important critic: Croft (2000,
2001, 2005). The latter proposes to define parts of speech in terms of a
broader set of criteria, including their semantic meaning and their ability
to express particular morphological categories. Croft’s approach yields a
typology in which part of speech classes are not directly comparable across
languages. On the other hand, the cross-linguistic comparability of parts
of speech defined according to Hengeveld’s approach is achieved at the
cost of ignoring specific sets of linguistic facts. In addition, the difference
between the Hengeveldian and the Croftian approaches has important
repercussions for the debate about so-called flexible languages, i.e. languages
that presumably lack all parts of speech distinctions, most significantly a
basic distinction between nouns and verbs. At the end of Chapter 2 I review
this particular debate, and propose an approach to lexical flexibility that
aims to integrate the insights of several earlier studies (see also Don & Van
Lier, forthcoming).
In Chapter 3 I turn to the theoretical and typological treatment of
dependent clauses, which is approached from two perspectives. First, it is shown
that dependent clause constructions, like parts of speech, can be defined in
terms of the communicative function(s) that they are able to express. Second,
as mentioned above, dependent clauses are complex constructions, which can
also be classified according to their internal morpho-syntactic properties. This
second perspective, which is most common in the existing literature, focuses
on the fact that dependent clause constructions often display a certain mixture
of formal features of independent clauses on the one hand, and features of
lexical or phrasal constructions on the other hand (cf. examples (4a-b) above).
I briefly review the functional-typological literature concerning these two
types of features and the various ways in which they may be combined in

actual dependent clause constructions. Two recent studies on the typology
of the internal structure of dependent clauses, by Cristofaro (2003) and
Malchukov (2004), are discussed in more detail. These formal features of
dependent clauses, as I will show, are expected to influence the hypothesized
functional dependency relation between these constructions and parts of
speech classes. Therefore, at the end of Chapter 3, I present a classification
system of dependent clause constructions that takes into account both their
functional properties and their internal formal properties.
Chapter 4 rounds off the theoretical and methodological part of the
book. In this chapter, I first present the composition of the language sample
investigated for this research. Second, I formulate and operationalize the
specific hypotheses that are tested on this sample. These hypotheses take
the form of predictions about dependency relations between the functional
patterns displayed by the part of speech classes of particular languages, and
the functional patterns of their (different structural types of ) dependent
clause constructions. Finally, Chapter 4 presents the method that will be
used to test these predictions.
The second part of the book contains the actual typological data.
First, Chapters 5 and 6 separately present the classifications of part of
speech classes and dependent clause constructions in the languages of
the sample. In addition, a number of issues related to the analyses of both
types of constructions are raised, and linked to the theoretical discussions
of Chapters 2 and 3, respectively. Subsequently, in Chapter 7, the two
data sets of Chapters 5 and 6 are combined in order to test the specific
predictions about their interrelationship, as formulated in Chapter 4. This
analysis yields a set of statistically significant correlations, which allow for
the formulation of a number of typological generalizations. It is shown that
these generalizations pertain in particular to languages with part of speech
classes displaying a high degree of flexibility or multi-functionality (as
opposed to languages with less or no lexical flexibility), and to the functional
behaviour of deranked (as opposed to balanced) dependent clauses.
In the third and final part of the book I further discuss the results of the study
and present its conclusions. First, In Chapter 8, the findings of Chapter 7 are
reconsidered, taking a specific functionalist perspective that makes reference
to the overall complexity of language systems and the various ways in which
this level of complexity can be attained. To quote Hockett once more, “all
languages have about equally complex jobs to do.” (Hockett 1958: 180) This
‘job’ – the encoding of human communication – can indeed be considered
Introduction | 5
universal. However, languages differ in the way in which they divide the
labour of expressing the total amount of functional complexity over the
available grammatical resources, including lexicon, morphology, and syntax.
It seems that flexibility or multi-functionality in one area of the grammar
must be counterbalanced or ‘traded off ’ by functional specificity or rigidity
in another area. In general, the functional flexibility of linguistic units tends
to gradually decrease when they become structurally more complex. When
flexibility is nonetheless retained in complex constructions, this requires
functional disambiguation at the highest level of phrase-structure, either
by means of ordering restrictions or through the use of function-indicating
elements. In general, it will be shown in Chapter 8 that an explanatory
framework based on functional transparency provides a complete account of
the findings of this research. Finally, Chapter 9 summarizes the main points
of the study as a whole and presents its overall conclusions.

Parts of Speech
2.1. Introduction
Parts of Speech (henceforth PoS) is the traditional term used to refer to
the major classes of words or lexemes that are distinguished in a language
(Schachter & Shopen 2007: 1). The issue of lexical classification or
categorization has a long and troublesome history in linguistics. Generally,
the nature of categorization in human language has been and continues to
be the subject of a debate that goes back to Artistotle and has raised the
interest of philosophers, psychologists, and linguists alike. In the present
chapter, I will not attempt to give an overview of the various conceptions
of categorization that have been on and off stage at various periods in the
history of these sciences 1. Rather, I will limit myself to a brief discussion of
the problems that have been at stake in defining lexical categories within
the sub-discipline of linguistics called functional typology 2. This is the topic
of section 2.2.
Subsequently, in section 2.3, I go on to discuss in more detail one specific
functional-typological approach to lexical categorization, namely the theory
proposed by Hengeveld (1992 a, b; Hengeveld et al. 2004; Hengeveld &
Rijkhoff 2005; Hengeveld & Van Lier 2008, 2009). I will explain that
1
For a recent overview of this type, see Aarts 2006.
2
For other overviews of approaches to lexical categorization, including approaches that do
not belong to the functional-typological framework, the reader is referred to Bisang (fc.),
Evans (2000), Haspelmath (2001), and Rijkhoff (2007).
7
the Hengeveldian approach provides a suitable basis for the present study,
because it defines PoS classes in terms of the propositional function(s) they
can fulfil in an utterance 3. These PoS definitions will serve as a basis for the
typology of PoS systems presented in Chapter 5.
However, Hengeveld’s approach is certainly not without problems. In
section 2.4 I discuss a number of objections that have been raised against
this theory, most elaborately by Croft (2000, 2001, 2005). In general terms,
these problems involve the tension between the language specific nature of
lexical categories on the one hand, and their cross-linguistic comparability
on the other hand. As such, they touch upon a fundamental discussion in
linguistic typology concerning the (non-)existence of universal linguistic
categories (see Haspelmath 2007, Newmeyer 2007).
Section 2.5 is specifically devoted to lexical flexibility, which for the
time being I will define as the possibility to use the members of one or
more lexeme class(es) of a language in more than one propositional function
without needing any morpho-syntactic adaptation. First, I will review
the discussion about lexical flexibility in recent literature. Subsequently, I
present an approach to lexical flexibility that attempts to integrate insights
of various earlier studies (see also Don & Van Lier, forthcoming). Finally,
section 2.6 provides a summary of this chapter.
2.2. Functional-typological attempts at defining parts of speech

As the binary term implies, a functional-typological approach to the study of
linguistics involves two basic ingredients: typology and functionalism. First,
typology deals with cross-linguistic comparison, under the assumption
that the structural variation displayed by languages across the word is
not random, but is rather constrained by a set of principles that are called
typological universals. Second, functionalism refers to the assumption that
these universals of language structure can be explained in terms of language
use, i.e. in terms of the semantic and pragmatic meanings expressed by these
structures in human communication 4.
Within the functional-typological literature, various approaches to lexical
categorization have been proposed. These can be characterised in terms of the
type of criteria on which the definitions of PoS classes are primarily based.
3
The term propositional function has been introduced by Croft (2001). Its use in the present
study will be clarified in the course of this chapter.
4
For a more detailed characterization of the functional-typological approach the reader is
referred to Croft (2003), or Cristofaro (2003: 5-9).

According to Givón (2001: 49) there are three different types of such criteria:
(i) Semantic criteria: the kind of meanings that tend to be coded by

words of a particular class.
(ii) Morphological criteria: the kind of bound morphemes that tend to
be affixed to words of a particular class.
(iii) Syntactic criteria: the typical syntactic slot that words of a particular
class tend to occupy.
To these, a fourth type may be added, namely:
(iv) Pragmatic criteria: the kind of discourse function that tends to be

coded by words of a particular class. (cf. Sasse 1993b:196)
Although most approaches acknowledge the relevance of all four types of

criteria for the definition of lexical categories, there are clear differences
in emphasis. First, there is the so-called notional approach, which takes the
semantic meaning of lexemes as the most important property of PoS classes,
i.e. the types of entities that groups of lexemes denote. Thus, nouns are
defined as lexemes denoting first-order entities or objects and individuals.
In contrast, verbs denote second-order entities, i.e. “events, processes, states-of-
affairs, etcetera, that are located in time and which are said to ‘occur’ or ‘take
place’ rather than to ‘exist’” (Lyons 1977: 443). Langacker (1987) provides a
notional definition of nouns and verbs from the perspective of Cognitive
Grammar. According to his definitions, nouns designate bounded regions,
typically in the domain of space, and are therefore cognitively perceived
as static and holistic. Verbs, on the other hand, designate processes, and are
cognitively perceived as sequences of states evolving through time.
The obvious problem of the notional approach, as has been pointed out
by many scholars (Dik 1989; Lehmann 1990, among others), is that the
same type of entity may be expressed by lexemes that belong to different
lexical classes, both within and across languages. The English word
destruction, for instance, is a noun, even though it denotes an event rather
than an object. Furthermore, property concepts such as ‘small’ are in some
languages lexicalized in the form of a separate class of adjectives, while in
other languages these concepts are expressed by verbs or nouns.
A second possibility is the discourse approach, advanced by Hopper and
Thompson (1984, 1985). It defines the differences between PoS classes
Chapter 2 – Parts of Speech | 9

in terms of their prototypical discourse function. Thus, a prototypical
verb expresses event-occurrence, whereas a prototypical noun introduces
a participant into the discourse. Whenever a lexeme does not express its
prototypical discourse function, it will show reduced categoriality, meaning
that it will have fewer possibilities to take morphological marking. However,
as in the case of the notional approach, the discourse approach makes
no principled distinction between universal discourse functions and the
language-specific lexical categories that express them.
In an attempt to overcome the problem of the notional and discourse-
based approaches, focus shifted more towards the morphological characteristics
of PoS. For the classical languages, it had been noted long ago that lexical
classes could be identified by means of the particular morphological
categories for which they are specified. Typical nominal categories would
include grammatical gender, number, definiteness and case distinctions.
Verbs rather express categories like tense, mood, aspect, and person
agreement. And adjectives take morphological markers expressing degrees
of comparison and may display gender, number and / or case agreement with
the head they modify 5.
As Croft (1991: 86) observes, the morphological categories mentioned
above are indeed “tailored to” a specific lexical class, because they express bits
of meaning that are relevant to the semantic meaning or the pragmatic-
syntactic function prototypically expressed by members of that class. Thus,
aspectual distinctions are relevant to verbs, which prototypically express
actions; number is relevant to nouns, which prototypically refer to countable
objects; and comparative marking is relevant to adjectives, which typically
express gradable property concepts. Similarly, agreement has the function
of showing the relationships between typically verbal predicates and their
typically nominal argument(s), or between typically nominal heads and
their typically adjectival modifiers.
However, morphological criteria are problematic too, because they are
no more than “diagnostic” (Beck 2002:14) or “symptomatic” (Langacker 2002:
60) for lexical categories. For one thing, not all PoS classes in all languages
express the same type and amount of morphological categories. Therefore,
morphological criteria do not provide a suitable basis for cross-linguistic
5
As far as (manner) adverbs are concerned, it seems that there are no typical inflectional
classes associated with this lexeme class. This is probably related to the fact that the status
of manner adverbs as a major word class is somewhat problematic in the first place (see
Haspelmath 2001: 16544).

generalizations. In addition, morphological categories are problematic
as intra-linguistic criteria as well, because words belonging to the same
lexical class often do not display the full range of morphological categories
belonging to that class.
Schachter (1985) tries to come up with more integrated definitions of
PoS classes. He argues that, apart from morphological criteria, the possible
syntactic function(s) of lexemes should be taken into account as well6.
However, this combination of formal criteria still yields definitions of PoS
classes that are language-specific rather than universal. At the same time,
Schachter uses notional labels to refer to lexeme classes, which according to
him, “appropriately reflect universal semantic considerations” (Schachter 1985: 4).
This suggests that some kind of common prototypical meaning of the
members of a lexeme class should be acknowledged and play a part in a
universal theory of lexical categorization.
Anward et al. (1997: 171) also argue in favour of an approach to PoS
definition that integrates semantic, morphological, and syntactic, as well
as phonological criteria. However, they clearly state that “given that such
properties of constituents differ from one language to another at least in part,
parts-of-speech classes will need to be set up separately for each language […]”.
Again, this approach has obvious repercussions for the cross-linguistic
comparability of PoS classes.
To sum up, there are several possible approaches to the definition of
PoS classes, each with its own problems. These problems can be broadly
divided into two types: either universal tendencies are defined without
accounting for language-specific categories, or the other way around. In the
next sections, I will discuss in more detail two recent functional-typological
theories of PoS classification that deal with these problems in very different
ways: Hengeveld’s syntactic-pragmatic approach (section 2.3) and Croft’s
Universal-Typological approach (section 2.4). These sections, together with
the present one, should make clear that there is no single answer to the
question of whether and how the four types of criteria – pragmatic, semantic,
syntactic, and morphological – should be combined into definitions of lexical
categories within and across languages. As Bisang (fc.) puts it: “There is no
final solution appearing on the horizon”.
6
See also Schachter & Shopen (2007), who explicitly distinguish between notional and
grammatical criteria for defining PoS.

2.3 Hengeveld’s syntactic-pragmatic approach
2.3.1 Slots and fillers

Hengeveld (1992) developed his PoS theory as part of his typological study
of non-verbal predication. He defines PoS classes in terms of the syntactic
slot or slots that their members can occupy in an utterance, without any
further morpho-syntactic measures 7. Hengeveld rules out semantics as a
component of PoS definitions, because, as explained in the previous section,
denotational meaning is not a water-tight criterion for lexical category
membership. The expression of morphological categories is used as an
analytic device only, which can help to identify possible candidates for
lexeme classes in specific languages. It does not, however, play a role in the
actual definition of PoS classes.
Working in the framework of Functional Grammar (henceforth FG;
Dik 1989, 1997), Hengeveld distinguishes four possible syntactic slots in
which lexemes may be inserted. Two of these slots are obligatorily filled in
their respective phrases, namely:
(i) The head of a predicate phrase and

(ii) The head of a referential phrase
The other two slots are optionally filled in each of the two phrase types.
They are:
(iii) A modifier within a predicate phrase and

(iv) A modifier within a referential phrase.
The four slots can thus be tabulated as in Figure 2.1:
Head Modifier
Predicate phrase (i) (iv)
Referential phrase (ii) (iii)
Figure 2.1: Syntactic slots as distinguished in Hengeveld (1992)
On the basis of their distribution over these four syntactic slots, the
7
As I will explain shortly, in more recent work (Hengeveld & Van Lier 2008, 2009) the
syntactic slots are re-interpreted as propositional functions.

definitions of the major PoS classes verb, noun, adjective, and manner
adverb are formulated as in (1):
(1)
(i) A verb is a lexeme that can be used, without further morpho-
syntactic measures, as the head of a predicate phrase only (slot (i));
(ii) A noun is a lexeme that can be used, without further morpho-
syntactic measures, as the head of a referential phrase (slot (ii));
(iii) An adjective is a lexeme that can be used, without further morpho-
syntactic measures, as a modifier in a referential phrase (slot (iii));
(iv) A manner adverb is a lexeme that can be used, without further
morpho-syntactic measures, as a modifier in a predicate phrase
(slot (iv)).
(Hengeveld 1992: 58; Hengeveld et al. 2004: 530).
In Figure 2.2 the four syntactic slots are represented once again; this time
with the four lexeme classes appearing in the relevant slots (from Hengeveld
et al. 2004: 530):
Head Modifier
Predicate phrase verb manner adverb
Referential phrase noun adjective
Figure 2.2: Syntactic slots and lexical categories
The definitions of the PoS classes in (1) give rise to a number of issues that
require some clarification. First, there is quite a bit more to say about the
proviso ‘without further morpho-syntactic measures’. For a proper understanding
of what constitute such a measure, I find it useful to refer to a distinction
made by Croft (1991, 2001), between structural coding and behavioural
potential 8. The first term, structural coding, refers to those morpho-syntactic
markers that specifically indicate the syntactic or propositional function of
a linguistic unit (a lexeme, in the case at hand). The other type of marking,
behavioural potential, involves the expression of morpho-syntactic categories
that are semantically, syntactically and / or pragmatically relevant to the
particular function in which a linguistic unit is used, but do not indicate that
8
In Croft’s earlier work (1991: 58) structural coding is termed “function indicating morpho-
syntax”.

function as such (cf. Croft 2001: 66). Thus, structural coding can be thought
of as a condition for using a linguistic (lexical) unit in a certain function,
whereas behavioural potential is the consequence of using it in that function.
Consider for example the Dutch noun vlees ‘meat’. If this lexeme is to express
the function of modifier in a referential phrase, it needs the affix -ig. This
affix is an instance of structural coding. However, when the resulting form
vlezig ‘meaty’ is used to modify a definite neuter noun like het gezicht ‘the
face’, then it takes the gender agreement suffix -e (het vlezig-e gezicht). This
suffix is an instance of behavioural potential. It is crucial that in Hengeveld’s
definitions of PoS classes only structural coding counts as further morpho-
syntactic measures. In contrast, behavioural potential corresponds to what has
been termed ‘morphological criteria’ in section 2.2, and does not play a role
in the definitions in (1) above.
Second, it is noteworthy that Hengeveld’s definition of verbs includes a
special ‘only’-condition, which is absent in the definitions of the other three
PoS. This is because in many languages nouns, adjectives, and adverbs can
have an additional function next to their defining function as specified in (1)
(ii)-(iv): They can also be used as the head of a predicate phrase9. This means
that verbs are the only PoS class whose exclusive, distinguishing function it
is to express the head of a predicate phrase; hence the ‘only’-condition in the
definition of this lexeme class.
Cross-linguistically, the use of a non-verbal lexeme as the head of a
predicate phrase (i.e. non-verbal predication) may trigger various formal
expression strategies. Following Hengeveld’s typology of non-verbal
predication, two main types of strategies are distinguished: those with a
copula and those without one. In the first case, the copula counts as structural
coding. The second strategy, without a copula, can be further sub-divided
into two types, termed the zero-1 and the zero-2 strategy, respectively. In
the case of a zero-1 strategy, the non-verbal predicate shows the same
behavioural potential as a verbal predicate in the language under analysis.
Thus, any categories that are expressed on a verb, such as TAM and person
agreement, should also appear on a non-verbal predicate. In contrast, in the
case of a zero-2 strategy the non-verbal predicate does not show the same
behavioural potential as a verbal predicate, but rather remains uninflected
and is simply juxtaposed to its argument 10.
9
This is then termed a non-verbal predication, which was in fact the main topic of
Hengeveld’s 1992 study. See further below.

A third point concerning the details of Hengeveld’s PoS definitions is that
the class of adverbs includes manner adverbs only. This is because other types
of adverbs do not modify the head of a predicate phrase but rather another
unit. This may be a larger unit, such as a state of affairs, a proposition, an
illocution, or an even larger stretch of discourse 11. Alternatively, it may be
another modifying unit, such as an adjective or another adverb 12.
Taking into account the issues discussed above, the functions of the four
PoS classes (verb, noun, adjective, and manner adverb) can be represented
as in Figure 2.3, in which ‘+’ stands for the defining use of a lexical class
(as mentioned in the left-hand column); ‘-’ for excluded uses; and ‘+ / -’ for
potential additional uses. The figure shows that verbs are uniquely used as
the head of a predicate phrase, while nouns, adjectives, and manner adverbs
may have additional uses as the head of a predicate phrase, next to their
respective defining uses as the head of a referential phrase, a modifier in a
referential phrase, and a modifier in a predicate phrase.
Head of Head of Modifier in Modifier in

Predicate Referential Referential Predicate
phrase phrase phrase phrase
Verbs + – – –
Nouns + / – + – –
Adjectives + / – – + –
Manner adverbs + / – – – +
Figure 2.3: The functional potential of parts of speech classes
2.3.2 The cross-linguistic perspective
2.3.2.1 General introduction

Individual languages across the world do not all have systems of lexical
categorization of the type represented in Figures 2.2 and 2.3 above, in
10
In many cases where a zero-2 strategy is used, a copula can optionally be added. The copula
often remains un-expressed in ‘default’ cases, i.e. involving present tense and third person
singular. In other, non-default cases, TAM and person distinctions are expressed on the
copula. See also Chapter 5, section 5.
11
See Hengeveld & Mackenzie (2008, Chapter 3) for examples.
12
Regarding this latter point, the modification of lexical units other than verbs, Hengeveld
& Van Lier (2008: 760-762) show that the definition of the class of lexemes functioning as
a modifier in a predicate phrase must make reference to the PoS class of the head of that
predicate phrase. That is to say, when the predicate phrase has a verbal head, it takes different
types of adverbial modifiers (namely manner adverbs) than when it has a nominal, adjectival,
or adverbial head (namely reference / referent modifiers or degree adverbs).

which there are four PoS classes, each of which is mapped neatly onto a
single defining function. Rather, many languages display different types
of mappings of their lexical categories onto the set of possible functions.
Hengeveld’s syntactic approach to lexical categorization provides a method
to compare these types of mapping systems across languages, called parts of
speech systems (henceforth PoS systems).
According to Hengeveld’s typology, PoS systems come in two basic
types: rigid and flexible ones. Rigid PoS systems are characterized by the
fact that all lexical classes are rigid, i.e. specialized for the expression of a
single function. When lexical items do appear in other functions, this must
be marked by means of further morpho-syntactic measures or structural
coding (with the possible exception of the predicative function, cf. the + / -
signs in Figure 2.3). Rigid PoS systems may display all four lexeme classes
that are represented in Figures 2.2 and 2.3. Alternatively, however, they may
lack one or more PoS classes. This means that one or more function(s)
cannot be expressed lexically, so that the relevant language must resort to an
alternative, non-lexical strategy. In Garo NPs, for instance, the translational
equivalent of an English adjective is always expressed by means of a relative
clause, as illustrated in (2):
Garo (Hengeveld et al. 2004: 531) 13

(2) da’r-gipa mande
be.big-ptc man
‘the big man’ (lit. the man who is big / the man who ‘big-s’)
Thus, in a rigid PoS system all major lexeme classes are functionally
specialized for the expression of a single function, even though there may
not be a PoS class available for every function.
In contrast, flexible PoS systems are characterized by the fact that one or
more of the available lexeme classes is flexible, meaning that its members can
express two or more functions, without further morpho-syntactic measures.
In Dutch, for instance, the translational equivalents of English adjectives
13
The suffix –gipa in this example is not a lexical derivational suffix, as can be seen from
example (i), in which the participial form marks a relative clause construction (Burling 2004:
301):
(i) [nok-o pïsa-ko nik-gipa] metra
house-loc child-acc see-ptc woman
‘the woman who saw the child at the house’

and manner adverbs belong to a single class of flexible lexemes that can be
used without further measures as modifiers in both referential and predicate
phrases. This is shown in examples (3) and (4), respectively.
Dutch
(3) [een mooi meisje]Referential Phrase
a beautiful girl
‘a beautiful girl.’
(4) Het meisje [zingt mooi]Predicate Phrase

The girl sings beautiful
‘That girl sings beautifully.’
Notice that in the English translation of (4) the adjective beautiful does
require a further measure in order for it to be usable in the function of
modifier in a predicate phrase, namely the suffix -ly 14.
2.3.2.2 The Parts of Speech Hierarchy (Hengeveld 1992)

Hengeveld (1992) shows that the two basic types of PoS systems, flexible
and rigid, come in several variants across languages. He claims that this
variation can be systematically described in terms of a Parts of Speech (PoS)
Hierarchy, which is given in (5):
(5) Head of Head of Modifier in Modifier in

Predicate Referential Referential Predicate
Proceeding through this hierarchy from left to right, the chance decreases
that a language has a separate, functionally specialized lexical class to
express that particular function. For a language with a rigid PoS system this
means that if it lacks a lexeme class for just a single function, this will be the
rightmost function in the hierarchy: modifier in a predicate phrase. In other
words, this language will have no class of manner adverbs. In addition, the
14
In Dutch, modifiers in referential phrases show gender agreement with the head noun in
definite contexts. Thus, if in a phrase like (4a) the definite article het (rather than indefinite
een) would be used, then mooi would carry the agreement marker -e. However, as mentioned
above, gender agreement does not count as ‘further measures’, since it constitutes behavioural
potential rather than structural coding.

hierarchy is implicational. As such, it predicts that, if a rigid PoS system
lacks a lexeme class for a specific function in the hierarchy, then it will also
lack lexeme class(es) for any function(s) further to the right in the hierarchy.
For example, a language without a lexeme class to express the function of
modifier in a referential phrase will also lack a lexical class for the function of
modifier in a predicate phrase. In other words, a language without adjectives
is predicted to also lack manner adverbs.
For languages with flexible PoS systems, the hierarchy predicts that
they are most likely to have a flexible lexeme class that can express the
two rightmost functions: modifier in a referential phrase and modifier in a
predicate phrase (as in the Dutch examples (3) and (4) above). Moreover, if
in a flexible PoS system a lexeme class can express two particular functions
in the hierarchy, it is predicted that this class will also be able to express any
function(s) further to the right in the hierarchy, as well as any function(s)
lying in between. For example, if a language has a class of lexemes that can
be used as the head of a referential phrase and as a modifier in a predicate
phrase, then it is predicted that this class will also be used for modification in
a referential phrase. In other words, if a language has a class of lexemes that
can function as nouns and as manner adverbs, then these lexemes should
also be usable as adjectives.
In its original formulation (as given in (5) above), the PoS Hierarchy
predicts seven different PoS systems, wich are represented in Figure 2.4
below (Hengeveld 1992: 69; Hengeveld et al. 2004: 537). Three PoS systems
are flexible, i.e. they include one lexeme class that can be used in multiple
functions. The other four systems are rigid, i.e. they involve functionally
specialized classes only. The labels for the flexible PoS classes are taken from
Hengeveld et al. (2004): A contentive is a member of a maximally flexible
class that can be used in all four functions; a non-verb can be used as the
head and modifier in a referential phrase, and as a modifier in a predicate
phrase; and a modifier is a lexeme that can be used as a modifier in both
predicate and referential phrases.

Parts of speech Head of Head of Modifier in Modifier in
system Predicate Referential Referential Predicate
1 concentive
Flexible 2 verb non-verb
3 verb noun modifier
4 verb noun adjective manner adverb
5 verb noun adjective –
Rigid
6 verb noun – –
7 verb – – –
Figure 2.4: Possible parts of speech systems according to Hengeveld 1992
2.3.2.3 Intermediate systems (Hengeveld, Rijkhoff & Siewierska 2004)

In more recent work by Hengeveld et al. (2004), the typology in Figure 2.4
is expanded in order to accommodate languages with so-called intermediate
PoS systems (as first proposed by Smit (2001)). Such intermediate systems
exhibit features of two ‘neighbouring’ PoS systems, i.e. two PoS systems that
are on contiguous rows in Figure 2.4.
In a flexible intermediate PoS system, the lexeme class that is positioned
rightmost in the hierarchy consists of two sub-classes: one containing basic
members and the other derived members. The sub-class of derived items
can express one function less than the subclass of basic items. In particular,
the left-most function that can be expressed by the basic items cannot be
expressed by the derived items. For example, Turkish has a PoS system with
classes of basic verbs and basic non-verbs. The non-verbs, the right-most
lexical class in the system, serve as the input category for derivation with the
suffix -CA, which creates a class of modifiers, i.e. derived items that can be
used as modifiers in referential and predicate phrases, but not as the head of
a referential phrase. This is represented in Figure 2.5, in which the shading
indicates the derived nature of the modifier class.

Predicate phrase Referential phrase Referential phrase Predicate phrase
Verbs Non-verbs
Derived modifiers
Figure 2.5: The flexible intermediate PoS system of Turkish

In (6a-c) the functional possibilities of the Turkish non-verb güzel are illustrated:
in (6a) it functions as the head of a referential phrase, in (6b) as a modifier in a
referential phrase, and in (6c) as a modifier in a predicate phrase.
Turkish (Göksel & Kerslake 2005: 49)

(6) a. güzel-im
beauty-1poss
‘my beauty’
b. güzel bir kopek

beauty art dog
‘a beautiful dog’
c. güzel konuştu
beauty s / he.spoke
‘S / he spoke well’
Examples (7a-b) show the distribution of CA-derived modifiers: in (7a)

erkek-çe functions as a modifier in a referential phrase, and in (7b) ıyı-ce
modifies a predicative head.
Turkish (Kornfilt 1997: 92)

(7) a. Erkek-çe birses
man-modif a.voice
‘a manly voice’
b. Hasan gömleğ-i ıyı-ce yıka-dı

Hasan shirt-acc good-modif wash-pst
‘Hasan washed the shirt well.’
A rigid PoS system is classified as intermediate when its rightmost lexical

category is a small, closed class, rather than a large, open one. For example,
Alamblak has large open classes of rigid verbs and nouns, and a reasonably
large class of adjectives (45 items). Its class of manner adverbs, however,
comprises only 9 items (Bruce 1984: 87, 88). This is represented in Figure
2.6, in which the darker shading is used to indicate the small, closed nature
of the manner adverb class.

Predicate phrase Referential phrase Referential phrase Predicate phrase
Verbs Nouns Adjectives Manner adverbs

Figure 2.6: The rigid intermediate PoS system of Alamblak
The case of Alamblak point to the lack of a sharp distinction between ‘large,
open’ and ‘small, closed’ word classes. The class of adjectives in Alamblak
is much smaller than the classes of nouns and verbs, but also substantially
larger than the class of manner adverbs. Also the notion of openness – the
ease with which new members are added to a lexeme class – seems to be of a
relative, rather than an absolute nature. As a result, there may be borderline
cases, the classification of which inevitably involves a degree of arbitrariness.
I will return to this issue in some more detail in Chapter 5 (section 5.4.2.1),
where the PoS systems of the individual sample languages are classified.
The intermediate PoS systems predicted by the PoS Hierarchy in (6)
are included in Figure 2.7, in which the cells with light shading are derived
classes, and darker shaded cells represent small, closed lexical classes (cf.
Hengeveld et al. 2004: 409).

Parts of speech Head of Head of Modifier in Modifier in
system Predicate Referential Referential Predicate
1 contentive
1 / 2 contentive
non-verb
2 verb non-verb
Flexible 2/3 verb non-verb
modifier
3 verb noun modifier
3/4 verb noun modifier
manner adverb
4 verb noun adjective manner adverb
4/5 verb noun adjective manner adverb
5 verb noun adjective –
Rigid 5/6 verb noun adjective –
6 verb noun – –
6/7 verb noun – –
7 verb – – –
Figure 2.7: Parts of speech systems, including intermediate types
2.3.2.3 More diversity (Hengeveld & Van Lier 2008, 2009)

Even more recent typological investigations make clear that the set of
PoS systems predicted by the PoS Hierarchy, including the ‘extended’
set represented in Figure 2.7 above, is too restricted to account for the
attested cross-linguistic variation (Hengeveld & Van Lier 2008, 2009). In
particular, there are languages that display intermediate PoS systems that
are not contiguous in Figure 2.4 above. Moreover, the ‘extra’ classes of these
intermediate systems do not necessarily consist of derived members in
flexible PoS systems, nor are they always small, closed classes in rigid PoS
systems (cf. Figure 2.7 above).
Accommodating these findings, Hengeveld’s earlier model has recently
been adapted in order to give it higher coverage, without seriously diminishing
its accuracy. In other words, the new model predicts a higher amount of
different types of PoS systems, and as such accounts for a larger range of
variation attested in the world’s languages, without predicting too many PoS
systems that have not (or at least not yet) been attested in any language.
Hengeveld and Van Lier (2008, 2009) argue that the four syntactic slots

distinguished in Figure 2.1 above should be re-interpreted as propositional
functions 15. These functions constitute dimensions of a space defined in terms
of two parameters, each of which represents a dominance relation between
two functional values. In addition, the two parameters stand in a dominance
relation with respect to one another. Together, these hierarchical relations
yield a set of predictions concerning possible mappings of PoS categories
onto the space of propositional functions. The details of this model are
presented in the remainder of the present sub-section.
The first and primary parameter involves the distinction between
predication and reference 16. These two functions are conceived of as the basic
communicative functions, with which a speaker builds up the communicative
content he or she wishes to express. It is assumed that the function of
predication is privileged in relation to the function of reference, since the
communicative act of referring presupposes that something is predicated of
a referent. For instance, in a noun phrase like ‘a yellow car’ the properties 'car'
and 'yellow' are predicated of the entity being referred to. The hierarchical
relation between predication and reference is represented in (8):
(8) Predication Reference
The second and also secondary parameter involves the distinction between
heads and modifiers. Heads are obligatory and therefore primary in relation
to optional modifiers 17. This dominance relation is also shown by the fact
that the lexical class of a modifier is dependent on the lexical class of its
head (see note 12). The hierarchical relation between heads and modifiers is
represented in (9):
(9) Head Modifier
15
Croft (2001) considers modification as a propositional function on a par with predication
and reference (see section 2.4.2 of the present chapter). In his earlier work (Croft 1991:
52), however, modification is defined as an “accessory function to reference and predication”.
This latter definition is in accordance with the treatment of the modifier function in the
Hengeveldian model.
16
Hengeveld and Van Lier (2008) use the term ascription where I use predication. The former
term is used in the framework of Functional Discourse Grammar (FDG), which is the follow-
up of Functional Grammar (FG), to distinguish between a propositional or interpersonal
function and a semantic or representational function (see Hengeveld & Mackenzie 2008).
17
cf. Bloomfield’s (1962: 194-196) definition of a head: the element of a construction that can
function like the whole construction.

In addition, as already mentioned, the two hierarchical relations in (8) and (9)
are in turn ranked with respect to one another. In view of the fact that there
are appositional languages that do not use modification at all, while there
are no languages that do not display the predication-reference distinction
(even though they may not do so at the level of lexical classification, but
rather at the level of morpho-syntactic constructions), it is expected that the
predication-reference parameter is primary in relation to the head-modifier
parameter. This is represented in (10):
(10) ((Predication / Reference) (Head / Modifier))
On the basis of these three hierarchical relations, three implicational

constraints can be formulated, as listed in (11)-(13):

a. If a language has a rigid class of lexemes that can be used as the
head of a referential phrase, it must also have a rigid class of
lexemes that can be used as the head of a predicate phrase.
b. If a language has a flexible class of lexemes that can be used

as the head of a referential phrase (but not as the head of a
predicate phrase, since then the restriction becomes irrelevant),
it must also have a flexible or rigid class of lexemes that can be
used as the head of a predicate phrase.
(12) Head Modifier

a. If a language has a rigid class of lexemes that can be used as
the modifier within a phrase, it must also have a rigid class of
lexemes that can be used as the head of that phrase.
b. If a language has a flexible class of lexemes that can be used as

the modifier within a phrase (but not as the head of that phrase,
since then the restriction becomes irrelevant), it must also have a
flexible or rigid class of lexemes that can be used as the head of
that phrase.
(13) ((Predication / Reference) (Head / Modifier)) 18

If a language has distinct (rigid or flexible) classes of lexemes for

heads and modifiers within any phrase, then it must also have
distinct (rigid or flexible) classes of lexemes for heads of predicate
versus referential phrases.
The parameters in (8)-(10) and the constraints in (11)-(13) are accounted

for in the layered implicational map in Figure 2.8 below. This figure shows
that the Predication-Reference parameter ranks higher than the Head-
Modifier parameter (as indicated by the symbol in the centre), and that
the Head-Modifier parameter applies in the domains of both Predication
and Reference, which does not exclude the possibility that these domains
share a single lexeme class.
Predication
U
Modifier
Head
U
U
Reference
Figure 2.8: The implicational map of parts of speech
The map in Figure 2.8 predicts all seven systems that were also consistent with
the original parts of speech hierarchy in (5). These systems are represented
in (14)-(20) below. In addition, the new model predicts another 10 systems,
which would have been counterexamples to the original model. These are
represented in (21)-(30). In what follows, I briefly discuss each of the systems
predicted by the map in Figure 2.8. Note that in all representations of PoS
18
It is explicitly not the case that the Head-Modifier distinction must apply within the
predicative domain before it can apply in the referential domain. Thus the combination
of (8) / (11) and (9) / (12) should not be read as (PredHead PredModifier) (RefHead
RefModifier). Thanks to Michael Cysouw for pointing this out.

systems the flexible PoS classes have dark shading, while rigid classes are in
a lighter shade. When there is no PoS class available to express a particular
function, then the cell corresponding to this function remains unshaded and
is marked with a bar.
The system in (14) is consistent with the new model because it involves
no lexical distinctions at all. All functions can be expressed by a single class,
called contentives (cf. Figures 2.4 and 2.7 above).
(14)
Head Modifier
Predication
Contentive
Reference
The system in (15) is consistent because (i) it has a flexible lexeme class
(non-verbs) that can be used as the head of a referential phrase and a rigid
lexeme that can be used as the head of a predicate phrase (verbs) (cf. (11b));
(ii) it has a flexible class of lexemes that can be used as a modifier within a
predicate phrase and a rigid class of lexemes that can be used as the head of
that phrase (cf. (12b)); and (iii) it has distinct classes for heads and modifiers
in the predicate phrase as well as distinct classes for heads of predicative
versus referential phrases (cf. (13)).
(15)
Head Modifier
Predication Verb
Reference Non-verb
The system in (16) is consistent because it involves a predication-reference

distinction for heads (verbs and nouns) but not for modifiers. This is in
accordance with constraints (11a), (12b) and (13).
(16)
Head Modifier
Predication Verb Modifier
Reference Noun

The system in (17) is consistent with the model because it involves head-
modifier distinctions in both the predicative and the referential domain.
Thus it has verbs, nouns, adjectives, and manner adverbs.
(17)
Head Modifier
Predication Verb Manner adverb
Reference Noun Adjective
The system in (18) differs from the one in (17) only in that it lacks a lexeme
class for the function of modification in predicate phrases (manner adverbs),
while the head function is this domain is expressed by verbs. This does not
violate any of the constraints.
(18)
Head Modifier
Predication Verb –
The system in (19) again involves a predication-reference distinction in the

head domain, but does not have lexical classes for either of the two modifier
functions. This system conforms to (11a), since it has rigid classes of heads
in both the reference and the predication domain (verbs and nouns), while
constraints (12) and (13) are not applicable.
(19)
Head Modifier
Reference Noun –
The system in (20), finally, has only one lexical class, namely one that fulfils
the function of head of a predicate phrase (verbs). This does not violate any
of the constraints.

(20)
Head Modifier
Reference – –
These were the PoS systems predicted by both the original PoS hierarchy in
(5) and the implicational map in Figure 2.8. As mentioned, the latter model
predicts another 10 possible PoS systems, which would have presented
counterexamples to the former model. These systems are represented in
(21)-(29) below.
The system in (21) is in accordance with constraint (11b), since it has a
class of flexible lexemes that can be used as referential heads (nominals) and
a rigid class for predicative heads (verbs). It is also in accordance with (12a)
in the predication domain, where it has rigid modifiers (manner adverbs) and
rigid heads. Finally, it conforms to (13), since it has distinct classes for heads
and modifiers in the predicate phrase, as well as distinct classes for heads in
predicative versus referential phrases.
(21)
Head Modifier
Reference Nominal
The system in (22) is in accordance with constraint (11b), since it has a

class of flexible lexemes that can be used as the head of a referential phrase
(nominals), and a class of rigid lexemes for the function of head of a predicate
phrase (verbs). Constraint (12) and (13) are not applicable.
(22)
Head Modifier
Reference Nominal
The system in (23) involves a lexical distinction between predication and

reference, but not between heads and modifiers. The PoS classes in this

system are termed predicatives and nominals (see also Hengeveld & Van Lier
2009). The system is in accordance with (11b). Constraints (12) and (13) are
not applicable.
(23)
Head Modifier
Predication Predicative
Reference Nominal
The system in (24) conforms to constraint (11b), since it has a flexible class
of lexemes that can be used predicatively (predicatives), while not having a
class of lexemes that can be used referentially. Constraints (12) and (13) do
not apply.
(24)
Head Modifier
Reference – –
The system in (25) involves a single flexible class of heads, i.e. lexemes that
can be used as the head of both predicative and referential phrases. This
system does not violate any of the constraints.
(25)
Head Modifier
Predication –
Reference Head –
The system in (26) also involves a single flexible class, but one that can be
used in all functions except modifier in a predicate phrase. This system does
not violate any of the constraints 19. For want of a good candidate, no special
term
19
Note that the lack of a lexeme class for the function of modifier in a predicate phrase does
not mean that there is a lexical head-modifier distinction within the functional domain of
predication.

is proposed for this type of PoS class. It is simply indicated as ‘flexible class A’
(Flex A); its functional distribution can be read off the representation.
(26)
Head Modifier
Predication –
Reference Flex A
The system in (27) is similar to the one in (26), except that the flexible class
can now be used in all functions except modifier in a referential phrase.
Again, no special term is introduced for this type of flexible lexeme class; it
is indicated as Flex B.
(27)
Head Modifier
Predication
Reference Flex B –
The system in (28) involves a rigid class of verbs and a flexible class of
lexemes, termed Flex C, which can be used as the head of a referential phrase
and a modifier in a predicate phrase, but not as a modifier in a referential
phrase. As such, it is in accordance with constraint (11b), with (12b) in the
predication domain, and with (13).
(28)
Head Modifier
Predication Verb Flex C
Reference Flex C –
The system in (29) involves two ‘cross-wise’ flexible lexeme classes: One
class (Flex C; see above) that can be used as the head of a referential phrase
and as a modifier in a predicate phrase, while the other class, termed Flex
D, can be used as the head of a predicate phrase and as a modifier in a
referential phrase. This system does not violate any constraint.

(29)
Head Modifier
Predication Flex D Flex C
Reference Flex C Flex D
Finally, the system in (30) has rigid lexeme classes for heads and modifiers
in the predication domain (verbs and manner adverbs), while in the reference
domain it has a rigid lexeme class for heads only (nouns). This system is in
accordance with constraints (11a), (12a) and (13).
(30)
Head Modifier
Reference Noun –
In sum, the implicational map model in Figure 2.8 predicts 17 possible PoS
systems (of which 7 were also predicted by the original PoS hierarchy),
which are expected to occur in actual languages. In addition however, the
model predicts that a large number of logically possible PoS systems will not
occur in any language. In particular, 34 out of 51 logically possible systems
are excluded by the model, as they violate one or more of the constraints
in (11), (12) and (13) above. For expositional reasons, these excluded systems
are listed separately in Appendix ii, where I specify for each system which
constraint(s) it violates. In Chapter 5 the complete set of predicted and
excluded PoS systems will be compared with the PoS systems that are (un-)
attested in the sample languages.
At this point, it should be noted that the PoS systems presented above
do not include any intermediate systems. It is expected, however, that the
predicted systems can indeed be supplemented with additional PoS classes,
as long as this does not violate the constraints in (11)-(13). As I mentioned
earlier, these ‘extra’ classes need not be contiguous to the rightmost large
basic class of a system, in terms of the original PoS hierarchy. Moreover, and
again contrary to the predictions of the original hierarchy, extra classes are
not necessarily derived, small and / or closed. Rather, they can also be large,
open classes consisting of simple members. This will be illustrated in full

detail in Chapter 5. However, to give a flavour of the attested cross-linguistic
variation in PoS systems, I provide some examples below.
First, languages with flexible PoS systems may have extra classes that
are large and open rather than derived. Santali, for instance, combines a
maximally flexible PoS class of contentives with a large, open class of rigid
verbs (Neukom 2001: 17). This is represented in Figure 2.9. Note that this
system would have been a counterexample to the original theory, which
predicts that an ‘intermediate’ system with contentives can only involve a
class of derived non-verbs (cf. Figure 2.7).
Head Modifier
Predication Derived nouns
Reference
Contentives
Figure 2.9: The PoS system of Santali
Second, consider Garo, a language with a rigid PoS system. It has large open
classes of verbs and nouns, no adjectives, a small class of simple manner
adverbs, and a large class of verb-derived manner adverbs (Burling 2004).
This is represented in Figure 2.10. The PoS system of Garo would have
been a counterexample to the original PoS hierarchy, since there is no lexical
class available for the function that is positioned between the functions of
modifier in a predicate phrase and head of a referential phrase. Garo’s PoS
system does not fit into the extended typology of PoS systems in Figure 2.7
either, since the latter does not take into account the possibility of derived
classes in rigid systems.
Head Modifier
Predication Verbs Derived / small class of
simple manner adverbs
Reference Nouns –
Figure 2.10: The PoS system of Garo
These examples give a glimpse of the diversity encountered in the realm of

PoS systems across languages, showing that the range of variation is much
wider than Hengeveld’s original typology suggests.

2.3.3 Summary
In this section I have presented Hengeveld’s theory of lexical categorization,
from its original formulation to its most recent version. I explained that the
Hengeveldian approach has two crucial characteristics. The first one is the
exclusion of semantics as a criterion for the definition of PoS. The second
important component involves the ‘further morpho-syntactic measures’
that do or do not accompany the use of a particular group of lexemes in a
particular functional environment. Such further measures include structural
coding only; behavioural potential, or morphological criteria, are not taken
into account.
Hengeveld’s approach avoids (but does not provide a solution to) the
problems associated with semantic and morphological criteria for PoS
definitions, as pointed out in section 2.2. In particular, the Hengeveldian
method allows for a straightforward cross-linguistic comparison of the
various ways in which languages may map groups of lexemes onto a set
of propositional functions. However, this convenient situation comes at a
cost. In the next section I discuss the problems associated with Hengeveld’s
method, and the alternatives that have been proposed for it.
2.4 Critique on the syntactic approach
2.4.1 Introduction
The most elaborate and critical alternative to Hengeveld’s approach is
Croft’s Universal-Typological Theory of Parts of Speech (2000, 2001). The
crucial difference between the two approaches is that Croft’s theory, unlike
Hengeveld’s, takes into account semantic and morphological criteria.
According to Croft, leaving out these two types of linguistic data implies
missing out on the cross-linguistic generalizations that apply to them.
First, in section 2.4.2, I start with a brief outline of Croft’s theory.
Subsequently, in section 2.4.3, I discuss in more detail the specific objections
that Croft has raised against Hengeveld’s approach. In fact, these objections
are not confined to Hengeveld’s PoS theory. Rather, they target any theory of
linguistic categories that fails to take into account the full range of linguistic
evidence, and that fails to acknowledge that such a complete distributional
analysis inevitably leads to the conclusion that cross-linguistically identical
categories do not exist 20.
20
For a recent general assessment of this fundamental issue in linguistic typology, the reader
is referred to Haspelmath (2007) and Newmeyer (2007).

Finally, one particular aspect of Croft’s critique on Hengeveld’s method
concerns the role of semantics in languages with flexible word classes.
Since this is a complex issue that has been the subject of a rather extensive
discussion in recent literature, I treat it separately in section 2.5.
2.4.2 Croft’s Universal-Typological Theory of Parts of Speech

Croft (2000, 2001) distinguishes three functions that lexemes may express:
predication, reference, and modification. These are termed propositional act
functions. Although this resembles the Hengeveldian model, in the latter the
modifier function is treated as a dimension of the head-modifier distinction,
which cross-cuts the predication-reference distinction, rather than as a basic
function in itself (cf. note 15).
A much more important difference between the two theories is that
Croft divides the lexemes that may fulfil the three propositional act functions
into three different semantic classes: objects, properties, and actions. These
semantic classes are in turn defined in terms of the following four semantic
properties (Croft 2001: 87):
(i) R elationality: whether a definition of a concept inherently requires

reference to another concept;
(ii) Stativity: whether the concept represents a state or a process;
(iii) Transitoriness: whether the concept represents a transitory state or
process or an inherent or permanent state of the entity in question;
(iv) Gradability: whether the entity is graded along a scalar dimension.
The definitions of the semantic classes in terms of these properties are

presented in Figure 2.11 below (taken from Croft 2001: 87)21.
relationality stativity transitoriness gradability

Objects non-relational state permanent non-gradable
Properties relational state permanent / gradable
transitory
Actions relational process transitory non-gradable
Figure 2.11: Properties of semantic classes
In principle, lexemes belonging to any of the three semantic classes

(objects, properties, and actions) may express any of the three propositional
act functions (reference, modification, and predication). However, the

assumption of the prototype approach (first formulated in Croft 1991) is
that the semantic classes of objects, properties, and actions are the typological
prototypes of referring, modifying, and predicating constructions, respectively.
Croft thus defines PoS as prototypical or unmarked combinations of
semantic class and propositional function. A prototypical noun denotes an
object and functions referentially; a prototypical adjective denotes a property
and functions as a modifier; and a prototypical verb denotes an action and
functions predicatively. This is represented in Figure 2.12 (adapted from
Table 2.3. in Croft 2001: 88). Any combination of propositional act function
and semantic class other than the three prototypical ones represented in
Figure 2.12 is claimed to be typologically marked.
reference modification predication

Objects Unmarked nouns
Properties Unmarked adjectives
Actions Unmarked verbs
Figure 2.12: Nouns, adjectives and verbs as prototypical combinations of propositional act
functions and semantic classes
Importantly, although Croft defines PoS as typological prototypes, rather

than as language-specific categories, this by no means implies that he regards
PoS-typology as a pointless enterprise. On the contrary, the conceptual
notion of typological markedness (versus unmarkedness or prototypicality),
as it applies to the combinations of propositional functions and semantic
categories, correlates cross-linguistically with formal phenomena. In
21
In an earlier version of his theory (1991), Croft uses different labels for some of the
semantic properties listed in Figure 2.11. Pustet (2000, 2003) draws heavily on Croft’s
work, but she proposes a slightly different semantic system. Apart from the property of
transitivity / valency, which is the same as Croft’s (1991), she introduces a new semantic
parameter, namely energy release (renamed dynamicity in Pustet 2003) and leaves out Croft’s
parameter of gradability. The latter is apparently integrated, together with perfectivity and
processuality (corresponding to Croft’s stativity and persistence / transitoriness), into the notion
of transcience. Just for the sake of clarity, this rather confusing terminological situation is
represented in the schema below:
Croft 1991 Croft 2001 Pustet 2000 Pustet 2003
valency relationality transitivity transitivity / valence
– – energy release dynamicity
persistence transitoriness perfectivity –
stativity stativity processuality transcience
gradability gradability – –

particular, Croft identifies two implicational universals, which constrain
cross-linguistic structural variation in terms of both structural coding and
behavioural potential.
The first universal, pertaining to structural coding, predicts that a
typologically marked combination of pragmatic function and semantic
class must be encoded by at least as many morphemes as an unmarked
combination. The second universal makes the opposite prediction for
behavioural potential: It states that the range of behavioural potential
displayed in the case of an unmarked combination of pragmatic function and
semantic class must be at least as wide as the range displayed in the case of a
marked combination (Croft 2001: 90, 91). Notably, both universals allow for
the possibility of a typologically marked combination showing exactly the
same amount of morpho-syntactic measures as an unmarked combination,
including zero-marking 22. What is excluded is a situation in which a marked
combination shows less structural coding or more behavioural potential than
an unmarked combination 23.
In sum, Croft’s PoS theory defines a universal conceptual space that
is made up of a pragmatic (propositional) and a semantic (denotational)
dimension. Onto this conceptual space, a set of language-specific lexical
categories is mapped, which are identified on the basis of a distributional
analysis of formal criteria, including both structural coding and behavioural
potential. Croft’s theory and Hengeveld’s theory thus resemble one another
to the extent that both are concerned with form-function mappings
(or ‘construction-filler interactions’, as Croft (2005: 437) calls them).
However, their approaches differ in two fundamental respects. First, Croft’s
functional space includes a semantic dimension, which is lacking in the
Hengeveldian model. Second, Croft is interested not only in structural
coding – as Hengeveld is in further morpho-syntactic measures –, but also
takes into account all other distributional properties of lexemes, i.e. their
behavioural potential.
According to Croft, the theoretical repercussions of using only a subset
of morpho-syntactic criteria are serious. He argues that constructions or
22
As far as structural coding is concerned, this would amount to Hengeveld’s definition of
flexibility.
23
Note that the amount of marking is measured in terms of the number of morphemes (Croft
2001: 90, 91). This is not a very straightforward method, though, since it is not clear how to
treat for instance portmanteau morphemes. Therefore, it would probably be better to count
the number of semantic categories expressed on the lexeme, either by bound (agglutinative or
fusional) morphemes or by analytic grammatical particles.

categories (including PoS classes) identified in this way are “epiphenomenal”
and that “one will not find language universals that way” (Croft 2005: 436). In
what follows, I discuss these claims in some more detail.
2.4.3 Lexical categories within and across languages:

The generality and the subclass problems
Croft’s main objection to Hengeveld’s PoS theory is that it is selective about
which distributional tests are taken into account (structural coding) and which
are not (behavioural potential). According to Croft, Hengeveld’s approach
amounts to “methodological opportunism”, which can be avoided only by
taking into account the complete set of distributional facts in each individual
language. This, he argues, will reveal “covert categories, that is, categories not
obligatorily flagged by overt structural coding” (Croft 2001: 75; 2005: 434) 24.
However, the approach advocated by Croft raises two potential problems,
which are in close relation to each other, and which have been termed the
subclass problem and the generality problem (Haspelmath 2001). The former
problem is concerned with intra-linguistic lexical categorization; the latter
with inter-linguistic comparability of lexical categories.
Starting with the subclass problem, if one takes into account the full set
of distributional facts about groups of lexemes in a specific language, this
gives rise to a “myriad” of different types of sub-categorizations, without any
means to decide which of these categories are to be considered ‘PoS classes’
and which ones are not (Croft 2001: 78ff, Haspelmath 2001: 16540). This
problem is also aptly described by Schachter and Shopen:
“It must be acknowledged […] that there is not always a clear basis
for deciding whether two distinguishable open classes of words that
occur in a language should be identified as different parts of speech or
as subclasses of a single part of speech. The reason for this is that the
open parts of speech classes must be distinguished from one another on
the basis of a cluster of properties […]. Typically there is some overlap,
some sharing of features, as well as some differentiation. […] What this
means is that there may in some cases be considerable arbitrariness in
the identification of two open word classes as distinct parts of speech
24
Croft (2001: 73) also points out that a PoS theory like Hengeveld’s, which lacks a semantic
dimension, is bound to overlook regularities in the relevant domain, such as those applicable
to semantic shifts in flexible lexemes. This issue will be discussed further in section 2.5.

rather than subclasses of a single part of speech.” (Schachter & Shopen
2007: 4; emphasis in the original)
Croft (2001: 81-83, following Uehara 1998) illustrates this point with an
example from Japanese, which I will discuss here in some detail. In Japanese,
the boundary between nouns and adjectives is not clear-cut. Traditionally,
three categories are distinguished: ‘nouns’, ‘adjectives’, and ‘nominal
adjectives’. ‘Nouns’ can function without structural coding as the head of
a referential phrase. When they are used as the head of a predicate phrase
they need a copula, and when they are used as a modifier in a referential
phrase they take a genitive particle no. ‘Nominal adjectives’ behave in the
same way as nouns when functioning as the head of a referential phrase
or the head of a predicate phrase, but they take the linking particle na
when used in modifier function. ‘Adjectives’ can not be used as heads in
referential phrases. In contrast, they can be used without structural coding
as referential modifiers and as heads of predicate phrases. In both functions,
the behavioural potential of adjectives consists of tense-marking: -i for non-
past and -katta for past (see Hinds 1986: 44) 25.
However, a number of other lexical classes may be distinguished in
Japanese with distributional patterns that seem to fall somewhere in between
the patterns of ‘nouns’, ‘nominal adjectives’, and ‘adjectives’. First, there is
a class of lexemes that sometimes behave like nouns and sometimes like
nominal adjectives. In particular, when functioning as modifiers, these items
can appear either with the genitive marker no, like nouns (see (31a)), or with
the linker na, like nominal adjectives (see (31b)):
Japanese (Croft 2001: 82)

(31) a. heiwa no sisya
peace gen messenger
‘a messenger of peace’
b. heiwa na kuni
peace(ful) lk country
‘a peaceful country’
25
In Chapter 5 I will interpret the fact that ‘adjectives’ inflect for tense when used as modifiers
as evidence that these lexemes are in fact verbs that may be used as dependent predicates in
(zero-marked) relative clauses. At present, however, I will follow the argument as presented
by Croft.

The members of another such ‘in-between class’ behave partly like nominal
adjectives and partly like adjectives. One part occurs either with a copula
when used as the head of a predicate phrase, or without a copula and with
tense-marking (see (32a)). In modifier function these items can take either
the linker na or zero structural coding with tense inflection (see (32b))).
(32) a. ataka-i / ataka da

warm-prs / warm cop
‘It is warm.’
b. ataka-i hi / ataka na hi
warm-prs day / warm lk day
‘a warm day’
The second part of this subclass displays alternative coding strategies only in
the modifier function (na versus zero structural coding and tense marking, as
in (33a), cf. (32b)). When used as the head of a predicate phrase the relevant
lexemes always occur without a copula and with tense marking (see (33b)
and cf. (32a)):
(33) a. tiisa-i hon / tiisa na hon

small-prs book / small lk book
‘a small book’
b. tiisa-i / *tiisa da
small-prs / small cop
‘It is small.’

These distributional facts are summarized in Figure 2.13:
PoS class Function / Coding

Head of Pred Head of Ref Mod in Ref
Phrase Phrase Phrase
Noun da (COP) Ø no (GEN)
Noun / nominal adjective da (COP) Ø no(GEN) / na (LK)
Nominal adjective da (COP) Ø na (LK)
Nominal da (COP) / Ø + Ø na (LK) / Ø + Tense
adjective / adjective 1 Tense
Nominal Ø + Tense Ø na (LK) / Ø + Tense
adjective / adjective 2
Adjective Ø + Tense X Ø + Tense
Figure 2.13: The noun-adjective distinction in Japanese
The crucial question is now whether one should interpret all these different
distributional patterns as evidence for as many PoS classes in Japanese. If
so, then a Hengeveldian typology of PoS systems would be split up into
a potentially unlimited number of sub-classifications. For Croft’s theory,
however, this problem is irrelevant: From his point of view, it is simply
not necessary to decide whether the classes identified above are nouns or
adjectives, or both, or neither. They just represent a set of language-specific
formal categories mapped onto a universal functional space.
A direct consequence of the subclass problem is the generality
problem: lexical categories like the Japanese ones identified in Figure 2.13
are not comparable with lexical categories attested in any other individual
language. Since languages differ in terms of the number and type of relevant
distributional criteria, a lexeme class that is defined on the basis of such
criteria in one language can never be identical to a lexeme class in another
language, which expresses different morpho-syntactic distinctions. Again,
this is a non-issue in terms of Croft’s prototype approach. The semantic
dimension of his theory accounts for cross-linguistic generalizations about
the ways in which individual languages cut up the universal functional
space cake: the less prototypical the combination of semantic meaning and
propositional function, the more structural coding will be required, and the
less behavioural potential will be displayed.
The Hengeveldian approach, in contrast, ignores fuzzy boundaries
between lexeme classes, such as between nouns and adjectives in Japanese.
According to this approach, the relevant facts for Japanese are that it has

(i) a class of lexemes that can be used without structural coding as the
head of a referential phrase (i.e. Hengeveld’s definition of nouns), and (ii)
a class of lexemes that can be used without structural coding as a modifier
in a referential phrase (i.e. Hengeveld’s definition of adjectives). Nouns and
adjectives (and other PoS) defined along these lines are not universal in the
way that Croft’s prototypes are universal, and they are not language-specific in
the Croftian sense either (i.e. they are not based on a complete distributional
analysis of a the particular language system of Japanese). However, these
PoS are cross-linguistically comparable, even though they can be considered
identical across languages only as far as Hengeveld’s narrow (or should one
say ‘broad’) definitions go. Adding more analytical primitives, as Croft does,
inevitably leads to more categorial differentiation.
2.4.4 Summary, outlook

In sum, Croft’s method of identifying lexical categories gives rise to the
subclass problem and the generality problem, which concern the intra-
linguistic definition and inter-linguistic comparability of PoS classes.
However, Croft avoids these problems by defining PoS as prototypes rather
than as language-specific categories, and by identifying cross-linguistic
generalizations about patterns of relative formal markedness, which make
reference to both structural coding and behavioural potential. In contrast,
Hengeveld’s approach to lexical categorization avoids the subclass and
generality problems by looking at a restricted set of distributional data,
namely at the presence versus absence of structural coding accompanying the
use of particular groups of lexemes in particular propositional functions.
In section 2.5 I turn to another problematic issue that has been raised
in relation to Hengeveld’s PoS theory. This issue concerns the absence of a
semantic dimension in relation to the problem of defining lexical flexibility.
In particular, since the Hengeveldian theory does not take into account the
semantic denotation of lexemes, it has nothing to say about the semantic
interpretation of supposedly flexible lexemes in the various functions that
these items may express. However, claims about the existence of lexical
flexibility (in general as well as with reference to particular languages) have
often been challenged on semantic grounds: Even though presumed flexible
lexemes do not change formally when used in various functions, they may
have different (if related) meanings in each of those functions. Scholars
disagree about the origin and theoretical significance of such semantic shifts.
In addition to the problem of semantic interpretation, a number of other

issues have entered the discussion about lexical flexibility. The following
section reviews this discussion.
2.5 Defining lexical flexibility
2.5.1 Introduction
This section addresses the debate surrounding so-called flexible languages.
As a point of departure for the discussion, I use the three criteria for
lexical flexibility proposed by Evans and Osada (2005). The first and most
complicated criterion, to be discussed in section 2.5.2, is concerned with semantic
(non-)compositionality, i.e. with the semantic interpretation of presumably
flexible lexemes in different functional environments. In section 2.5.3
I take up the criterion of exhaustiveness, which relates to the problem of
quantitative measurement of lexical flexibility. Finally, in section 2.5.4,
I discuss the criterion of equivalent combinatorics, which pertains to the issue
of bi-directional lexical flexibility.
2.5.2 The criterion of compositionality: flexibility and semantic shift
2.5.2.1 Evans and Osada’s proposal and some alternatives

Presumably flexible lexemes, i.e. lexemes that can be used without further
measures in two or more propositional functions, receive slightly different
semantic interpretations, depending on the particular function in which they
appear. Although these meanings are (closely) related, particular meaning
shifts may differ in terms of their degree of regularity: while some are
apparently very predictable, others seem to be rather more idiosyncratic.
Evans and Osada’s (2005) criterion of semantic compositionality states
that, for a lexeme to be ‘truly’ flexible,
“[A]ny semantic differences between the uses of a putative ‘fluid’ [i.e.

‘flexible’, EvL] lexeme in two syntactic positions must be attributable to
the function of that position.” (Evans & Osada 2005: 367)
A corollary of this criterion is that the meaning shift, or what Evans and
Osada call the coercion effect, should be the same for any member of a
flexible lexeme class that is used in a particular function, apart from possible
semantic interactions attributable to behavioural potential that belongs to
the function, such as for instance aspect markers. According to Evans and

Osada, if the compositionality criterion is not satisfied, i.e. if the semantic
interpretation of a flexible lexeme in different functional environments is
not entirely compositional, then what looks like flexibility should in fact
be analyzed as ‘rampant’ zero-conversion: large-scale derivation from one
lexical category to another without any formal reflection.
Similarly, Croft (2000, 2001, 2005) argues that cases of non-
compositional semantic shifts are not instances of flexibility. Rather, he
argues, they are instances of polysemy, i.e. they involve two or more related
meanings associated with a single form. Since the relation between these
various meanings is not predictable, each of them must be stored individually
in the lexicon, and must be labelled as appropriate in a particular functional
environment. This amounts to the same kind of categorization that is found
in languages with rigid PoS systems.
As we have seen, Hengeveld’s original PoS theory makes no reference
to semantics, and therefore it does not account for meaning shifts in flexible
lexemes. In Croft’s words: “Hengeveld ignores what happens to a lexical root’s
meaning when used in more than one function” (Croft 2000: 69). This point is
illustrated with examples from Tongan, which were adduced by Hengeveld
(1992: 66) as evidence for analyzing Tongan as a flexible language. According
to Hengeveld, the examples in (34) demonstrate that the lexeme si’i is a
flexible lexeme, because it can be used without further measures as the head
of a predicate phrase (34a), the head of a referential phrase (34b), and a
modifier in a referential phrase (34c) 26. In contrast, Croft argues that this
analysis is problematic, since it does not account for the different meanings
of si’i that are associated with each of its various uses.
(34) a. [Na’e si’i] ‘ae akó

pst small abs school:def
‘The school was small.’
b. ‘i [‘ene si’í]
in 3sg.poss childhood:def
‘in his / her childhood’
26
Note in passing that ako presumably also represents an instance of lexical flexibility.

c. Na’e ako [‘a tamasi’i si’i] iate au
pst study abs child small loc 1sg
‘The small child studied at my house.’
In response to Croft’s critique, Hengeveld and his colleagues argue that

there is no need to assume multiple meanings for flexible lexemes like si’i
(Hengeveld et al. 2004, Hengeveld & Rijkhoff 2005). In other words, they
regard flexible lexemes as monosemous rather than polysemous. Meaning
shifts, they argue, come about because the use of a flexible lexeme in a
particular function activates certain aspects of the lexeme’s general meaning.
This view is clearly stated in the following quotation:
“[E]ach flexible lexeme has a single (vague) sense. By placing the flexible
lexeme in a particular slot or by providing it with certain morphological
markers 27, the speaker highlights those meaning components of the
flexible lexemes that are relevant for a certain lexical (verbal, nominal,
etc.) function. Thus we contend that the meaning of a flexible lexeme
always remains the same, and that morpho-syntactic and other contextual
clues signal to the addressee how to interpret this lexeme in an actual
utterance. In other words, it is the use of a vague lexeme in a certain
context (an actual linguistic expression) that brings out certain parts of
the meaning, giving the category neutral lexeme a particular categorical
(verbal, nominal, etc.) flavour.” (Hengeveld et al. 2004: 541)
However, Croft (2001: 71, 72) argues that problems with this vagueness
approach arise in defining the meaning of a lexeme in such a way that it is
neutral with respect to the various interpretations associated with it in each
of its possible functions. He thus claims that it is impossible to come up with
a definition that covers all and only the meanings of a flexible lexeme like
Tongan si’i. A definition such as ‘some concept associated with smallness’
is too general because it may include all kinds of concepts that have the
property of ‘smallness’. As mentioned above, the only alternative, according
to Croft, is to specify by convention two or more different meanings of the
lexeme in question, and to store these meanings together with the function to
which they belong. This analysis undermines the very notion of flexibility.
27
The term ‘morphological markers’ refers to behavioural potential, not to structural coding;
the presence of the latter would obviously block the analysis as a flexible lexeme. (Kees
Hengeveld, p.c.)

Considering the Tongan data in (34) in light of Evans and Osada’s criterion
of semantic compositionality, one may argue that the shift in the meaning
of si’í in (34a) (‘be small’) versus (34c) (‘small’) is indeed attributable to the
different functional environments in which the lexeme appears, namely
predication and attribution, respectively. In contrast, this type of analysis
cannot easily be applied to (34b), where si’í means ‘childhood’ (or ‘period
in one’s life during which one is small’). This meaning does not follow
predictably from the fact that si’í functions as the head of referential phrase
in this example. Meanings such as ‘smallness’ or ‘small one’ appear to be at
least as plausible in this function. This would suggest that Tongan does not
meet the compositionality criterion, at least not across the board.
However, Hengeveld and Rijkhoff (2005) also disagree with Evans and
Osada’s assumption that a functional environment adds semantic content
to a flexible lexeme that it did not have before it was used. As mentioned,
Hengeveld and his colleagues claim that flexible lexemes are semantically
vague (cf. the above quotation). This means that both the verbal and
the nominal senses are already contained in the semantics, and that the
functional slot “only highlights meaning components that are already there (…)”
(Hengeveld & Rijkhoff 2005: 514). Thus, according to Hengeveld and
Rijkhoff, non-compositional meaning shifts are not an argument against
flexibility. To the contrary, they say that semantic idiosyncrasies should be
expected, since each lexeme has its own unique set of meaning components,
which in combination with a particular functional environment yields the
intended semantic interpretation.
This line of thought links up with work in the framework of Cognitive
Grammar, as developed by Langacker (1987). As mentioned briefly in
section 2.2, in Cognitive Grammar the meanings of lexemes are defined as
image-schematic concepts. Verb meanings are conceptualized as temporal
sequences and are therefore sequentially scanned. Nouns, on the other hand,
are conceptualized as bounded regions. If the meaning of a noun includes
the notion of some temporal unfolding (like a verb), then it is scanned
summarily, as a whole, rather than sequentially. Within this cognitive
framework, flexible lexemes can be defined as having a single conceptual
structure that is compatible with either a verbal or a nominal meaning.
Depending on the functional frame in which the lexeme is used, a profiling
of the concept is triggered that corresponds to the appropriate semantic
interpretation of the word.

Farrell (2001) applies this cognitive approach to English lexemes. Even
though English is probably most often thought of as a language with rigid
PoS classes, it has long been noted that many lexemes in this language can
indeed be used as nouns and verbs without any overt derivation (Clark &
Clark 1979)28. Farrell argues that any English lexeme – irrespective of the
behaviour of other lexemes in the language – that can occur in two different
functions without derivational marking does not belong to a part of speech
until it is inserted into a syntactic frame. In other words, he claims that at
least some proportion of the English lexicon is flexible.
Consider for instance the lexeme kiss. Farrell analyzes this item as
having a conceptual structure that defines contact between the lips of one
entity, the agent, with another entity, the patient. This is shown in the upper
box of Figure 2.14 (taken from Farrell (2001: 114, Figure 2)). When kiss is
used as the head of a predicate phrase, its action-interpretation is profiled,
as represented in the lower left box of Figure 2.14 by the bold lines of the
temporal t-axis and the oval and circle designating the participants. If, on
the other hand, kiss is used as the head of a referential phrase, then the
abstract region defining the overall event is profiled, as represented in the
lower right box by the bold print of the outermost oval.
According to Farrell, kiss is an example of a so-called process-centred
word. There are also thing-centred words. An example of the latter would be
hammer. The analysis of thing-centred words differs to the extent that using
the lexeme in referential function triggers the profiling of one of the entities
involved in the conceptual representation – the hammer in this case –, rather
then the overall process, as in the case of kiss. As Farrell himself admits,
however, there are many other possible types of semantic shifts, which do
not correspond to the two types just discussed.
In fact, this seems to be exactly what also constitutes the flaw in
Hengeveld and Rijkhoff ’s analysis of flexibility: Even if both the lexeme’s
particular semantic features and its functional environment are taken into
account, it is not always predictable exactly which meaning elements are
to be profiled in a particular function. On the other hand, an advantage
of Hengeveld and Rijkhoff ’s vagueness analysis and Farrell’s cognitive
analysis is that neither of them needs to posit a process of conversion, that is,
28
See also Evans & Osada (2005) (and the discussion further on in this section), who
analysed samples of 120 lexemes in Mundari (a presumably flexible language) and English,
and found that around 75% of the Mundari lexemes could function as verbs and nouns,
against around 65% of the English items.

zero-derivation from one lexical category to a new one, without any formal
evidence for such a transition. In other words, under both analyses the lack
of overt morphological marking of a category change is explained by the fact
that there is no such change going on in the first place.
lips
kiss
lips lips
t t
Predicating Referring
Figure 2.14: The conceptual structure of kiss.
In general, the above discussion makes clear that some types of meaning shifts
accompanying the use of formally identical lexemes in different functional
environments appear to be highly regular, while others seem to be less
predictable. In addition, it has been shown that there are, broadly speaking,
two views concerning the relation between lexical flexibility and semantic
shift: One claiming that ‘true’ flexibility involves purely compositional
semantics, and another claiming that idiosyncratic interpretations are
expected on the basis of the specific meaning components that are part of
the unique semantics of any flexible lexeme. In the next section, I will show
that in fact both regular and irregular types of semantic interpretations occur
in languages for which pervasive lexical flexibility has been claimed. It will
be argued that the relevant data can be understood in terms of a mismatch
between lexical and phrase-structural categorization in these languages.
2.5.2.2 An integrative approach to semantics in flexible languages

Theories about PoS often take for granted that there is a one-to-one
correlation between lexical categories and syntactic categories. However,
Sasse (1993a, b) explicitly advocates a clear distinction between these two
types or levels of categorization. Following Sasse’s view, Himmelmann
characterizes the levels as follows:

“One level is the level of terminal syntactic categories where lexical items
are categorized according to their phrase-structural properties. (…).
The second level is the level of lexical categories proper where lexical
items are categorized according to those grammatical features which
are not directly relevant for phrase structure […] (i.e. phonological,
morphological, and syntactic features).” (Himmelmann 2007: 263)
Making this distinction implies that there may or may not be a correlation
between the categories distinguished at one level and those distinguished at
the other level. The former situation, i.e. a one-to-one correlation between
lexical and syntactic categorization, is typically attested in rigid languages,
including many Indo-European languages. The latter option involves a
mismatch between lexical and syntactic categories. A well-known example
of a language for which this has been claimed is Tongan (see also the previous
section). According to Broschart (1997), lexical categories in Tongan are
defined in terms of morphological features, which have no relevance for
syntactic distribution. Himmelmann (2007) makes a very similar claim for
Tagalog: According to him, there are a number of morphologically defined
lexical categories in Tagalog, which are not projected into the syntax. There
is some kind of correlation between these so-called morpho-lexical categories
and purely ontologically defined categories, but this correlation is far from
perfect. This means that the lexical categories of Tagalog are definitely
grammatical categories, even though they are not syntactic categories.
Don and Van Lier (forthcoming) relate the issue of compositional
and non-compositional semantic interpretation in flexible languages to a
mismatch between lexical and syntactic categorization in these languages.
They assume that, at a sufficiently abstract level of representation, all languages
have a set of un-categorized roots with certain ontological meanings. In
rigid languages, the operation that assigns these roots to a lexical category
also automatically determines their phrase-structural destination: Once
it is categorized, a lexical item can be used in only one syntactic function
(without further measures). Flexible languages, in contrast, are characterized
by the fact that they can lexically categorize and re-categorize their roots
without affecting the phrase-structural possibilities of the output forms.
Such non-syntactic (re-)categorization may or may not involve overt formal
marking. The crucial point is that, since this type of lexical categorization
occurs prior to syntactic categorization, it may involve non-compositional
semantic interpretations. In contrast, syntactic categorization, i.e. the

ultimate insertion of lexical material into a syntactic function, should yield
fully compositional semantic interpretations. Typically, the latter type of
categorization involves syntactic frames made up of tense-aspect-mood
(TAM) operators (for verbal / predicative usage), or determiners and / or case
markers (for nominal / referential usage).
Consider a language like Samoan. In this language, according to Mosel
and Hovdhaugen (1992: 77), “the categorisation of words into nouns and verbs
is not given a priori in the lexicon. It is only their actual occurrence in a particular
syntactic environment which gives them the status of a verb or a noun”. When
a lexical item is preceded by a TAM particle it functions syntactically as a
verb, and when it combines with one of the following particles it functions
syntactically as a noun: le (determiner), e (ergative; absolutive is zero-
marked), o (possessive), or i (marker of non-core arguments and locative-
directional adjuncts).
The semantic interpretation of lexemes in verbal versus nominal
syntactic function is often completely predictable, as is illustrated in examples
(35) and (36). In example (35b), the action-denoting root alu ‘go’ occurs in
nominal function and means ‘(the action of ) going’. In example (36a) the
object-denoting root uō ‘friend’ occurs in verbal function and is interpreted
compositionally as ‘be friends’.
Samoan (Mosel & Hovdhaugen 1992: 73, 77)

(35) a. E alu le pasi i Apia
genr go det bus dir Apia
‘The bus goes to Apia.’
b. le alu o le pasi i Apia

det go poss det bus dir Apia
‘the going of the bus to Apia.’
(36) a. E uō Tanielu ma Ionatana

genr friend Daniel and Jonathan
‘Daniel and Jonathan are friends.’
b. E alofa Taniel i l=a=na uō

genr love Daniel dir det=poss=3sg friend
‘Daniel loves his friend.’

However, there are also many other possible semantic shifts, with less
predictable interpretations. Some examples are given in (37). Note that the
semantic relations between the paired forms make reference to actions and
participants involved in those actions (actors, instruments, undergoers) or
specific instances of those actions.

(37) a. fana: ‘gun’ (N) or ‘shoot’ (V)
b. lama: ‘torch’ (N) or ‘fish by torch light’ (V)
c. gaoi: ‘steal’ (V) or ‘thief ’ (N)
d. solo: ‘move forward’ (V) or ‘procession’ (N)
e. tusi: ‘write’ (V) or ‘letter / book’ (N)
The semantic shifts that go with the conversions in the examples in (37) are
not marked by any overt formal change. However, overt lexical derivation
does occur in Samoan, in particular with the suffix -ga. Some examples are
given in (38). Interestingly, the output forms of this process have irregular
semantic patterns, which at least partly resemble the ones in (37). The derived
forms may denote a participant, a location, or an institution associated with
the action denoted by the base form, but also a specific instance of that
action.
Samoan (Mosel & Hovdhaugen 1992: 195)

(38) a. amo ‘carry’ - āmo-ga ‘person(s) carrying loads’
b. a’o ‘teach’ - ā’o-ga ‘school’
c. tīpi ‘cut’ - tīpi-ga ‘surgical operation’
d. pule ‘control’ - pulē-ga ‘unit of church administration’
Since the semantic meanings of the derived forms in (38) are typically
‘nominal’, while those of the un-derived ones are ‘verbal’, -ga has traditionally
been termed a nominalization suffix (Mosel & Hovdhaugen 1992: 194).
However, according to Mosel (2004: 267) “words carrying the so-called
nominalization suffix can occur in the head position of a verb complex.” In other
words, -ga is not a syntactically categorizing derivational suffix: it creates a
new lexical item with a particular semantic meaning, but does not determine
the phrase-structural destiny of the item. This is illustrated in (39) with the
form to-gā ‘plantation’, derived from the root to ‘plant’. The derived form has
an unpredictable semantic interpretation, which may be described roughly as:

‘location associated with the action denoted by the base form’. After entering
into a compound with niu ‘coconut’, the derived form togā is inserted into a
verbal syntactic function by combining it with the perfective particle ‘ua. This
syntactic categorization, as expected, yields a fully compositional semantic
interpretation, namely ‘was a (coconut) plantation’.
Samoan (Mosel 2004: 267):

(39) ‘Ua to-gā-niu ‘ātoa le mea maupu’epu’e
tam plant-nmlz-coconut whole art place hill
‘The whole hilly place was now a coconut plantation.’
A very similar process occurs in Tongan, which has the suffix -Canga that can
be applied to action-denoting, property-denoting, and object-denoting roots.
According to Broschart (1997: 146), this suffix, just like Samoan -ga, is not a
nominalizer in the phrase-structural sense, but rather a derivational marker
whose output forms are flexible and mean something like “domain where
something is or takes place”. However, the exact semantic interpretation of
Canga-derived forms is far from fully predictable, as the examples in (40a-c)
make clear:
Tongan (Broschart 1997: 146)

(40) a. pule ‘govern’ – pule-’anga ‘government’
b. motu’a ‘old’ – motu’a-’anga ‘reason for having aged’
c. api ‘home’ – api-tanga ‘homestead’
Despite their ‘nominal’ semantics, Tongan -Canga derivations can be used as

verbs by combining them with a TAM particle, as is illustrated in example (41):
Tongan (Broschart 1997: 145)

(41) ‘oku ‘ikai ke pule’anga
prs neg subj government
‘It does not belong to the government.’
(lit. ‘it is not that it government-s.’)
I argue that the semantic shifts in the Samoan lexemes listed in (37) above
are due to a process of derivation that must be analyzed as the zero-marked
counterpart of -ga derivation. The same analysis applies to Tongan -Canga
derivation, as compared with Tongan zero-derivation of the type illustrated

in (34b) above. Let us take the Samoan form gaoi in (37c) as a concrete
example. The proposal is that there are two homophonous but structurally
different forms gaoi: (i) an un-derived form with the meaning ‘steal’, and (ii)
a zero-derived form with the meaning ‘thief ’. Alternatively, the items in (37)
can be analyzed as polysemous, i.e. as involving two meanings belonging to
a single form. The crucial point, which is captured by both analyses, is that
the cases in (37) are not instances of flexibility. As such they stand in contrast
to the examples in (35) and (36), which involve syntactic rather than lexical
categorization, and accordingly behave semantically fully regularly.
Notably, it is predicted that both forms of the examples in (37), just
like -ga derivations and their base forms, are flexible, i.e. are open to usage
in verbal and nominal syntactic functions, and that these usages will yield
compositional semantic interpretations. Taking gaoi as an example again,
the un-derived form with the meaning ‘steal’ would be interpreted as ‘to
steal’ in verbal function, and as ‘(the act of ) stealing’ in nominal function.
The zero-derived form with the meaning ‘thief ’ would be interpreted as
‘to be a thief ’ in verbal function, and as ‘a / the thief ’ in nominal function.
Unfortunately, the available data are insufficient to confirm this prediction.
Interestingly, the Samoan -ga suffix can also be used as a syntactic
marker, rather than as a marker of lexical derivation. As would be expected,
in the former function its output receives fully compositional semantic
interpretations, namely ‘the act of X-ing’, where X is the action denoted
by the base. In addition, the two different levels at which -ga may operate
correlate with a phonological difference: Lexical derivation involves
vowel-lengthening in the base form, while syntactic derivation leaves
the phonological structure of the base unaffected. The phonological and
semantic differences between lexical and syntactic -ga formations are
illustrated in (42), using the same examples as in (38) above.

(42)
Base Lexical derivation Syntactic derivation
amo ‘carry’ āmo-ga ‘person(s) carrying loads’ amo-ga ‘carrying’
a’o ‘teach’ ā’o-ga ‘school’ a’o-ga ‘education’
tīpi ‘cut’ tīpi-ga ‘surgical operation’ tipi-ga ‘cutting’
pule ‘control’ pulē-ga ‘unit of church administration’ pule-ga ‘controlling’

A similar phenomenon is attested in Maori. This language also has a very
high degree of lexical flexibility, and it has a suffix -tanga, which can be used
as a marker of lexical and syntactic derivation (Bauer 1993: 254). Example
(43) nicely shows both types of derivations from the same base form tangi
(‘cry’) within a single clause. The lexical process yields the idiosyncratic
interpretation ‘funeral’, while the syntactic formation is compositionally
interpreted as ‘the crying (of the people)’.
Maori (Bauer 1993: 48)

(43) I rongo au i te [tangi-hanga hotuhotu-tanga
t / a hear 1sg dir.obj art cry-nmlz sob-nmlz
O ngaa taangata i te tangi-hanga ki a
poss art people at art funeral loc art
Maui Poomare]
Maui Poomare
‘I heard the people’s crying at Maui Poomare’s funeral.’
So far, we have seen that syntactic categorization of putatively flexible

lexemes in languages like Samoan and Tongan involves fully compositional
semantic interpretation. This is in accordance with Evans and Osada’s
compositionality criterion for ‘true’ flexibility. However, the data discussed
above also show that such flexibility does not exclude the presence of lexical
(re-)categorization processes. When such lexical processes are not overtly
marked, they may be confused with flexibility. However, on the basis of the
irregular semantic characteristics of these lexical processes, I have argued
that they do not involve flexibility, but should rather be analyzed as instances
of conversion.
We may now compare the Tongan and Samoan data discussed above with
similar patterns found in other supposedly flexible languages. Consider for
instance Tagalog, a language in which, according to Himmelmann (2005: 361),
“content words do not have to be sub-classified with regard to syntactic categories. They
all have the same syntactic distribution, i.e. they all may occur as predicates [and] as
(semantic) heads of noun phrases […]”. Himmelmann analyzes voice-marking in
Tagalog as a derivational process, which produces “actor- or undergoer-oriented
action expressions” that are still flexible in terms of their syntactic possibilities.
This process is characterized by “pervasive formal and semantic idiosyncrasies”
(Himmelmann 2007: 288). Some examples of idiosyncratic semantic
interpretations of voice-marked forms are provided in (44):

Tagalog (Foley 1998)
(44) Base Voice-marked form
a. abogado ‘lawyer’ – mag-abogado ‘become a lawyer’
b. bus ‘bus’ – mag-bus ‘ride a bus’
c. kamay ‘hand’ – mag-may-an ‘shake hands’
d. langgam ‘ant’ – langgam-in ‘be infested with ants’
e. lubid ‘rope’ – lubir-in ‘be made into rope’
Derivation with voice markers may also be irregular in terms of phonology:

It may involve unpredictable deletion of root vowels, as illustrated in (45a),
or sporadic insertion of / n /, as in (45b). There are also completely irregular
forms, such as in (45c):
Tagalog (Himmelmann3007: 288)

(45) Base Voice-marked form
a. kain – kan-in
‘consumption of food’ ‘to eat something’
b. tawa – tawa-n-an
‘laugh, laughter, laughing’ ‘to laugh at someone’
c. kuha – kun-in
‘getting, a helping’ ‘to get something’
As already mentioned, even though voice-marked forms denote actions, they

can still be used in nominal syntactic function without further measures.
As predicted, such syntactic categorization yields fully predictable semantic
interpretations: An actor-voice form in nominal function denotes ‘the agent
of the action’, a patient voice form in nominal function denotes ‘the patient
of the action’, and so on (Himmelmann 2007) 29. An example of a locative
voice form in nominal syntactic function is given in (46):
29
These Tagalog data show that a compositional interpretation of flexible lexemes is not
necessarily ontologically equal across languages: The interaction with the specific feature of
voice-marking in Tagalog accounts for the fact that in this language action-denoting forms
are interpreted as objects or individuals in referential function, while in Samoan they are
interpreted as denoting the action as a whole. See also the examples from Kharia further on.

Tagalog (Himmelmann 2007: 266)
(46) i-u-uwi’ niya ang
cv-rdp-returned_home 3.sg.poss spec
a-alaga’-an nya
rdp-cared_for-lv 3sg.poss
‘He would return the ones he was going to care for.’
Like the other flexible languages discussed above, Tagalog also seems to
display cases in which two or more unpredictably related meanings are
associated with a single form. According to Himmelmann, when action-
denoting roots are used in nominal function without voice-marking, they
may have one of the following kinds of meanings: (i) a state that results from
the performance of an action, (ii) the result of the performance of an action,
(iii) an object typically involved in the performance of an action, or (iv) the
name of an action. I argue that interpretation (iv) is the result of syntactic
categorization, while the meanings (ii) and (iii) are the result of lexical (re-)
categorization 30. Consider now example (47), in which the root lakad ‘walk’
has meaning type (ii):

(47) Iyón ay ma-haba’-ng lakad
dist pm stat-length-lk walk
‘That was / is a long walk
In parallel with the analysis of Samoan gaoi, I propose that there is a simple
root lakad1, meaning ‘walk’, and a homophonous, zero-derived form lakad2,
meaning ‘the result of walking’ 2 ‘a walk’. It may be objected that a zero-
derivation analysis is somewhat problematic, since, unlike in Samoan, there
seems to be no overt counterpart of this particular process. This is why above
I have described the phenomenon in terms of polysemy. The crucial point is
that the meaning shift is not due to a syntactic, but to a lexical process, and
as such does not involve flexibility.
It is predicted that both lakad1 and lakad2 are flexible, i.e. can be used
in verbal and nominal syntactic functions, with compositional semantic
interpretations. The base form lakad1 would mean ‘to walk’ when used as a
verb, and ‘(the act of ) walking’ when used as a noun. The latter interpretation
30
Meaning (i) is associated with attributive function, and I will not discuss it here.

is illustrated in example (48) below. In contrast, lakad2 would mean ‘a / the
walk’ when used as a noun, and ‘be a walk’ when used as a verb (as is in fact
the case in example (47) above).

(48) Ma-husay ang lakad ng mákiná
stat-orderliness spec walk gen machine
‘The walking of the machine is good.’
These data show that, like the other languages discussed in this section,
Tagalog displays ‘true flexibility’ in the sense that it freely uses simple
and derived lexical material in all syntactic functions with compositional
semantic interpretations. This flexibility is combined with either overtly or
zero-marked lexical processes, which yield semantically non-compositional
outcomes.
Finally, consider Kharia, a South Munda language that is unrelated to
the Malayo-Polynesian languages discussed so far. Peterson (2005, 2006,
forthcoming) claims that Kharia does not possess the lexical classes ‘noun’
and ‘verb’, but instead has two syntactically defined categories, which he calls
‘predicate’ and ‘complement’. According to Peterson (2006: 60), “lexemes in
Kharia do not appear to be either inherently predicates or their complement but can
generally appear in both functions, without any overt derivational morphology.”
The categories ‘predicate’ and ‘complement’ are expressed through the
combination of a so-called ‘content head’ with a functional head. Content
heads may consist of any single lexical root, but also of multiple roots or
complex phrases. A predicative functional head is expressed through enclitic
markers for tense / voice and person, while a complement functional head is
expressed through enclitic case marking (other than possessive / genitive).
The semantic interpretation of a content head in an actual utterance
requires a combination of the intrinsic meaning of the root(s) that it consists
of, the syntactic function in which it appears, and the morpho-syntactic
distinctions belonging to that function (such as tense / voice in predicative
function). According to Peterson (2006: 70), this combination yields “entirely
predictable” semantic outcomes. In particular, if an object- or individual-
denoting root is used as a predicate and marked for middle voice, it gets
the meaning ‘to become X’ (where X is the denotation of the root). When
the same type of lexeme in the same function is marked for active voice it
means ‘to turn something into X’. Action-denoting roots that are used in

referential function get the meaning ‘(the act of ) X-ing’ (where X is again
the denotation of the root). Examples are given in (49) and (50):
Kharia (Peterson 2006: 60, 68)

(49) a. lebu ɖel-ki b. bhagwan lebu=ki
man come=m.pst God man=m.pst
‘The / a man came.’ ‘God became a man.’ (= Jesus)
(50) a. u kayom onɖor=kon raʈa=yaɁ ayo=ɖom,

this talk hear=seq Rata=gen mother=3.poss
daɽhi=yaɁ saw-ɽay=ɖom gam=te:…
Darhi=gen spouse-woman=3.poss say=act.prs
‘Hearing this talking, Rata’s mother, Darhi’s wife says: …’
b. ni lebu=ki khoɽ i=ki=te kayom=ta=ki

story people=pl village.section=pl=obl talk=m.prs=pl
‘The people tell [this] story in the villages.’
The syntactic nature of Kharia categorization and the phrasal clitic-hood

of the functional heads becomes even more evident when considering an
example like (51), in which the complex content head bidesaɁ ebukiyaɁ rupraŋ
‘the appearance of people from abroad’, is used in the syntactic function of
predicate. This involves combining the content head with a functional head,
which is expressed by clitic forms for voice, tense and person, and attaches
to the phrase’s final element.
Kharia (Peterson 2006: 85)

(51) bharat=yaɁ lebu=ki [bides=aɁ lebu=ki=yaɁ
India=gen person=pl abroad=gen person=pl=gen
rupraŋ]=ki=may.
appearance=m.pst=3pl
‘The Indians took on the appearance of people from abroad.’
Just like the other flexible languages discussed above, Kharia also has
lexical processes that involve semantically and phonologically unpredictable
outcomes. Consider the derivational process marked by nasal-vowel-
infixation in Kharia. The output forms of this process may denote an object,
an instrument, or a location typically involved in the action denoted by the

base form, as well as a specific instance or result of that action. With respect
to phonology, the infixed vowel has the same quality as the vowel preceding
the nasal, but the infixed nasal is of indeterminate quality: it is usually
realized as /n /, but / m / is also occasionally found (Peterson 2006: 79). Some
examples of the process are given in (52):

(52) Base Lexical derivation
a. bel ‘spread out’ be-ne-l ‘bedding’
b. bui ‘keep, raise’ bu-nu-i‘ ‘pig’
c. joɁ ‘sweep’ jo-no-Ɂ ‘broom’
d. jin ‘touch’ ji-ni-b ‘touch’
e. rab ‘bury’ ra-na-b ‘burial ground’
f. kuj ‘dance’ ku-nu-j ‘dance’
g. muɁsiŋ ‘rise’ mu-nu-Ɂsiŋ ‘east’
According to Peterson, and in line with the lexical (zero-)derivational

processes discussed for other flexible languages, Kharia infixed forms can be
used freely in nominal and verbal syntactic function, despite their ‘nominal’
meanings. As is the case for any other type of content head, this yields
entirely compositional semantic interpretations. Thus, a derived form like
bunui ‘pig’ (see (52b) above) will get the meaning ‘to become a pig’, when
used as a predicate in middle voice, but ‘to turn something / / someone into a
pig’ when used as a predicate with active voice. In nominal syntactic function
it will mean ‘a / the pig’. Obviously, the predicative uses of a form like bunui
are pragmatically rather odd, and will therefore be seldom attested, if at
all. However, examples like those in (49b) above make clear that, given the
appropriate pragmatic context, the use of object- or individual-denoting
lexemes in verbal syntactic function is grammatically totally unproblematic.
To summarize, I argued in this section that the patterns of semantic
interpretation of lexical material in presumably flexible languages can be
understood when taking into account the mismatch between lexical and
syntactic categorization in these languages. In flexible languages, lexical
categorization is not relevant for phrase structure. This stands in contrast to
typical rigid languages, in which there is a one-to-one correlation between
lexical categorization and syntactic categorization. Possibly, these two
scenarios represent the extreme ends of a scale or cline. This would mean
that there are also languages in which syntactic categorization takes place

somewhere in between the level of simple lexical root and the level of phrase
structure. In Chapter 8 I will discuss this idea and its repercussions for the
present study in more detail.
2.5.3 The criterion of exhaustiveness: How flexible should a flexible

language be?
The second criterion proposed by Evans and Osada (2005) concerns the
amount or the proportion of flexible lexical items that is required in order
for a language to qualify as flexible. This criterion is described as follows:
“It is not sufficient to find a few choice examples which suggest word
class flexibility. Since word classes are partitionings of the entire lexicon,
equivalent statements need to hold for all relevant words in the lexicon
that are claimed to have the same class”. (Evans & Osada 2005: 378)
It seems, in fact, that there is considerable agreement between scholars on

this point. For instance, Hengeveld et al. (2004: 538) remark that “in order
for a language to qualify as flexible, it has to show systematic flexibility (…)
for an entire class of lexemes.” And Baker (2003: 117) makes a similar point
saying that “an important typological difference exists only if categorial ambiguity
extends to an entire open class of inflectionally similar words”.
However, Evans and Osada (2005: 378) rightly observe that “in practice
(…) it is difficult to check every one of tens and thousands of lexical items”. In spite
of this rather obvious practical restriction, it seems that in many reference
grammars and other descriptive sources quite unambiguous statements can
be found about the pervasiveness of multi-functionality of (certain groups
of ) lexemes, be it not exactly in the quantified fashion that Evans and Osada
aim at. This is clearly illustrated by the quotations in the previous subsection
about such languages as Samoan, Tagalog, and Kharia. In Chapter 5 I will
further discuss the criterion of exhaustiveness and its application to the
relevant languages of the sample.
2.5.4 The criterion of equivalent combinatorics: Bi-directional flexibility

The third and final criterion for flexibility proposed by Evans and Osada is
the criterion of equivalent combinatorics. It states that
“[M]embers of what are claimed to be merged classes [i.e. flexible classes,

EvL] should have identical distributions […]. All members of both

putative classes should be equally acceptable in both primary functions
as argument and predicate”. (Evans & Osada 2005: 366-367)
A corollary of this criterion is that flexibility should be bi-directional:
“To establish that there is just a single [i.e. flexible, EvL] word class, it is
not enough for Xs to be usable as Ys without modification: It must also
be the case that Ys are usable as Xs.” (Evans & Osada 2005: 375)
The part of this criterion that makes reference to the actual possibility to use a
flexible lexeme in more than one function seems quite unproblematic. Probably
anyone would agree that this is an essential part of any definition of flexibility.
Rather less straightforward is the part of the bi-directionality criterion that refers
to equal acceptability of every lexeme in all functions. In particular, the frequency
with which flexible lexemes appear in predicative or referential function is bound
to be influenced by semantic and pragmatic factors. Regarding this point, I
follow Hengeveld and Rijkhoff, who argue as follows:
“There is no reason to assume that all our concepts are equally

symmetrical with respect to predicating and referring functions in a
particular language. We may expect certain flexible lexemes to occur
more as predicates than as arguments, whereas other lexemes are used
more often as arguments than as predicates.”
(Hengeveld & Rijkhoff 2005: 412)
In other words, the semantic denotation of a root affects its relative

markedness with respect to a particular propositional function. This is in
fact convincingly shown by Croft’s prototype approach to PoS classes, as
discussed in section 2.4.2. Also, recall in this respect the examples from
Samoan, Tongan, and Kharia mentioned in the previous subsection, which
involve semantically ‘nominal’ but syntactically flexible (derived) lexemes.
Crucially, the fact that the use of a lexeme with a particular semantic meaning
in a particular function is pragmatically marked and therefore infrequently
attested does not mean that there is a syntactic restriction that excludes
such usage. In Chapter 5 I will further illustrate this point for the relevant
languages in the sample.
A final aspect of bi-directionality is specifically related to noun-adjective
flexibility. There are languages in which property-denoting lexemes

(i.e. semantic adjectives) can be the sole element in a referential phrase 31.
However, this does not necessarily warrant the conclusion that there
is no lexical distinction between adjectives and nouns. Often, such cases
of apparent noun-adjective flexibility in fact involve so-called absolutively
used adjectives, i.e. adjectives which function as the modifier of a head that is
understood from the discourse context and therefore remains unexpressed.
Moravcsik (2001) distinguishes two ways in which property-denoting
lexemes can function as the sole element in a referential phrase, namely
referentially dependent and referentially independent. In the former case, the
head noun must be available from the discourse or situational context. In the
latter case, the intended referent is invariant, such as ‘people’ in (53) below.
This possibility of referentially independent use of property-denoting words
seems to be a requirement for ‘true’ noun-adjective flexibility.
Hungarian (Moravscik 2001: 339)

(53) Sok gazdag van itt
many rich is here
‘There are many rich (people) here.’
However, it is often problematic to determine whether referentially

independent usage of property-denoting lexemes as the head of a referential
phrase is a real option in an individual language. This is especially so when
the analysis is based on isolated examples from written sources. Sometimes
however, evidence in favour of an independent usage analysis can be found
in the domain of agreement. In particular, the understood head may trigger
the expression of agreement morphology on the absolutively used adjective.
This happens for instance in Spanish, as shown in example (54), where
moderna takes a feminine singular agreement suffix because it modifies the
un-expressed but understood head la casa ‘the house’.
Spanish (Hengeveld 1992: 61-62)

(54) prefiero es-a modern-a
prefer:1s.prs that-fem.sg modern-fem.sg
‘I prefer that modern (one).’ [e.g. the house]
31
Schachter & Shopen (2007: 17) use the term ‘adjectival-noun languages’ for this type of
language.

Note furthermore that, if one takes seriously the criterion of bi-directionality,
this implies that, in order to establish ‘true’ noun-adjective flexibility, it
should be possible not only to use property-denoting lexemes as heads of
referential phrases, but also to use object- or individual-denoting lexemes
(i.e. semantic ‘nouns’) without further measures as modifiers in referential
phrases. Again, this is not always easily determinable, since the distinction
between nominal apposition or composition on the one hand, and attributive
modification on the other hand may be unclear. In Chapter 5, where the PoS
systems of the individual sample languages are categorized, I will discuss the
relevant cases in more detail.
2.5.5 Lexical flexibility: A summary

In this section, I discussed the three criteria for lexical flexibility proposed
by Evans and Osada (2005): Semantic compositionality, exhaustiveness,
and equivalent combinatorics. Especially the first criterion has given rise to
disagreement in the literature: While Evans and Osada contend that ‘true’
flexibility involves fully compositional semantic interpretation of lexemes in
particular functional environments, other scholars argue that the occurrence
of semantic idiosyncrasies is not an argument against flexibility. I have shown
that a number of languages, which figure prominently in discussions on lexical
flexibility, display both semantically compositional and non-compositional
(zero-marked) derivational processes. The difference between these two
types of processes lies in the fact that one pertains to syntactic, and the other
to lexical (re-)categorization. The special characteristic of flexible languages,
then, seems to be the lack of a one-to-one correspondence between these
two levels of categorization.
As regards the remaining two criteria, the one concerning exhaustiveness
seems theoretically unproblematic, but is hard to verify empirically. Finally,
with respect to the criterion of equivalent combinatorics, it was argued that
while bi-directionality is probably a central notion in flexibility, this does
not mean that flexible lexemes are expected to appear equally frequently in
all of their possible functions.
2.6 Summary
This chapter presented a number of functional-typological approaches to
lexical categorization. I started out with a brief outline of some early studies,
and the criteria that were used in them to define PoS classes. Subsequently,
I provided a detailed discussion of Hengeveld’s PoS theory. In its most

recent version, this theory proposes an implicational map consisting of
four propositional functions defined by two binary parameters: predication-
reference, and head-modifier. Hengeveld’s definitions of PoS classes are
based on the presence versus absence of structural coding accompanying
the use of lexemes in each of these four possible functions. The dominance
relations holding between the two parameters and between their respective
values predict a number of PoS systems that are expected to occur in actual
languages.
Hengeveld’s theory excludes two types of defining criteria for PoS,
namely morphological criteria (or behavioral potential) and semantic
criteria. The advantage of this approach is that it provides the possibility to
compare PoS classes and PoS systems across languages. However, exactly this
aspect of the Hengeveldian method has been criticized, most elaborately by
Croft (2000, 2001, 2005). The latter objects that Hengeveld’s theory, since
it ignores morphological and semantic data, amounts to methodological
opportunism and misses out on cross-linguistic universals to be found in
these two areas. On the other hand, the consequence of Croft’s method,
namely taking into account the full set of distributional data, is that lexical
categories, like any other type of language-specific formal category, can no
longer be compared across languages. Croft’s solution to this problem is
to define PoS as typological prototypes, i.e. as unmarked combinations of
semantic classes and pragmatic functions, which correlate cross-linguistically
with relative degrees of formal marking.
Another major point of discussion concerns the status of so-called
flexible languages, and especially the role of semantics in defining what
lexical flexibility really entails. Evans and Osada (2005, and cf. Croft
2005) propose that in truly flexible languages the interpretation of lexical
material in different functional environments should be fully compositional.
I discussed data from supposedly flexible languages, which show that
this criterion is indeed met. At the same time, however, such languages
display lexical processes of (zero-)derivation and polysemy, and these are
semantically (as well as phonologically) non-compositional. I argued that
the presence of both compositional and non-compositional semantic (re-)
categorization in flexible languages can be understood in terms of the fact
that these languages, unlike in the more familiar rigid languages, lack a one-
to-one correlation between lexical and syntactic categorization.
In the next chapter, I will use Hengeveld’s method for lexical classification
to develop a typology for a different set of constructions, namely dependent

clauses. Therefore, it seems appropriate at this point to emphasize that I fully
acknowledge the limited scope of the Hengeveldian approach in terms of the
type and amount of analytic primitives that it takes into account. However,
unlike Croft, I do not regard this as methodological opportunism. Depending
on the aim of one’s research, it seems perfectly defendable to focus on certain
linguistic facts while ignoring others, as long as this is made explicit. As already
mentioned, the lexical categories defined within Hengeveld’s framework are
claimed to be cross-linguistically comparable only as far as his very specific set
of criteria goes. And, as Cysouw (2007: 227) puts it: “Cross-linguistic identity
is always just a matter of granularity of analysis”.

Dependent Clauses
3.1 Introduction
In the previous chapter, parts of speech classes were defined as mappings
onto a space consisting of four propositional functions, defined in terms of
two hierarchically ordered implicational parameters: the predication-reference
parameter and the head-modifier parameter. The aim of the present chapter
is two-fold. First, it demonstrates that dependent clauses (henceforth DCs)
can be defined in the same way as parts of speech (PoS), namely as mappings
onto this functional space. Second, on the basis of many earlier functional-
typological studies, it is shown how DCs can be classified according to their
internal morpho-syntactic properties.
The central point of section 3.2 is that propositional functions constitute
the common ground between PoS classes and DC constructions. More
specifically, complement clauses have referential function, like lexical nouns;
relative clauses function as modifiers in referential phrases, like adjectives;
and adverbial clauses function as modifiers of predicates, predicate phrases
or clauses, in the same way as lexical adverbs 32. Based on this parallelism, I
show in section 3.2.1 that DCs can be defined, like PoS, in terms of the (set
of ) propositional function(s) that they express.
This implies that the notions of flexibility and rigidity, as introduced
for PoS in Chapter 2, can also be applied to DCs. Recall that, according to
32
Apart from these three traditional types of DCs, there are so-called predicate clauses which
function predicatively, like lexical verbs. I will come to these shortly.
65
the Hengeveldian approach, these notions depend crucially on the presence
versus absence of structural coding accompanying the use of a particular
construction in a particular propositional function. In order to arrive at
parallel definitions for PoS and DCs, a definition is required of structural
coding of DCs. This issue is addressed in section 3.2.2. Subsequently, in
section 3.2.3, I define a number of rigid and flexible DC types, in parallel
with the various types of PoS classes that figure in the PoS systems predicted
in Chapter 2.
In section 3.3 I turn to the internal morpho-syntactic properties of DCs,
i.e. their behavioural potential. In section 3.3.1 I explain that these properties
are, at least in part, motivated by the fact that DCs involve a marked
combination of semantic class and propositional function. In particular,
DCs denote states of affairs (henceforth SoAs), or propositional contents 33.
As such, they are marked as expressions of reference and modification, since
these propositional functions are prototypically expressed by object-denoting
and property-denoting (lexical) constructions, such as nouns, adjectives, and
manner adverbs (see Chapter 2, section 2.4.2). The marked combination
of semantic class and propositional function displayed by DCs is reflected
in their formal encoding, which in many cases shows a certain mixture of
morpho-syntactic characteristics of independent clauses on the one hand,
and features of lexical or phrasal constructions on the other hand.
There are cross-linguistic differences as regards the set of possible
features that may be involved in the behavioural potential of DCs in the
first place; not all languages display the same morpho-syntactic distinctions.
In addition, language-specific DC constructions may differ in the amount
and type of features that is selected from the available set. However, the
variation is not random, and functional-typological research has come a
long way in bringing to light universals in these areas. Among other factors,
the possible combinations of morpho-syntactic features in the DCs of the
world’s languages are constrained in terms of universal hierarchical ordering
patterns displayed by certain feature sets. In section 3.3.2 I give a general
overview of the research on these patterns, which are usually referred to as
verbal and nominal feature hierarchies (Bybee 1985; Dik 1997; Lehmann
& Moravcsik 2000; Rijkhoff 2002). Subsequently, in section 3.3.2, I will
discuss studies that focus particularly on the impact of these hierarchies on
33
There are also DCs that do denote objects or individuals (i.e. first order entities), such as
headless relative clauses and actor / object nominalizations. However, the present study is not
concerned with these.

mixing patterns of verbal and nominal features in DCs, and the functional
motivations underlying these effects (Cristofaro 2003; Malchukov 2004).
At the end of this chapter, in section 3.4, I present a typological
framework for DCs that combines the parameters of internal morpho-
syntactic properties (as discussed in section 3.3) with the classification system
of rigid and flexible DCs (as developed in section 3.2). This integrated
typology will be applied in Chapter 6 to the DCs of the sample languages.
Finally, section 3.5 is a brief summary of the present chapter.
3.2 Dependent clauses and propositional functions
3.2.1 Slots and fillers

Traditionally, as mentioned in the introduction to this chapter, three types of
DCs are distinguished: complement clauses, relative clauses, and adverbial
clauses (Thompson et al. 2007: 238). They can be defined in terms of their
respective pragmatic-syntactic functions: A complement clause functions as
a referential unit, expressing an argument of the matrix clause predicate; a
relative clause functions as a modifier of a matrix clause argument; and an
adverbial clause functions as a modifier of the main predicate of the matrix
clause. As such, the functions of these DC types correspond to the defining
propositional functions of lexical nouns, adjectives, and manner adverbs, as
presented in Chapter 2, section 2.2 34.
In addition to these three DC types, there are also clausal constructions
that function predicatively. Such constructions are usually called predicate
clauses. Recall from Chapter 2 that the use of non-verbal lexemes in predicative
function may trigger one of three expression strategies: (i) a copula, (ii) a
zero-2 strategy (i.e. no copula and no verbal behavioural potential), or (iii)
a zero-1 strategy (i.e. no copula but full expression of verbal behavioural
potential). Only when a particular PoS class triggers the third, zero-1 strategy,
it is regarded as having the possibility to function predicatively, in addition
to its defining, non-predicative use(s) (see also Chapter 5, section 5.5).
34
Note that, as far as adverbial clauses are concerned, this study is restricted to adverbial
manner clauses (cf. lexical manner adverbs in Chapter 2, section 2.3.1), excluding other types
of adverbial clauses, such as purposive, temporal, and conditional clauses. This means that,
in terms of DCs, I will take into account only adverbial clauses that specify how the event
expressed by the predicate of the matrix clause is performed (Kortmann 1998: 467). However,
the borderline between adverbial manner clauses and other, semantically related types of
adverbial clauses is not always easily drawn. This issue is taken up in Chapter 6, where the
actual DCs of individual languages are discussed.
Chapter 3 – Dependent Clauses | 67

In the present chapter (and in Chapter 6) the same procedure will be
followed for DCs 35.
In Chapter 2 an implicational map model was developed to formulate
restrictions on the typology of PoS classes and PoS systems. The functional
space pertaining to this model is re-presented in Figure 3.1a. In Figure
3.1b, the four functional slots of the space are filled with the four DC types
mentioned above, instead of with lexeme classes, as was the case in Chapter
2. In other words: lexical units are substituted by configurational units.
The term ‘configurational’ refers to a complex linguistic construction, in
which the elements stand in a nucleus-dependent (or predicate-argument)
relationship (cf. Hengeveld & Van Lier 2008: 769-770).
Head Modifier
Predicate phrase
Referential phrase
Figure 3.1a: The space of propositional functions
Head Modifier
Predicate phrase Predicate clause Adverbial manner clause
Referential phrase Complement clause Relative clause
Figure 3.1b: Propositional functions and dependent clause types
The definitions of the four types of dependent clauses in Figure 3.1b can
now be formulated as in (1):
(1)
(i) A predicate clause is a configurational unit that can be used as the
head of a predicate phrase;
(ii) A complement clause is a configurational unit that is used as the
head of a referential phrase;
(iii) A relative clause is a configurational unit that is used as a modifier in
a referential phrase;
(iv) An adverbial manner clause is a configurational unit that is used as a
modifier in a predicate phrase.
35
In Chapter 6 it will become clear that zero-1 strategies are hardly if ever attested in
combination with clausal constructions. Nevertheless, in the present discussion the possibility
of a predicate clause triggering a zero-1 strategy is included for reasons of theoretical
consistency.

The parallel between these definitions and the definitions of nouns,
adjectives, and manner adverbs presented in Chapter 2 will be obvious:
DCs are regarded as the clausal equivalents of PoS classes, as far as their
respective functions in the matrix clause are concerned.
Recall, however, that in the definition of the PoS classes in Chapter 2
the phrase ‘without further morpho-syntactic measures’ was included (see
section 2.3.1). As was explained, these further measures make reference to
structural coding rather than to behavioural potential. In order to complete
the parallelism between PoS and DCs, the notion of structural coding needs
to be defined with reference to the latter construction type. This is the topic
of the next subsection.
3.2.2 Structural coding in dependent clauses:

Subordinating conjunctions and dependent verb forms
Two general types of structural coding of DCs can be distinguished:
(i) Subordinating conjunctions, further sub-divided into

complementizers, relativizers, and adverbial conjunctions;
(ii) Special markers on the dependent predicate, further sub-divided
into nominalizers and infinitive markers, participles, and converbs 36.
These two groups of markers are regarded as structural coding because

they function primarily as indicators of subordination, and, when they are
specialized markers, of the function in which the DC appears. In the case of
subordinating conjunctions, the structural coding occurs at the level of the
DC construction as a whole, and the internal morpho-syntactic structure
of the DC remains the same as in independent clauses (in the particular
language under analysis). In contrast, special morphemes on the dependent
predicate mark the construction at the level of its nucleus. This affects the
categorial status of the predicate, which is reflected in a certain degree of
deviation of the internal morpho-syntax of the DC from the structure
of an independent clause. Structural coding by means of a subordinating
conjunction is illustrated in (2a) below with the English that-construction.
Example (2b) shows structural coding with a dependent predicate marker,
namely the English -ing form. Note that in the DC in (2b) there is no tense
expression, and that the subordinate subject is encoded as a possessor.
36
The term converb refers to “a non-finite verb form whose main function is to mark adverbial
subordination” (Haspelmath 1995: 3).

A number of issues concerning the two types of structural coding of DCs

require some discussion. First, examples (2a-b) show that subordinating
conjunctions are typically freestanding forms, which appear at the periphery
(i.e. either the beginning or the end) of a DC, whereas dependent predicate
markers are (obviously) bound forms that attach to the predicate. However,
as Cristofaro (2003) explains, in some languages this distinction between
the two types of structural coding is somewhat obscured, namely when
subordinating conjunctions are clitics. These clitics may appear attached to
the dependent predicate for one of the following independent reasons:
(i) They can be attached to any constituent in the clause and so may
happen to be attached to verbs;
(ii) They are usually placed close to verbs because of word order rules;
(iii) There is a rule stating that they have to be attached to verbs.
(Cristofaro 2003: 58)
In such cases one might be tempted to regard the complex [dependent

predicate + cliticized conjunction] as an instance of structural coding by
means of a special marker on the dependent predicate. However, special
markers on the dependent predicate, as mentioned above, affect the internal
morpho-syntactic structure of the DC, while subordinating conjunctions
do not. Therefore, Cristofaro argues, a cliticized subordinating conjunction
can be recognized as such when the structure of the DC remains unaltered.
Consider example (3), from Cofán. The relativizer =’cho cliticizes to the
dependent predicate because (i) it must attach to the final element of the
unit on which it operates, and (ii) DCs in Cofán are obligatorily predicate-
final. Despite its position, =’cho can be identified as a conjunction because
the internal structure of the DC is the same as in an independent clause 37.
Note that the DC in (3) is marked off by square brackets. This will be done
in examples of DC constructions throughout the rest of the book
Cofán (Fischer & Van Lier, forthcoming)

(3) Ingi=ta [avuja-tsh=e canse='cho] a'i-’fa=gi.
we=indef happy-stat=advr live=rel person-pl=foc.1
‘As for us, we were people that lived happily.’

Second, it should be mentioned that not all DCs are overtly structurally
coded by means of a subordinating conjunction or a special marker on the
dependent predicate. Rather, DCs can also be zero-marked 38. Consider for
instance example (4) from Bukiyip:
Bukiyip (Conrad & Wogiga 1991: 179)

(4) Énan n-a-kli [yek i-wich umu énaniny moul]
he he-real-say I I-irr-enter ben his work
‘He said that I would have his job.’
Third, structural coding may be fused with the expression of a category that
pertains to behavioural potential. For instance, determiners may function as
complementizers. This occurs in Dhaasanac, as illustrated in (5):
Dhaasanac (Tosco 2001: 288)

(5) yú [kúun hí fuɗɗu=a] feɗe
I you 3(pron ) open.dep=det want.ipfv
‘I want you to open it.’
Another common case of accumulation of structural coding and behavioural

potential is the fusion of a TAM marker with a special dependent predicate
marker. This is attested in Imbabura Quechua, where different forms of the
nominalization suffix express different relative tenses, as shown in (6):
37
Clauses marked with =’cho can also be used as object complements. In that function they are
marked for accusative case, as illustrated in (i):
(i) Atesu-mbi ke ña=nga tevaen=’cho=ma

know-neg you I=dat write=comp=acc
‘I didn’t know that you had written to me.’
According to Cristofaro, the expression of case would be evidence of categorial change and
would as such be a reason to analyze =’cho as a special dependent predicate marker. However,
following Malchukov (2004; see section 3.3.3.3 of the present chapter and Fischer & Van Lier
(forthcoming)), I regard case as an external category, which is in principle independent of the
categorial status of the construction on which it operates.
38
DCs without morpho-syntactic marking may be marked through prosody. For practical
reasons however, the latter coding strategy is not taken into account in the present study.
Another possibility is of course structural coding by means of fixed constituent order.

Imbabura Quechua (Cole 1982: 33)
(6) ñuka-ka [Juan kay-pi ka-j / -shka / -na]-ta ya-ni
I-top Juan this-in be-nmzl.prs / pst / fut-acc think-I
‘I think that Juan is / was / will be here.’
Finally, as mentioned above, DCs marked by subordinating conjunctions

typically have an internal structure that resembles the structure of
independent clauses, while the internal structure of DCs marked by
means of a special dependent predicate form deviates from the structure of
independent clauses. However, this correlation between structural coding
type and internal morpho-syntax is not absolute. In particular, the structure
of DCs marked with subordinating conjunctions may differ in certain
respects from that of independent clauses. Typically, this involves the use
of dependent or subjunctive forms for the expression of verbal categories.
Consider for instance example (7), which shows a Georgian complement
clause marked with the complementizer rom and with a predicate that bears
a subjunctive aorist suffix.
Georgian (Hewitt 1995: 628)

(7) Masp’injel-s u-nd-a, [rom male c’a-vid-e]
host-dat ov-want-it comp soon prev-go-aor.subj
‘The host wants me to go soon.’
Subjunctive constructions tend to have fewer inflectional possibilities than

their indicative counterparts, but this is not necessarily the case. As will become
clear in section 3.4, I will regard subjunctive constructions as sentence-like
DCs, i.e. as DCs with roughly the same syntactic form as independent clauses
(cf. Hengeveld 1998; Malchukov 2006: 974; Noonan 2007: 59-63).
Taking these issues into account, I round off the discussion of structural
coding of DCs and proceed to the next step: applying the notions of
flexibility and rigidity to DC constructions.
3.2.3 Rigid and flexible dependent clauses

In Chapter 2, it was shown that PoS can either be rigid, i.e. specialized for
the expression of a single propositional function, or flexible, i.e. usable in two
or more functions. If DCs are assumed to be the configurational equivalents
of PoS, then it follows that there are also rigid and flexible DCs. Like PoS,
DCs are regarded as rigid when they are used in only one propositional

function, and as flexible when they can fulfil two or more functions without
further measures, i.e. without any change in their structural coding. In this
section I will define a number of possible flexible and rigid DC types. These
DCs are the functional counterparts of the different types of PoS classes
that figure in the systems predicted in Chapter 2, section 2.3.2.2 39.
The four predicted rigid DC types were in fact already presented
in Figure 3.1b above, since they correspond to the three traditional DC
types, plus predicate clauses. For the sake of completeness, these DCs are
represented separately in (8)-(11) below. Notice that, as in Chapter 2, rigid
constructions are presented with light shading and flexible ones (as in shown
in (12) and further) with darker shading.
(8)
Head Modifier
Predication Predicate clause
Reference
(9)
Head Modifier
Predication
Reference Complement clause
(10)
Head Modifier
Predication
Reference Relative clause
39
As shown in Chapter 2, PoS classes enter into larger PoS systems. If DCs indeed mirror
the functional possibilities of PoS classes, then it may be expected that the former are subject
to the same restrictions as the latter in terms of the types of systems into which they may
combine. However, DC systems are as such not the focus of this study. Rather, I will be more
generally concerned with the question whether flexibility / rigidity in the realm of lexical
classification is reflected in flexibility / rigidity in the realm of DC constructions. In addition, I
will investigate whether particular PoS classes have DC counterparts with the same functional
possibilities. Neither of these two research questions requires the definition of specific systems
of DC constructions. Moreover, as I will argue shortly, languages often display a range of
different DCs that can be used in (partly) the same propositional function(s).

(11)
Head Modifier
Predication Manner adverbial clause
Reference
Turning to flexible DCs, the predicted types are represented in (12)-(20).

The first flexible DC-type can be used in all four functions of the space.
This type will be called a contentive clause, in accordance with the term
contentive used for maximally flexible lexemes. The functional potential of
contentive clauses is represented in (12):
(12)
Head Modifier
Predication Contentive clause
Reference
The second flexible DC type can be used in all functions except the head
of a predicate phrase. Thus, it has the same functional potential as lexical
non-verbs. However, the term ‘non-verbal clause’ seems counter-intuitive,
since dependent clauses typically do contain a verb. Therefore, I will use
multifunctional clause instead. This type is represented in (13):
(13)
Head Modifier
Predication
Reference Multi-functional clause
Third, a flexible DC type is predicted that can be used to express both

modifier functions, in the same way as lexical modifiers. This type will
therefore be termed a modifier clause. It is represented in (14):

(14)
Head Modifier
Predication Modifier clause
Reference
Fourth, a DC type is predicted that can be used as the head and modifier
in a referential phrase, in parallel with lexical nominals. This type, termed
nominal clause, is represented in (15):
(15)
Head Modifier
Predication
Reference Nominal clause
The DC type in (16) represents the clausal counterpart of the PoS class
of predicatives. This DC type, which will be called predicative clause, can
express the functions of head and modifier in a predicate phrase:
(16)
Head Modifier
Predication Predicative clause
Reference
The DC type represented in (17) parallels the distribution of lexical flexible

heads. This head clause can function as the head of both predicative and
referential phrases.
(17)
Head Modifier
Predication
Reference Head clause

Finally, in (18)-(21) I represent four types of flexible clauses with the same
distribution as four types of PoS classes predicted in Chapter 2 and indicated
there as Flex A, B, C, and D. As the representations show, I have simply
named the corresponding DC types Flex clause A, B, C, and D, respectively.
(18)
Head Modifier
Predication –
Reference Flex clause A
(19)
Head Modifier
Predication
Reference Flex clause B –
(20)
Head Modifier
Predication Flex clause C
Reference Flex clause C
(21)
Head Modifier
Predication Flex clause D
Reference Flex clause D
Note that there is one logically possible type of flexible DCs that is expected
not to occur in actual languages. It is represented in (22)* below (the asterix
indicates that that this type of DC is a ‘forbidden’ type). This system is the
functional counterpart of a PoS class that figures only in systems that violate
one or more of the constraints formulated in Chapter 2 (cf. Appendix ii:
(2)* / (3)*).

(22)*
Head Modifier
Predication Flex clause
Reference
Having defined these DC types in terms of propositional functions, in

parallel with the PoS classes in Chapter 2, a difference between DCs and
PoS should be noted. In particular, subordination relations are semantically
complex and this is reflected by structural variation in the DCs that express
them. A consequence of this is that in many languages there are multiple DC
constructions that can be used in the same or partly the same propositional
function(s) (cf. note 39). For instance, Turkish has two types of nominal
clause constructions: One in which the predicate is marked with -AcAK, and
another that takes the suffix -DIK. Examples (23) and (24) show that both
constructions can be used as complement clauses and as (object) relative
clauses. The choice between the two forms is semantically motivated and
concerns tense: -DIK indicates past and -(y)AcAK future tense.
Turkish (Göksel & Kerslake 2005: 423, 442)

(23) a. [Orhan -ɪn bir şey yap-ma-yacağ-i]
Orhan-gen anything do-neg-nmlz-3sg.poss
belliydi.
it.was.obvious
‘It was obvious that Orhan wouldn’t do / wasn’t going to do
anything.’
b. [Fatma-‘nın yarın gör-eceğ-i] film

Fatma-gen tomorrow see-ptc-3sg.poss film
‘the film that Fatma is going to / will be seeing tomorrow’

(24) a. (ben) [Ahmed-in öl-düg-ün]-ü duy-du-m
I Ahmed-gen die-nmlz-3sg-acc hear-pst-1sg
‘I heard that Ahmed died.’

b. [adam-ın git-tiğ-i] okul
man-gen go-ptc-3sg school
‘the school that the man goes / went to’
It is also possible that two DC constructions in the same language show

partial overlap in terms of the propositional functions that they can
express. For instance, Turkish has, apart from relative clauses marked by
-DIK / -AcAK, another participial construction with the suffix -An. Unlike
the -DIK / -AcAK construction, however, -An clauses cannot function as
complement clauses. Moreover, -An clauses express subject relative clauses,
whereas -DIK / -AcAK clauses express object relative clauses. An example of
an -An clause is given in (25):

(25) [ögretmen ol-an] haydar
teacher be-ptc Haydar
‘Haydar, who is a teacher’
These examples show that, when there are multiple DC constructions that
can express (partly) the same propositional function(s), the choice between
these constructions depends on the specific semantic and / or syntactic
characteristics of the subordination relation that holds between the DC and
the matrix clause.
3.2.4 Summary
In this section I have established the first part of a typology of DC
constructions, namely the part that makes reference to the functional
possibilities of DCs. These functional possibilities have been defined in
parallel with the PoS classes in Chapter 2, i.e. in terms of the number and
type of propositional function(s) that a particular DC can express without any
difference in structural coding. In the next section I turn to the second part
of the typology, namely the part that is concerned with the internal morpho-
syntactic properties or the behavioural potential of DC constructions.

3.3 Behavioural potential in dependent clauses:
De-categorization and re-categorization
3.3.1 Introduction
The internal structure of DC constructions has been investigated in a fair
number of functional-typological studies. Many of these studies focus on the
‘mixed’ patterns of behavioural potential attested in DCs across languages,
i.e. on the various combinations of morpho-syntactic properties typical of
independent clauses on the one hand, and of lexical or phrasal constructions
on the other hand.
These mixed patterns of morpho-syntactic properties can be understood
as the outcome of two functionally motivated processes: de-categorization
and re-categorization. The former process involves the non-expression of
behavioural potential associated with the primary propositional function
of a construction (Hopper & Thompson 1984). For a DC, i.e. a clausal
construction denoting a SoA or a proposition, this primary function would
be a predication. The categories that belong to this primary function, and
which may be lost as a result of de-categorization, are often called verbal
categories. The obvious reason for this is that the nucleus of a clause or
predication – the predicate – is prototypically a verb, which expresses the
relevant categories in the form of inflectional markers. Such verbal categories
include tense-aspect-mood (TAM) distinctions and / or person marking
(including agreement and cross-reference) 40.
The process of re-categorization, on the other hand, is reflected by
the expression of behavioural potential associated with the secondary
propositional function in which a construction appears (Bhat 1994). In the
case of a DC, this secondary function is reference or modification. As we
have seen, reference and modification are prototypically expressed by object-
denoting and property-denoting constructions, i.e. nouns, adjectives, and
manner adverbs (or phrases headed by these PoS). Cross-linguistically, the
range of categories associated with nouns and NPs is more extensive than the
range of adjectival or adverbial categories. Accordingly, the features that are
expressed in a DC as a result of re-categorization are often termed nominal
categories. They include gender, number, case, and definiteness. There are
also some typical adjectival features, such as degree of comparison and head-
modifier agreement. However, the former category is semantically marked
40
The process of de-categorization in DCs has alternatively been termed desententialization
(Lehmann 1988: 193) and deverbalization (Croft 1991: 79).

with respect to event-denoting DCs, while the latter is cross-linguistically
not very common (Haspelmath 2001: 16541). For manner adverbs there
seem to be no prototypical features at all. This asymmetrical distribution of
features over lexeme classes is also reflected in the literature on the internal
structure of DCs, which is often concerned with clausal nominalizations.
Both verbal and nominal features have been shown to display universal
patterns as regards the order in which they are expressed relative to their
base unit, a verbal or nominal head. In what follows I first discuss functional-
typological studies that provide evidence for these hierarchical ordering
patterns (section 3.3.2). Subsequently, I turn to studies that investigate how
these hierarchies motivate the particular mixtures of verbal and nominal
features attested in DCs across languages (section 3.3.3).
3.3.2 Verbal and nominal feature hierarchies
3.3.2.1 Introduction
Different studies have taken into account somewhat varied sets of verbal and
nominal features, depending on their theoretical orientation and on whether
or not they distinguish between morphological and lexical expressions of
categories. In this section I adopt the versions of the feature hierarchies
proposed by Malchukov (2004), since they explicitly integrate various
strands of functional-typological work 41.
3.3.2.2 The verbal feature hierarchy

Starting with (Malchukov’s version of ) the verbal feature hierarchy, its
most important ingredients are the work of Bybee (1985), combined with
insights from research carried out in the framework of Functional Grammar
(FG). Bybee’s well-known study shows that bound morphemes expressing
verbal categories exhibit a universally preferred order in relation to their
stem. This order is iconically motivated: it reflects the degree of relevance of
each category for the interpretation of the verbal stem. In particular, those
categories that occur more closely to the stem – the internal categories –
affect the verb’s semantics, whereas those that are expressed further away
from the stem – the external categories, in particular agreement phenomena
– do not affect the meaning of the stem but are rather relevant to its syntactic
41
I will adopt the visual representations of the hierarchies that are used by Nikitina (2007) in
her review of Malchukov’s work.

and / or pragmatic function 42. The verbal feature hierarchy as proposed by
Bybee is represented in (26):
(26)
STEM Valency Voice Aspect Tense Mood Agreement
Considering the hierarchy in some more detail, the most external category,
agreement, is purely inflectional to the extent that its sole raison d’être is
the existence of a syntactic relation between the verb and its argument(s)
(Haspelmath 2002: 72). Agreement does not have any impact on the
semantics of the verb. In contrast, the categories of tense, aspect and mood
are not directly triggered by syntactic relations. These categories can be
ordered with respect to each other in terms of their degree of semantic
relevance to the verbal stem. Mood and tense have relatively little impact
on the meaning of the verb, because the nature of the expressed action is
in principle independent of the speaker’s evaluation of that action (mood),
and of its temporal relation to the speech event (tense). Nevertheless, the
categories of mood and tense do convey a certain amount of independent
semantic content; a property that agreement typically does not have. The
category of aspect is again of a different type: since aspectual distinctions
modify the internal temporal constituency of an event, they directly affect
the semantics of the verb. Finally, the two most internal categories, valency
and voice, have an even higher amount of semantic impact on the verb, since
they relate to the number and orientation of the participants involved in the
event expressed by the verb.
Clearly, the functional difference between internal and external categories
relates to the traditional distinction between derivation and inflection.
However, this distinction is notoriously hard to make, since many of the
criteria proposed to distinguish between the two types of morphological
processes do not involve absolute properties. This is reflected in Bybee’s
treatment of verbal categories as points on a scale ranging from typically
inflectional (external) to increasingly derivational (internal).
The hierarchical structure of verbal categories also forms part of the
theory of Functional Grammar (FG; Dik 1997), and its successor: Functional
42
In fact, the way in which verbal categories are expressed is also iconically motivated: More
internal features tend to reveal more fusion with the stem. I will return to the functional
principle of Iconicity in section 3.3.3.2.

Discourse Grammar (FDG; Hengeveld & Mackenzie 2008). According to
F(D)G, utterances consist of several representational layers. The innermost
layer hosts the predicate with its arguments. On each higher layer a new set
of semantic distinctions is added, pertaining respectively to the domains of
aspect, tense, modality, and illocutionary force 43.
As Malchukov (2004: 16) remarks, the verbal feature hierarchy used
in F(D)G’s layered model of the clause is quite similar to the hierarchy
proposed by Bybee (1985), but there are also some differences. First, the
F(D)G approach takes into account both morpho-syntactic and lexical
expressions of TAM categories, whereas Bybee considers bound morphemes
only. In F(D)G, morpho-syntactic expressions of categories are termed
operators, whereas lexical expressions are called satellites. The inclusion of
the latter implies that adverbs of manner, time, etcetera, count as expressions
of verbal TAM categories.
A second difference between Bybee’s hierarchy and the framework of
F(D)G is that agreement is included as a verbal category in the former but
not in the latter. F(D)G represents agreement at a different grammatical
level than TAM categories, because, as mentioned above, agreement reflects
the syntactic relation between the predicate and its argument(s). As such, it
has no semantic representation and belongs to the level of morpho-syntactic
encoding. In contrast, TAM distinctions concern the semantics of their base
and therefore belong to the representational level of grammar. As we will
presently see, Malchukov includes agreement as a category in his version of
the verbal feature hierarchy, and he applies the F(D)G distinction between
operators and satellites to it. In particular, Malchukov regards an agreement
marker as an operator expressing argument realization, and the actual
encoding of the corresponding argument as the satellite-type expression of
the same category.
A complicating factor in this respect is that the category of agreement
can be decomposed into subject agreement and object agreement. According
to Malchukov (2004: 19), these two sub-categories occupy different locations
on the verbal feature hierarchy. Object agreement pertains to verbal valency,
i.e. to the innermost level of the feature hierarchy. Subject agreement, on
43
The FDG model rather works top-down and outside-in. Thus, the structural building of
an utterance starts from the speaker’s communicative intention, and works down to its actual
morpho-syntactic and phonological encoding. In addition, the underlying representation of
an utterance is build up starting with the most external categories, while FG representations
start with the innermost categories followed by the more external ones.

the other hand, is more relevant to perspectivization, i.e. to the pragmatics
of the base unit, and as such qualifies as an external category. Moreover, as
just mentioned, the morphological expression of agreement is linked to the
lexical expression of the corresponding argument(s). This means that lexical
subject and object expression occupy the same respective locations on the
verbal feature hierarchy as the concomitant agreement morphology.
A final, less important difference between Bybee’s feature hierarchy and
FG’s layered clause structure involves illocutionary force. The fact that FG
includes this category is a consequence of the fact that this framework takes
the clause as its basic unit of analysis, whereas Bybee looks more narrowly at
the verb phrase 44.
Integrating the points discussed above, Malchukov (2004: 20) proposes
the verbal hierarchy represented in (27). This is the version of the hierarchy
that will be adopted in the remainder of this chapter.
(27) VERB stem

voice / valency, direct object, object agreement

aspectual operators, adverbial satellites with aspectual value
(manner adverbs)

tense and mood operators and corresponding satellites
(temporal / modal adverbs)

subject agreement, clausal subject

illocutionary force markers
3.3.2.3 The nominal feature hierarchy

As regards the nominal feature hierarchy, Malchukov (2004) again integrates
several earlier proposals. One of them is the hierarchy of Lehmann and
Moravcsik (2000: 753), which is represented in (28):
44
In FDG, the basic unit of analysis is a communicative Move, which may consist of several
discourse Acts. These discourse acts do not necessarily correspond to clauses at the morpho-
syntactic level. For a full discussion, see Hengeveld & Mackenzie 2008.

(28)
STEM gender / noun class number possessive determiners case
As in the case of Bybee’s verbal feature hierarchy, the ordering of the nominal
features is presumably motivated in terms of iconicity: it reflects the relative
degree of relevance of the morphological categories for the interpretation
of the stem.
This becomes clearer when considering a second, more detailed
proposal, namely Rijkhoff ’s (2002, 2008a, b) hierarchical model of the
noun phrase. This model consists of five concentric layers at which nominal
features may operate. Working in the F(D)G framework, Rijkhoff includes
both morpho-syntactic and lexical expressions of nominal categories, i.e.
both operators and satellites. Furthermore, he proposes a basic distinction
between descriptive modification of the nominal base, which involves the four
most internal levels of nominal features, and one external level of discourse-
referential modification. In other words: the features belonging to the four
internal levels influence the semantic meaning of the base, while the most
external features are relevant to its pragmatics, i.e. “the status of the thing
(or event) as a discourse entity” (Rijkhoff 2008a: 67). First, consider in more
detail the four layers of descriptive modification:
(i) The kind layer, which has scope over the lexical head noun and
accommodates classifying modifiers that further specify what kind
of entity is being referred to by the speaker. In other words, they
specify a particular subclass of the entity denoted by the base. They
can be satellites, as in ‘a corporate lawyer’ or ‘an electric toothbrush’.
Classifying operators are so-called ‘nominal aspect markers’, such as
singulative / collective markers.
(ii) The quality layer, which has scope over the kind layer and
accommodates lexical modifiers (satellites) that specify properties
concerning size, color, weight, value, age, etcetera 45.
(iii) The quantity layer, which has scope over the quality layer and
accommodates grammatical and lexical expressions of number.
(iv) The location layer, which has scope over the quantity layer and
accommodates modifiers that specify properties concerning the
45
According to Rijkhoff (2008a: 794), qualifying operators, i.e. grammatical expressions of
quality do not exist.

location of the referent. These can be satellites and take the form of
case-marked or adpositional phrases, including possessive phrases 46.
Operators that belong to the locational level are demonstratives.
The highest or outermost layer distinguished in Rijkoff ’s model is the layer

of discourse-referential modification. Operators at this layer are markers of
definiteness, i.e. articles or determiners. Examples of discourse-referential
satellites in English would be same or other, i.e. lexical expressions that
“provide the addressee with information about the referent as a discourse entity”
(Rijkhoff 2008a: 789). The five layers of nominal features are represented
in (29):
(29) [N-STEM]
Kind: Nominal aspect, quantifying adjectives]

Descriptive Quality: Qualifying adjectives]]
Modification Quantity: Lexical / grammatical Number]]]
Location: Demonstratives,
Locative / possessive phrases]]]]
Discourse-Referential
Modification: Determiners, lexical discourse-related
modifiers]]]]]
Note that Rijkhoff, unlike Lehmann and Moravcsik (2000), does not take
into account case marking as a nominal feature. This is because (core) case
marking reflects the syntactic function of the unit it operates on. Therefore,
like verbal agreement, it is represented at the level of morpho-syntactic
encoding in F(D)G.
The empirical justification for the iconic ordering of nominal features
is somewhat fragmented. Early evidence for the hierarchy comes from
Greenberg in the form of his Universals 20 and 39, which refer to the
relative orderings of case markers, demonstratives, numerals and adjectives
(Greenberg 1966: 87, 95). These findings are further supported by Hawkins’s
research (1983) based on a sample of more than 300 languages, as well as by
46
Rijkhoff (2008a: 797) notes that (adnominal, restrictive) relative clauses can also be used for
modification at the the locational level, as in the book, that’s lying on the table. For the purpose
of the present study, however, it is assumed that relative clauses are equivalent to lexical
adjectives in terms of their propositional function.

Rijkhoff ’s (2002) study of a balanced sample of 52 languages. Malchukov
(2004: 23) also tests the full range of nominal features on his 50-language
sample and reports confirmation of the hierarchical pattern.
Malchukov (2004: 47) integrates the findings from the studies discussed
above into the nominal feature hierarchy represented in (30) 47:
(30) NOUN stem

classifying / qualitative operators / satellites: singulative / collective
markers, noun classifiers, nominal class markers, adjectives

quantitative operators / satellites: number markers, numerals

locative / possessive phrases

determiners

case markers / adpositions
3.3.2.4 Summary, outlook

Having introduced the hierarchies of verbal and nominal categories, the next
question is how these hierarchies are reflected in the behavioural potential
of DC constructions. In other words, what kinds of mixing patterns of
verbal and nominal features are actually displayed by DC constructions of
the world’s languages? This is the topic of the next subsection.
3.3.3 Feature mixing in DCs across languages:

Cross-linguistic generalizations and functional motivations
Since the 1970s, a considerable number of typological studies have been carried
out on the behavioural potential of DCs (e.g. Comrie 1976; Noonan 1985/2007;
Lehmann 1988; Koptjevskaja-Tamm1993; Mackenzie 1996; Croft 1991, 2001;
Dik 1997; Cristofaro 2003; Malchukov 2004, 2006; Dixon & Aikhenvald
47
Note that the most recent version of Rijkhoff ’s model, as presented in Rijkhof 2008a / b, is
not incorporated in Malchukov’s 2004 study. This is why the latter does not distinguish the
innermost level of classifying modification. Since the distinction is not directly relevant to my
research, I have not added it to Malchukov’s model as represented in (30).

2006). Together, these studies provide a body of evidence showing that the
cross-linguistic variation in DC constructions takes the form of a continuum,
ranging from DCs that express the full set of available verbal categories and no
nominal features, to DCs that have lost (almost) all verbal features and express
(almost) all nominal features. In the literature, starting with Stassen (1985),
DCs of the first type are called balanced; they can express the same set of verbal
features as independent clauses in the particular language under analysis. All
other DC types are called deranked, meaning that their behavioural potential
deviates to a greater or lesser extent from the coding of an independent clause,
in terms of de-categorization (i.e. elimination of verbal categories) and/or re-
categorization (i.e. expression of nominal categories). As already anticipated
by the discussion of the verbal and nominal feature hierarchies in the previous
subsection, formal deviation from independent-clause-coding may affect
different elements of the DC. The individual structural phenomena that may
obtain are listed in (31a-f ) (cf. Croft 2001):
(31) a. (Partial) elimination of tense / mood / aspect marking (operators

and / or satellites);
b. (Partial) elimination of person marking;
c. Expression of definiteness (determiners) and / or
case / adpositions;
d. Expression of number and / or class;
e. Non-expression of argument(s) in the dependent clause, or
expression of argument(s) in a different manner than in an
independent clause;
f. Adjectival rather than adverbial expression of lexical
modifier(s) 48,49.
Even though there are some minor differences, the typological studies
mentioned above reveal broadly the same generalizations as regards the (co-)
occurrence in DC constructions of the features listed in (31). In what follows
I will not give an overview of this literature 50. Rather, I will discuss two of its
48
For the time being I ignore the possibility that TAM and / or agreement are coded by means
of dependent / subjunctive forms. This phenomenon is of minor importance to the present
discussion, but I will return to it in section 3.4.
49
Usage of different word order than in simple main clauses is another possible form of
structural deviation. However, since it is quite marginal and usually occurs in combination
with some other strategy, I do not take it into account in the present study.
50
See Malchukov (2004: 9-12) for a brief overview of this kind.

most recent exponents in some detail, namely the work of Cristofaro (2003),
and Malchukov (2004).
3.3.3.2 The typology of subordination (Cristofaro 2003)

First, Cristofaro’s work probably represents the broadest typological study
of subordination to date: It takes into account complement clauses, relative
clauses, and adverbial clauses in a sample of 80 languages. Cristofaro identifies
a number of correlations between various deranking phenomena in DCs.
These correlations are listed in (32a-h) (Cristofaro 2003, chapter 10) 51:
(32) a. Agreement not expressed 2 TAM not expressed

b. Case / adposition marking 2 TAM not expressed
c. Case / adposition marking 2 Agreement not expressed
d. Arguments expressed as POSS 2 TAM not expressed
e. Arguments expressed as POSS 2 Agreement not expressed
f. Arguments expressed as POSS 2 Case / adposition marking
g. Arguments not expressed 2 TAM not expressed
h. Arguments not expressed 2 Agreement not expressed
Cristofaro proposes three types of functional factors underlying these

correlations, two of which are the general functional principles of
(Syntagmatic) Economy and Iconicity (Haiman 1983). These principles
account for the non-expression of verbal features in DCs, including the
non-expression of arguments.
First, consider the principle of Economy, which refers to “the tendency to
reduce as much as possible the length or complexity of any utterance” (Cristofaro
2003: 9). As Cristofaro shows, in some subordination relations the semantics
of the matrix clause predetermine certain TAM values and / or a participant of
the SoA expressed by the dependent clause. There are three possibilities in
this respect:
(i) The matrix clause predetermines both TAM value(s) and a

participant of the dependent clause;
51
The generalizations in (32b) and (32c) are not supported by my data. In addition, as I will
become clear shortly, they are not in accordance with the typology proposed by Malchukov
(2004). See Lehmann (1988) for a proposal similar to Malchukov’s, and Mackenzie (1996)
for a proposal that supports Cristofaro’s findings. See also Fischer & Van Lier (forthcoming)
for more discussion.

(ii) The matrix clause predetermines TAM value(s) (but no participant)
of the dependent clause;
(iii) The matrix clause predetermines neither TAM value(s) nor any
participant of the dependent clause.
There are no subordination relations in which TAM values are predetermined,

while participants are not, which means that predetermination of participants
implies predetermination of TAM values. The Economy principle predicts
that predetermined information may be left unexpressed (this is called the
Principle of Information Recoverability by Cristofaro (2003: 258)). Thus, this
principle explains the correlation mentioned under (32g) above, namely that
non-expression of arguments implies non-expression of TAM distinctions.
Moreover, the non-expression of arguments in turn implies the non-
expression of concomitant agreement morphology, as is reflected in the
correlation mentioned under (32h) above.
The principle of Iconicity refers to “the tendency to shape linguistic structure

in conformity with the structure of conceptual experience” (Cristofaro 2003:
8). Iconicity explains the same structural phenomena as the principle of
Economy, namely non-expression of verbal categories and arguments in
DCs, but it does so in terms of semantic integration, rather than in terms
of semantic predetermination 52. Increased semantic integration between the
SoAs expressed by the two linked clauses arguably leads to reduced linguistic
independence of the DC, which is in turn reflected in the non-expression
of verbal features and / or arguments in the latter. This point was already
made by Givón (1980 and later work), who proposed the so-called binding
hierarchy or scale of event integration, stating that “the stronger the semantic
bond between two events, the more extensive will be the syntactic integration of
the two clauses into a single though complex clause.” (Givón 2001: 40)
However, neither Economy nor Iconicity can account for the
expression of nominal features in DCs. Therefore, Cristofaro proposes
a third explanatory principle, which has a basis in Cognitive Linguistics
(Langacker 1987). According to this principle, DCs are conceptually not
52
This suggests that the principle of Iconicity is superfluous, to the extent that it does not
explain any more than the Economy principle does. Indeed, Haspelmath (2008) argues
that Iconicity is irrelevant for the explanation of cross-linguistic generalizations found in
the coding of DCs. In his view, they should rather be explained in terms of frequency of
use. Since frequency implies predictability, this amounts to an explanation in terms of the
Economy principle.

construed as independent processes, and may instead be construed as things
or properties. This means that they are interpreted ‘as wholes’, without
any internal temporal development (cf. Chapter 2, section 2.2). Using the
terminology of Cognitive Linguistics, DCs are said to be “summarily rather
than sequentially scanned.” (Cristofaro 2003: 284)
In fact, according to Cristofaro, this principle of cognitive thing-construal
explains both the non-expression of verbal features and the expression of
nominal features in DCs 53. First, verbal categories such as TAM values
express meanings that are semantically relevant to processes or SoAs, but not
to things. Therefore, if a DC is construed as a thing, then TAM distinctions
can be omitted. A similar argument can be made for verbal agreement: It
marks the relationship between a predicate and its arguments, and as such
it is relevant to an SoA, but not to a thing. The fact that non-expression of
TAM and non-expression of agreement have a parallel motivation accounts
for the correlation between these two phenomena, mentioned under (32a)
above and repeated here for convenience as (33):
(33) Agreement not expressed 2 TAM not expressed
Second, Cristofaro claims that the conceptualization of DCs as things

explains the expression of nominal properties such as case or adposition
marking, and the coding of arguments as possessors. In particular, she
argues that a DC may “display the same properties as the grammatical entities
that prototypically code things, that is, nouns.” (Cristofaro 2003: 262-63)
Now, conceptualization as a thing presupposes non-sequential scanning.
Thing-construal is formally reflected by the expression of nominal categories,
while non-sequential scanning is reflected by non-expression of verbal
categories. This explains the universals of co-occurrence, as listed under
(32b-f ) and repeated here as (34a-e) (Cristofaro 2003: 284-285). Specifically,
the expression of the nominal features of case / adposition and possessive
argument coding implies the non-expression of the verbal features TAM
and agreement. The correlation between coding of arguments as possessors
and the expression of case / adpositions (see 34e) is explained by the fact that
these phenomena both reflect conceptualization of the DC as a thing.
53
For ease of reference, in the remainder of the discussion I will use thing-construal, rather
than to thing- or property-construal. Basically, however, the same claims are made for DCs that
are cognitively construed as properties as for those that are construed as things. For further
details on property-construal see Cristofaro (2003).

(34) a. Case / adposition marking 2 TAM not expressed
b. Case / adposition marking 2 Agreement not expressed
c. Arguments expressed as POSS 2 TAM not expressed
d. Arguments expressed as POSS 2 Agreement not expressed
e. Arguments expressed as POSS 2 Case / adposition marking
Clearly, this third explanatory principle – thing-construal of DCs – is most

relevant to the present study, since it hinges on a functional parallel between
DCs and PoS. Specifically, Cristofaro’s explanation makes reference to the
notional similarity between nouns and DCs to explain the attested formal
patterns: both are construed as ‘things’. However, recall from Chapter 2
that Croft defines prototypical nouns as lexical constructions that combine
the semantic denotation of thing (or ‘object’ in Croft’s terms) with the
pragmatic function of reference (see section 2.4.2). Recall further that, as
argued in section 3.2, it is the pragmatic or propositional function, rather
than the semantic denotation that DCs have in common with PoS. In fact,
DCs do not denote things or properties; they denote higher order entities
such as SoAs and propositions 54,55. Therefore, it seems that the expression
of nominal (or other non-verbal) features in DCs should be explained in
terms of the fact that DCs share the propositional function of reference or
modification with prototypical nouns, adjectives, and manner adverbs. Such
a pragmatic explanation for patterns of nominal and verbal feature mixing
in DCs is proposed in the second typological study of subordination that I
want to discuss: Malchukov’s work on nominalizations, as outlined in the
next subsection.
3.3.3.3 The typology of nominalization (Malchukov 2004)

Malchukov’s typological study of clausal nominalizations shows that verbal
and nominal categories at the external end of their respective feature
hierarchies (as discussed in section 3.3.2.3 above) are more readily affected
by processes of de-categorization and re-categorization in DCs than
54
cf. Noonan, who defines complementation the as situation in which “a notional sentence or
predication is an argument of a predicate” (Noonan 2007: 52; emphasis added, EvL).
55
It may be argued that cognitive thing-construal is not the same semantic thing-denotation.
However, as Cristofaro (2003: 302-303) herself admits, there is no independent evidence for
cognitive thing-construal. Cristofaro argues that the cross-linguistic generalizations attested
in her study can be regarded as supporting her cognitive explanation. I would argue, however,
that the linguistic facts to be explained in terms of cognitive construal cannot at the same
time function as evidence for such cognitive construal.

internal features. According to Malchukov, the functional explanation for
this generalization must be sought in the fact that external categories reflect
the syntactic and / or pragmatic function of the linguistic unit on which they
operate, while internal categories are relevant for the semantic interpretation
of their base. Specifically, since “the ultimate task of nominalization is to
produce a referential expression” (Malchukov 2004: 26), this explains why
external nominal features, which are relevant to the pragmatic function of
reference, are expressed in nominalizations before internal ones. In addition,
it explains why external verbal features, which are relevant to pragmatic and
syntactic aspects of (independent) predications, are the most likely ones to
be lost in nominalization processes.
At this point, it should be mentioned that the idea of a pragmatic
motivation for feature mixing in DCs ties in with Dik’s Principle of Formal
Adjustment (PFA). This principle states that:
“Derived, secondary constructions of type X will be formally assimilated

(adjusted) to non-derived, primary constructions of type X.”
(Dik 1997: 158)
Applying the PFA to the case at hand, we can say that DCs are derived,
secondary referential or modifying constructions (rather than derived,
secondary thing-denoting or property-denoting constructions). Their
primary, non-derived referential or modifying counterparts are lexical nouns,
adjectives, and adverbs. Thus, Dik’s PFA predicts that DCs, as secondary
constructions, will adopt formal features of primary lexical constructions.
Turning back to Malchukov’s study, his finding that external features are
affected in nominalizations before internal ones is further specified in terms
of two implicational hierarchies. These implicational hierarchies are based
directly on the feature hierarchies discussed in section 3.3.2 above. Regarding
nominal categories, the generalization is that the expression of a particular
feature in a nominalization implies the expression of any feature that occupies
a more external position on the nominal feature hierarchy. Regarding verbal
categories, in contrast, the loss (or non-expression) of a particular feature in
a nominalization construction entails the loss of any feature that occupies a
more external position on the verbal feature hierarchy.
In addition to these implicational effects of the verbal and nominal feature
hierarchies, Malchukov claims that there are other factors that constrain cross-
linguistic variation in verbal and nominal feature combinations. The most

important factor involves so-called blocking-effects. Basically, ‘blocking’ refers
to the fact that certain combinations of nominal and verbal categories are
incompatible as a result of their functional equivalence. In particular, arguments
and modifiers can be expressed either verbally or nominally, but not in both
ways at the same time. Therefore, the verbal expression of an argument or a
modifier blocks the nominal expression of that element, and vice versa.
Malchukov (2004: 66-67) distinguishes three types of blocking effects:
subject blocking, object blocking, and modifier blocking. In combination
with the implicational effects of the verbal and nominal feature hierarchies,
this yields a three-way typology of nominalizations, consisting of the
following types:
(i) nominalizations with no blocking;

(ii) nominalizations with subject blocking;
(iii) nominalizations with object blocking or modifier blocking.
First consider nominalizations with no blocking. As far as nominal

features are concerned, this type of nominalization may express only the
two outermost nominal categories (in terms of the feature hierarchy), i.e.
determiners and case markers. This is because these categories create no
blocking effects, as they have no verbal functional counterparts. This means
that all verbal features can be retained (with the exception of illocutionary
force, which is unexpressed by definition in dependent clauses), including
the expression of arguments and modifiers as in independent clauses 56. The
feature combination that thus characterizes nominalizations with no blocking
is represented in (35) below, where the dashed line indicates the boundary
between verbal features (above the line) and nominal features (below the
line), i.e. the point where the two feature hierarchies intersect. The brackets
around the nominal features indicate optionality: case / adposition and
determiners can but need not be expressed in nominalizations of this type.
56
Note that this is contra Cristofaro’s generalizations in (34a) and (34b) above, according to
which case or adposition marking implies the loss of verbal features (TAM and AGR).

(35) Nominalizations with no blocking:
STEM


aspect

tense and mood

subject agreement, clausal subject V
--------------------------------------------------------------------------
determiners N

The second type of nominalization involves subject blocking. This means

that one argument – the subject or agent – is expressed as a possessor.
This nominal subject encoding blocks verbal subject encoding and as such
also disrupts concomitant verbal subject agreement. This in turn implies
the non-expression of any verbal features external to subject agreement,
i.e. the category of illocutionary force. All other verbal categories can be
retained, and any arguments other than the subject / agent are expressed as
in independent clauses. The feature combination of nominalizations with
subject blocking is represented in (36), again with a dashed line marking the
boundary between the verbal and the nominal feature hierarchies:

(36) Nominalizations with subject-blocking:
STEM


aspect

tense and mood V
--------------------------------------------------------------------------
possessive subject N

determiners

The third and final type concerns nominalizations with object blocking or
modifier blocking. As its label makes clear, this type is in fact a combination of
two types, which are merged because they look formally very much alike. In
these nominalizations, the object or patient argument is coded as a possessor
and / or modification is adjectival. These two features involve blocking
effects that pertain to the innermost layers of both hierarchies: Object
blocking affects the valency layer of the verbal hierarchy, while modifier
blocking affects the qualitative layer of the nominal hierarchy. Therefore,
they imply the loss of all other verbal categories and the expression of all
other nominal categories, respectively. The morpho-syntactic effects of both
types of blocking are thus quite similar. The only difference is that in the
case of adjectival modification the valency layer of the verbal hierarchy need
not be affected, as it always is in the case of possessive object expression.
The feature combination of nominalizations with object / modifier blocking
is represented in (37):

(37) Nominalizations with object / modifier-blocking:
STEM

[voice / valency, direct object, object agreement] V
--------------------------------------------------------------------------
classifying / qualifying operators / satellites (adjectival modification) N

quantitative operators / satellites: number marking / numerals

possessive subject

determiners

Malchukov’s framework thus predicts a three-way typology of nominalization

constructions. However, not all nominalizations attested in the languages of
his sample fit the above typology. According to Malchukov, this is because
there are several structural factors that may complicate the combined effects
of blocking and implicational hierarchies.
First, there is a difference between the behaviour of morpho-syntactic
versus lexical instantiations of the same categories, i.e. between operators
and satellites. In particular, the possibility to express a specific nominal or
verbal category by means of an operator implies the possibility to express that
category with a satellite, but not vice versa: the expression of a satellite does
not mean that the corresponding operator is also available. Malchukov (2004:
37) calls this the Operator-Satellite Asymmetry principle (henceforth OSAP).
Recall from section 3.3.2.2 that Malchukov regards lexical arguments and
agreement morphology as, respectively, the satellite and the operator type
expressions of the same category. Thus, since nominal argument expression
blocks verbal argument expression (satellites), the OSAP predicts the loss of
concomitant verbal agreement (operators). In contrast, verbal expression of
arguments (satellites) can go together with either retention or loss of verbal
agreement (operators). With respect to modifier expression, the OSAP
explains why adverbial modification (satellites) does not entail the availability
of morphologically expressed TAM distinctions (operators). This means that
for nominalizations without modifier blocking no strong predictions can be
made about TAM operators; they may be either lost or retained.

Second, the outcome of feature mixing in nominalization constructions
can be complicated by cumulative expression of multiple categories through
a single morpheme. If such categories belong to different levels of either the
verbal or the nominal feature hierarchy, then a conflict arises that will result
in either the retention or the loss of all categories expressed by the relevant
morpheme. (See Malchukov (2004: 111) for more discussion and examples,
and Malchukov (2007) for an Optimality Theory implementation of the
interaction between functional and structural factors in determining the
expression of nominalization constructions.)
It is noteworthy in the context of the present study that Malchukov
explicitly claims his typology of nominalizations to be applicable to
languages without a lexical distinction between nouns and verbs. He argues
that the model works as long as a noun-verb distinction is present on
the syntactic level, i.e. as long as verb phrases and noun phrases involve
different categories (Malchukov 2004: 126). In other words, the expression
of ‘nominal’ categories in a DC does not necessarily imply the existence of a
class of rigid lexical nouns; it merely means that there is a certain phrase type
– headed by a lexical element whose categorial value may be indeterminate
in isolation – that is used to express the function of reference.
This is also the reason why in the present study the relevant set of DC
constructions is initially determined on the basis of structural coding, rather
than behavioural potential. The expression of behavioural potential in a
DC reflects its non-prototypical combination of propositional function and
semantic denotation; it does not have a direct connection with the lexical
categories available in a particular language. However, once a particular DC
construction is determined on the basis of the structural coding criterion, it
may be further classified according to its behavioural potential. This allows
one to address a second question, namely whether the internal structure of
DC constructions is a factor of influence on the relation between DCs and
PoS, in terms of their ability to express particular propositional functions.
3.3.4 Summary, outlook

In this section I discussed the typology of the internal morpho-syntax
of DCs. Starting out with a presentation of verbal and nominal feature
hierarchies, I subsequently discussed cross-linguistic patterns pertaining to
the mixing of these two feature types in DC constructions, and the functional
motivations that have been proposed to explain these patterns. This section
thus provides the basis for the second part of the typology that I will present

in the next section, and that will be used in Chapter 6 to classify the DCs of
the languages in my sample.
3.4 A two-step typology of dependent clauses

The classification of DCs that I propose involves a two-step procedure, which
integrates the two types of characteristics of DCs that have been at stake in
the present chapter: structural coding and behavioural potential. First, DCs
are identified purely on the basis of their structural coding, i.e. the form of
either the subordinating conjunction or the special marker on the dependent
predicate. For every DC construction identified in this way, it is determined
which propositional function(s) it can express. This yields a classification of
DCs in terms of the rigid and flexible types presented in section 3.2.2.
Secondly, every rigid or flexible DC construction can be classified
according to its internal morpho-syntactic properties, i.e. its behavioural
potential. For this part of the typology I will adopt a three-way distinction
between DC types, to be presented shortly. This typology is quite similar
to Malchukov’s classification of nominalizations discussed above, but also
differs from it in some respects. A first difference is that Malchukov’s
study does not include fully balanced DCs as a separate category. His
nominalization type with no blocking must have verbal arguments, but it
does not necessarily express all TAM / agreement categories. In contrast, I
will distinguish between:
(i) DCs that express their arguments as in independent clauses (i.e.
verbally) and retain all verbal (TAM / person) categories that can be
expressed in independent clauses;
(ii) DCs that express their arguments as in independent clauses (i.e.
verbally) but do not retain (all) the verbal (TAM / person) categories
that can be expressed in independent clauses.
Note that both these DC types may or may not combine with case markers
or determiners, an issue to which I return below.
Second, Malchukov uses the verbal versus nominal encoding of subject
and object arguments as one of the decisive criteria for his classification of
nominalizations. This means that his model does not provide an obvious
locus for DCs in which one or more argument(s) remain un-expressed. The
typology that will be used here does take this possibility into account, as I
will explain below.

Third, while Malchukov makes a principled distinction between
constructions with one possessive argument (nominalizations with subject
blocking) and constructions with two possessive arguments (nominalizations
with object blocking), I will regard these two types as belonging to a single
DC type, in which at least one argument is expressed nominally.
In line with previous studies on subordination, the three-way typology of
DCs that I will adopt is based on three formal parameters, as listed in (38):
(38) a. The (partial) (non-)expression of TAM operators and / or person

marking 57;
b. The (non-)expression of the nominal categories Determiner
and / or Case / adposition;
c. The coding of the argument(s).
Some remarks are required concerning these three parameters. First, it

may be noted that the verbal versus nominal expression of lexical modifiers
(i.e. adverbial versus adjectival modifiers) is not taken into account. This is
because this feature is not often well-documented in reference grammars.
Second, in terms of verbal operators (parameter (38a)), the following
features will be taken into account: voice and valency marking; tense, aspect,
and mood marking; person marking; and illocutionary force. Following
Malchukov (2004; and see also Noonan 2007: 68), object marking will be
regarded as pertaining to the innermost level of verbal valency.
Third, with regard to nominal features (parameter (38b)), I take into
account only the expression of determiners and case markers or adpositions.
Like Cristofaro (2003), but unlike Malchukov (2004), I do not systematically
consider other nominal features, such as number and gender / class marking,
since they are very infrequently attested in the DCs of the sample languages
(cf. Chapter 6).
Finally, consider parameter (38c) in some more detail. Argument coding
in DCs can take three possible forms, as listed in (39):
(39) a. An argument is expressed as it would be in an independent

clause;
57
As mentioned earlier, I use the term person marking to include both agreement and cross-
referencing on the predicate.

b. An argument is expressed in a different way than it would be in
an independent clause;
c. An argument is not expressed.
The possibilities in (39a) and (39c) are straightforward enough. Following

the terminological convention of Koptjevskaja-Tamm (1993), I will call the
type of expression in (39a) ‘sentential’, abbreviated as SENT, and the type
in (39c) ‘zero’, indicated by Ø. The parameter value in (39b), however, may
manifest itself in several different forms. Most typically, it involves coding
of the argument as a possessor, as in example (40) from Turkish (note that
the possessive subject triggers concomitant nominal agreement on the
dependent predicate).

(40) [Kerkes-in birier kikaye anlat-ma-sɪ]
everyone-gen one.each story tell-nmlz-3sg.poss
iste-n-iyor-muş
want-pass-ipfv-ev.cop
‘It seems that they want everyone to tell a story.’
In other cases, however, the argument may be coded as an oblique. This may
involve an adposition, as in example (41) from Georgian, or a special case
marker, such as the dative in example (42), also from Georgian.
Georgian (Hewitt 1995: 542, 609)

(41) [mokalake-ta mier upleb-eb-is
citizen-pl(gen ) by rights-pl-gen
ga-mo-q’en-eb-a]
prev-prev-use-ths-nmlz
‘the making use of their rights by the citizens’
(42) [salit’eraturo ena-s sa-upl-eb-ul-i]

literary language-dat prev-master-ths-pst.ptc-nom
‘(he) who has mastered the literary language’
A third possibility is the encoding of an argument in the form of an attributive

or associative (nominal) modifier, as in example (43) from Lango.

Lango (Noonan 1992: 214)
(43) [Nè-kkò gwôkk à dákô] ràc
kill-nmlz dog attr.prt woman 3sg-bad-hab
‘The killing of a dog by a woman is bad.’
Finally, an argument may be incorporated into the dependent predicate, as

in the Ket example in (44):
Ket (Vajda 2004: 78)

(44) [ə́k-nà qús-bèt] bínùt
1pl-anim.pl.gen tent-make.inf it.ended
‘We finished making the tent.’ (lit. ‘Our tent making ended.’)
In this study, all sub-types of non-sentential argument coding

illustrated in examples (40)-(44) – possessor-coding, oblique-coding,
attributive / associative coding, and incorporation – will be regarded as a
single strategy. This strategy is characterized by the fact that it deviates
from argument coding in independent clauses, and will be called alternative
coding, abbreviated as ALT 58.
A final issue, related to argument coding, concerns alignment patterns. In
this study I will distinguish between first and second arguments only. In this
way I generalize over constructions with nominative-accusative and ergative-
absolutive alignment. This is unlike the approach taken by Koptjevskaja-
Tamm (1993, 2003, 2005), who distinguishes between so-called possessive-
accusative and ergative-possessive constructions. In the former type, the S and
A arguments are expressed as possessors, while the P argument is expressed
as in an independent clause (i.e. with accusative coding). In the latter case,
the S and P arguments are expressed as possessors, in contrast to the A,
which is coded as in an independent clause (i.e. with ergative coding). In
my classification, these two types of alignment are collapsed into a single
strategy, characterized by the fact that in a transitive DC construction
one argument gets alternative coding (ALT), while the other one retains
sentential coding (SENT).
58
cf. Koptjevskaja-Tamm (1993, 2003, 2005), who treats possessive coding, oblique coding,
and incorporation all as separate types. Note that initially I have followed this more
finegrained distinction, but abandoned it again in a later stage, as it did not influence the
results in any interesting way.

Taking these remarks into account, I present below the three-way typology
of DC constructions to be used in this study. Note that each type is indicated
with both a number and a mnemonic label.
• Type 1: Balanced constructions (abbreviated as B)

Characterized by the following three parameter values:
a. All TAM distinctions and person marking are expressed as in
independent clauses;
b. All overt arguments are coded as in independent clauses
(SENT-SENT);
c. Determiners (DET) and / or case markers / adpositions
(CASE) 59 can but need not be expressed.
With respect to the second parameter value (argument expression), I regard

relative clause constructions with a gap-strategy as balanced, even though
one of the arguments (the relativized one) of the DC remains unexpressed.
Relative clauses with a gap are defined by Comrie & Kuteva (2005: 494)
as those involving “no overt case-marked reference to the head noun within the
relative clause”.
The third parameter value, according to which balanced DCs may
combine with determiners and / or case markers / adpositions, also requires
some clarification. Recall that in Malchukov’s typology a balanced clause
with a case marker or adposition would be a nominalization with no
blocking, which allows for the retention of all verbal categories (see (35)
above). This type of construction would constitute a counterexample to the
universal claimed by Cristofaro (2003), namely that the expression of case
or adposition marking implies the loss of TAM and agreement (see (32b-
c) / (34a-b) above). Nonetheless, there are several instances of this type of
construction in my sample. For instance, Guaraní has a complement clause
construction that takes a determiner but is otherwise completely balanced,
as can be seen in (45). Basque has balanced complement clauses with
bound complementizers and determiners, as is illustrated in (46). A similar
construction, with a bound complementizer and a case marker, is attested in
Cofán, as shown in (47).
59
The abbreviation CASE is meant to include both case markers and adpositions.

Guaraní (Gregores & Suárez 1967:158)
(45) ai-kwaá [la [n o-ù moɁá̃ i há]]
I-know art neg he-go mod neg comp
‘I know that he does not intend to go.’
Basque (Hualde & Ortiz de Urbina 2003: 646)

(46) Entzun dut [[Amaiaren neba hil d-en]-a]
hear aux Amaia.gen brother die aux-comp-det
‘I heard that Amaia’s brother died.’
Cofan (Fischer & Van Lier, forthcoming)

(47) Atesu-mbi [[ke ña=nga tevaen='cho]=ma]
know-neg you I=dat write=comp=acc
‘I didn’t know that you had written to me.’
The fact that these constructions are fully balanced, apart from the expression
of a determiner or case marker, is in accordance with the nominal feature
hierarchy and with Malchukov’s typology: Determiners and case-markers
are external features that mark the syntactic-pragmatic function of their
base unit, without affecting its internal morpho-syntactic structure.
Note that determiners and case markers may also be expressed on balanced
DCs that are not marked by a complementizer60. For determiners this was
already illustrated with an example from Dhaasanac (see (5)) in section 3.2.2).
A similar construction is attested in Cofán, as is illustrated in (48):
Cofán (Fischer & Van Lier: forthcoming)

(48) [Doscientos ochenta dolar=ma gana-’je-’fa]=ma
two.hundred eighty dollar=acc earn-ipfv-pl=acc
in'jan kan-'se.
think try-mit
‘Imagine that they earn 280 dollars.’
Finally, it must be noted that I take balanced clauses (type 1) to include two
sub-types that are strictly speaking not fully balanced. Both deviate slightly
from independent clause coding, particularly in terms of the expression of
verbal features. These two subtypes are:
60
In fact, case markers and adpositions are a common historical source for complementizers.

(i) DCs that are fully balanced except for the fact that they cannot
express illocutionary force marking, while this is an obligatory
category in independent clauses.
(ii) DCs in which the same or almost the same range of verbal
categories can be expressed as independent clauses, but by means of
dependent or subjunctive forms.
The first sub-type is attested for instance in Nama. In this language, the
declarative marker ke occurs in every independent clause, while it cannot
be expressed in DCs. In all other respects, however, DCs and independent
clauses have the same structure. Abkhaz presents an example of the second
subtype. In this language, certain DC constructions make use of the so-
called ‘non-finite’ verbal paradigm, as illustrated in (49). This paradigm
includes almost the same distinctions as the independent paradigm, but
expresses them with different forms. As already mentioned in section 3.2.2
(see example (7)), I will regard DCs with dependent or subjunctive forms as
belonging to the group of balanced DCs.
Abkhaz (Hewitt 1987: 238)

(49) [d-şaa-wà] (Ø-)xà-s-c’a-wa-yt’.
he-come-dyn(n.fin.prs) it-head.in-I-put-dyn-fin(prs)
‘I believe that he will come.’
We have now rounded off the discussion of the first DC type, and may
consider the second one:
• T
ype 2: Deranked DCs with SENT arguments (Abbreviated as
D-SENT)
Characterized by the following parameter-values:
d. (Partial) non-expression of TAM distinctions and / or Person
marking;
e. All overt arguments are expressed as in independent clauses
(SENT-SENT);
f. DET / CASE can but need not be expressed.
This D-SENT type also (like type 1) resembles Malchukov’s nominalizations

with no blocking (see (35) above), but differs from it in two respects, namely:

(i) Malchukov’s Nominalizations with no blocking do not necessarily
involve (partial) loss of TAM and / or person distinctions, whereas
my Type 2 D-SENT DCs do;
(ii) Malchukov’s Nominalizations with no blocking do not explicitly
include constructions in which one or more arguments remain un-
expressed, whereas my Type 2 D-SENT DCs do.
An example of a D-SENT construction with two overt arguments is given

in (50). This Tamil nominalization involves loss of agreement and partial
loss of aspect and mood distinctions, but it retains tense.
Tamil (Asher 1982: 20)

(50) [tiruʈan nakeyellaam eʈuttukkiʈʈirukkar-ate]
thief jewellery.all take.progr.prs-nmlz:acc
naan patteen
I see-pst-1sg
‘I saw the thief taking the jewels.’
D-SENT constructions can but need not be marked for CASE or DET.
Some D-SENT constructions are never marked for CASE / DET, such as
the Polish infinitive construction illustrated in (51):
Polish (Bielec 1998: 19)

(51) Postaniwiɫem [kupić dom].
I.decided buy.inf house
‘I decided to buy a house.’
Other D-SENT constructions are marked for case in some functions but
not in others. For instance, Imbabura Quechua has a flexible D-SENT
construction, which is case-marked in the function of complement clause
(52a), but not in the function of relative clause (52b); except if the relative
clause is extra-posed, then the head and the relative clause are both marked
for case (52c).

Imbabura Quechua (Cole 1982: 14, 47, 51)
(52) a. Marya nin-n [Juzi jatun wasi-ta cari-shka]-ta
María say-3 José big house-acc have-nmlz:pst-acc
‘Maria says that José had a big house.’
b. [Marya riku-shka] runa

María see-ptc:pst man
‘the man whom Maria saw’
c. Kwitsa-ta juya-ni [Juan-wan tushu-shka

girl-acc love-1 Juan-with dance-ptc:pst
ka-shka]-ta
be-ptc.pst-acc
‘I love the girl who had danced with Juan.’
These Polish and Imbabura Quechua examples also show that in D-SENT
constructions one argument may remain unexpressed: Example (51)
illustrates the zero-expression of a co-referential subject argument, and
(52b-c) illustrate the use of a gap strategy in a D-SENT construction. Other
D-SENT constructions with an unexpressed argument include converbal
constructions with co-referential subjects, such as the one illustrated in (53)
from Turkish:

(53) Ben [etraf-ım-a bak-arak] yür-ür-üm
I around-1sg-dat look-conv walk-aor-1sg
‘I walk looking around (myself ).’
Finally, D-SENT constructions may lack arguments altogether, even if they

are derived from transitive clauses. Such cases typically involve past / passive
or future participles, in which the agent argument is demoted and the
patient is gapped (Haspelmath 1994). This is illustrated in (54) with a future
participle construction from Georgian.

Georgian (Hewitt 1995: 609)
(54) Ager m-i-k’av-i-a
here I-ov-hold-prs.stat-3
[xel-mo-sa-c’er]-i kagald-eb-i
hand-prev-fut.ptc-sign-nom paper-pl-nom
‘I am holding here the papers which are to be signed.’
Having discussed DC constructions of type 2, I turn now to the third type:
• Type 3: Deranked DCs with at least one alternatively coded argument

(Abbreviated as D-ALT)
Characterized by the following parameter-values:
– (Partial) non-expression of TAM distinctions and / or person
marking;
– At least one argument gets alternative (ALT) expression; the
other one, if available / overt, gets either sentential (SENT) or
alternative (ALT) expression;
– DET / CASE are expressed (if relevant in the language).
The Turkish example in (40) above illustrates a DC of type 3 with two overt
arguments, one of which has ALT coding, while the other remains SENT.
An example of a Type 3 DC with two ALT-coded arguments appears
in (41) above; a Georgian nominalization construction, in which the
(transitive) subject takes genitive case, and the object takes oblique coding,
with the proposition mier ‘by’ (which also requires genitive case). Tense,
mood, and person / number agreement are lost, but aspect is retained. This
construction can be case-marked or combined with a postposition according
to its function in the matrix clause.
Non-expression of an argument may also occur in a Type 3 DC, namely
when it combines with a gap strategy. An example is given in (55); a participial
construction (again from Georgian), in which the agent argument is in the
genitive case and the relativized patient is gapped:
Georgian (Hewitt 1995: 611-612)

(55) [razik’ašvil-is ča-c’er-il]-i varian’-i
Razikashvili-gen prev-record-pst-ptc-nom variant-nom
‘the variant recorded by Razikashvili.’

We have now defined three structural types of DCs. In Table 3.1, these three
types and their respective parameter values are summarized:
DC type Argument expression TAM / Person DET / CASE

number label 1 2
Type 1 B SENT / Ø SENT / Ø + +/–
Type 2 D-SENT SENT / Ø SENT / Ø – +/–
Type 3 D-ALT ALT SENT / ALT / Ø – +
Table 3.1: DC types and their internal formal properties
3.5 Summary
In this chapter I have discussed two approaches to the typology of dependent
clauses: One based on structural coding of DC constructions and their
possibilities to express certain propositional functions; the other based on the
internal morpho-syntactic properties of DCs, i.e. their behavioural potential.
Concerning the first approach, I showed that DCs can be defined as
mappings onto a space of four propositional functions, in parallel with
the definitions of PoS classes developed in Chapter 2. This involved the
application of the notions of flexibility and rigidity to the domain of DC
constructions, which in turn required a definition of structural coding of
these constructions: subordinating conjunctions and dependent predicate
markers. A number of expected rigid and flexible DC types were defined in
this way.
Second, I discussed the internal morpho-syntax of DCs against the
background of the functional-typological literature on verbal and nominal
feature hierarchies, and the way they interact in DC constructions across
languages. Based on this literature, I developed a classification system of
three structural DC types, defined in terms of the (non-)expression of TAM
distinctions and Person marking, the type of argument encoding, and the
(non-)expression of determiners and / or case markers.
In Chapter 6 the integrated DC typology, based on structural coding and
behavioural potential, will be used to classify the DC constructions attested
in the languages of the sample.

Sampling, Research Questions,
Method
4.1 Introduction
The aims of this chapter are as follows: First, in section 4.2, it presents
the sample of languages investigated in this study. Second, in section 4.3,
the concrete research questions that will be addressed in this study are
formulated and operationalized. Third, section 4.4 explains the method that
will be used to answer these questions. Finally, section 4.5 provides a brief
outlook on the remaining part of the study.
4.2 The language sample

The sample used for this study consists of 50 languages and is composed by
means of the Diversity Value Technique (henceforth DVT; Rijkhoff et al.
1993, Bakker forthcoming). This technique can be used with any language
classification. A sample composed with the DVT consists minimally of one
language per family, according to the classification chosen. Absolute isolates
(as opposed to isolates within a genetic grouping) are always part of any
sample. Once such a minimal sample is put together, it can be expanded
– depending on the researcher’s wishes – on the basis of so-called diversity
values. These values are assigned to all nodes in the language family tree.
They express the complexity of the tree below the node in terms of the
amount of daughter nodes and the way they are embedded. The diversity
value determines the proportion of languages to be drawn from under the
relevant node in relation to its sister nodes, given a certain desired sample
109
size. Once the number of languages to be selected from a particular family
is determined, the linguist chooses those languages for which the best
descriptive grammars (and / or other data sources) are available. Preferably,
the languages should come from different subgroups within their respective
families (Bakker, forthcoming).
In principle, the DVT composes so-called variety samples, which are
used to explore the range of diversity in a relatively little-studied linguistic
domain. Variety samples stand in contrast to probability samples, which
are used to determine significant correlations between grammatical traits
(Croft 2003: 23). In a sample of the latter type, the languages represent
independently selected cases. Thus, this type of sample explicitly avoids
the situation in which grammatical traits are shared between languages as
a result of descent from a common ancestor or through language contact.
Samples that are selected without controlling for these factors are said to be
genetically and / or areally biased.
Although the DVT thus aims primarily at maximal genetic diversity,
it can also be used for the composition of what Bakker (forthcoming) calls
pseudo-probability samples. These are relatively small variety samples with a
relatively high degree of independence between the selected cases. They can
be composed by combining the DVT (applied to an up-to-date language
classification) with an areal classification as a stratifying dimension. The
rationale behind the use of small variety samples as pseudo-probability
samples is that languages that belong to different families vary along certain
structural parameters. Therefore, in the initial stage of the sampling process,
the goal of covering maximal diversity (relevant for a variety sample) and
the goal of avoiding genetic bias (relevant for a probability sample) can be
reached by the same procedure: picking one language per family. Combining
this method with an areal stratification allows one to avoid the second major
source of bias for a probability sample: feature sharing as a result of language
contact. Only when variety samples are expanded in order to cover a wider
range of variation (which may be attested in specific genetic or areal groupings)
do they become fundamentally ill-suited for probabilistic research, since the
languages can no longer be regarded as independent cases 61.
Hengeveld et al. (2004) use a pseudo-probability sample for their study
on the relation between PoS systems and word order constraints. This
50-language sample is composed by applying the DVT to Ruhlen’s (1991)
classification and then choosing languages that are spoken in non-contiguous
areas, at least whenever the bibliographical situation would permit it.

Notably, Ruhlen’s classification is controversial to the extent that it
distinguishes a relatively small number of (large) language families: 19
families and a number of isolates 62. This number is much smaller than for
instance in the Ethnologue classification, which distinguishes 120 families
(Gordon 2005). As mentioned above however, the DVT requires a minimum
of one language per family (see Rijkhoff & Bakker 1998). In view of this
requirement, Ruhlen’s classification provides a suitable basis for composing
a relatively small language sample, which is desirable in the case of a pseudo-
probability sample, and also in practical terms of feasibility. The Ethnologue
classification, in contrast, yields a sample of minimally 120 languages. Note
also that the specific methodology of the DVT partly makes up for the more
extensive ‘lumping’ in Ruhlen’s classification by dictating the selection of
more languages from families with larger internal diversity.
The sample used in the present study is very similar to the sample
used by Hengeveld et al. (2004). The first reason for this is a practical one:
It allowed me to take as a point of departure for most languages the PoS
system classifications as proposed by Hengeveld et al. (2004). Nevertheless,
I have adapted my sample in some respects. First, I have replaced a few
languages from the sample of Hengeveld et al. for bibliographical reasons.
The relevant cases are listed below:
• Hdi instead of Gude (Chadic, Afro-Asiatic)

• Dhaasanac instead of Oromo (Cushitic, Afro-Asiatic)
61
Maslova and Nikitina (submitted) have recently challenged the assumption behind
‘classical’ probability sampling, namely that language family size is a ‘historical accident’, and
that therefore every family should be represented in the sample with an equal number of
languages, in order to avoid disproportial representation of certain language types. They argue
that family size is determined by transition probabilities, which are in turn determined by
language constants, i.e. universal properties of languages and their users. Therefore, they argue
for the use of ‘truly’ random samples (so-called R-samples), rather than samples containing
one randomly selected language per family (so-called I-samples). R-samples will contain a
higher percentage of more stable language types. On the other hand, frequency distributions
in R-samples may be due to an accidentally higher initial frequency of some source type,
rather than to a systematically higher amount of transitions to this type. Therefore,
R-samples and I-samples should be compared in order to find out whether the distribution
attested in the modern language population is indeed the outcome of differences in transition
probabilities. If yes, then it constitutes evidence for a statistical universal. If not, then the
comparison can still indicate tendencies towards a linguistically meaningful equilibrium: If
a more stable trait occurs less frequently synchronically, then there will be a gradual increase
of languages with this trait. Conversely, if a less stable trait occurs more frequently, then the
number of languages with this trait will decrease over time.
62
In the second edition of Ruhlen, Korean-Japanese-Ainu and Kartvelian are distinguished
as separate families, while in the first edition they were considered sub-branches of Altaic and
Caucasian, respectively (Hengeveld et al. 2004: 528, note 4).
Chapter 4 – Sampling, Research Questions, Method | 111

• Lavukaleve instead of Nasioi (East Papuan, Indo-Pacific)
• Abun instead of Tidore (West Papuan, Indo-Pacific)
• Ma’di instead of Ngiti (Central Sudanic, Nilo-Saharan)
• Slave instead of Navaho (Na-Dene)
• Gooniyandi instead of Ngalakan (non-Pama-Nyungan (Bunaban),
Australian)
• Kharia instead of Mundari (Munda, Austro-Asiatic, Austric)
The second difference between my sample and that of Hengeveld et al.

(2004) is that I have expanded mine with two languages with a flexible
PoS system, since the most interesting data were obtained for this type of
languages. The added languages are:
• S antali, which comes from the same sub-family as Kharia: Munda,

Austro-Asiatic, Austric;
• Kambera, which belongs to the same family as Samoan and Tagalog:
Malayo-Polynesian, Austronesian.
Obviously, adding these languages results in areal and genetic biasing of the
sample 63. For this reason, Santali and Kambera are not taken into account
when calculating statistical dependencies between grammatical traits (see
Chapter 7). Rather, they are used as a further back-up to the patterns found
in the other flexible languages of the sample 64.
Finally, two further details should be noted. First, the present study
does not include any data on the extinct languages that are part of the
sample used by Hengeveld et al. (2004), namely Etruscan, Hurrian, Hittite,
Nahali, and Sumerian. Furthermore, in two cases I use a different language
name than Hengeveld et al. (2004): Bukiyip instead of Mountain Arapesh,
and Hmong Njua instead of Miao.
63
As Hengeveld et al. (2004: 529) note, the inclusion of Tagalog in their sample (and also in
mine) represents a violation of the genetic criterion, since it does not belong to the Formosan
branch of the Austric family, which would have been the right choice. It was included in
Hengeveld et al. in order to represent a language with a maximally flexible PoS system. Even
though this choice is relevant in view of the present study as well, it undermines the value of
the sample as a pseudo-probability sample, since it involves deliberately choosing a case (i.e. a
language) on the basis of the grammatical phenomenon being studied. I thank Elena Maslova
for emphasizing this point.
64
With the same goal in mind, some data from some other flexible languages have been and
will be discussed, including Tongan, Maori, and Mundari (see Chapters 2 and 5).

The total number of languages in the sample is 50, as can be seen in Table
4.1 below. In Appendix i, a more complete specification of the sample is
provided, in which the languages are also classified for family and subfamily
according to the Ethnologue, and for Genus according to the World Atlas
of Language Structures (WALS) 65.
As this section makes clear, a language sample is composed with a
specific typological research question in mind. Variety samples are in
principle not appropriate to answer probabilistic research questions, since
they are specifically designed to manipulate chances of occurrence: They
intend to maximize the likelihood of capturing all the linguistic diversity for
the phenomenon under study (Croft 2003: 21). However, small, genetically
and areally balanced variety samples, of the type used in Hengeveld et al.
(2004) and in the present study, can indeed be used as pseudo-probability
samples, on which to test hypotheses about dependency relations between
structural traits, in this case between various parameter values related to PoS
and DCs. Bearing this in mind, in the following section I turn to the specific
formulation and operationalization of the research questions of this study.
Family (Ruhlen 1991) Language(s)

Afro-Asiatic Hdi, Dhaasanac
Altaic Turkish
Korean-Japanese Japanese
Amerindian Pipil, Hixkaryana, Tuscarora, Koasati, Guaraní
(Paraguayan), Warao, Imbabura Quechua
Australian Kayardild, Nunggubuyu, Gooniyandi
Austric Tagalog, Samoan, Kambera, Paiwan, Garo, Thai,
Santali, Kharia, Hmong Njua
Caucasian Abkhaz
Kartvelian Georgian
Ckukchi-Kamchatkan Itelmen
Elamo-Dravidian Tamil
Eskimo-Aleut West Greenlandic
Indo-Hittite Polish
Indo-Pacific Wambon, Alamblak, Lavukaleve, Abun, Bukiyip
Isolates Ket, Burushaski, Basque, Nivkh
65
See www.ethnologue.org (or Gordon 2005) and http:/ / wals.info / (or Haspelmath et al.
2005).

Family (Ruhlen 1991) Language(s)
Khoisan Nama
Na-Dene Slave
Niger-Kordofanian Babungo, Kisi, Bambara, Krongo
Nilo-Saharan Lango, Ma’di
Pidgins and creoles Berbice Dutch Creole
Sino-Tibetan Nung, Mandarin Chinese
Uralic-Yukaghir Hungarian
Table 4.1: The language sample
4.3 Research Questions
4.3.1 Introduction
In the previous chapters, it was shown that PoS classes and DCs can be
defined as primary and secondary constructions, respectively, expressing the
propositional functions of predication, reference, and / or modification. On
the basis of this parallel, it can be hypothesized that there will be a match
between the distributional patterns of PoS classes and DC constructions in
any language. In particular, it is expected that the functional possibilities of
the (secondary) DCs constructions in a language are dependent on those of
the (primary) PoS categories available in that language. This hypothesis will
be investigated from two converging perspectives:
(i) The perspective of global matches, i.e. in terms of the presence of

functional flexibility versus rigidity in PoS systems as compared with
systems of DCs constructions (section 4.3.2);
(ii) The perspective of specific matches between particular types of flexible
and rigid PoS classes and DCs constructions, in terms of set of
functions that can be expressed by them (section 4.3.4).
In addition, and cross-cutting these two perspectives, I will investigate

the influence of another parameter, namely the internal morpho-syntactic
characteristics of the DC construction(s) under scrutiny (see sections 4.3.3
and 4.3.5). I will take this parameter into account by testing each of the
hypotheses to be formulated in the next subsections in two ways:
(i) Considering DCs as a single, undifferentiated construction type,

and

(ii) Splitting up the hypothesis into several sub-hypotheses, each of
which addresses a different structural DC type.
In what follows, a set of fully explicit hypotheses will be formulated and

operationalized according to the general research design just outlined.
4.3.2 Global functional matching

First, it is predicted that a global match exists between flexibility or rigidity as
attested in the PoS system of a particular language, and flexibility or rigidity in
the set of DC constructions in that language. This hypothesis is operationalized
in the form of the two-fold prediction A1 / A2, as given in (1a, b):
Predictions A1 / A2
(1) a. If a particular language has one or more flexible DCs, then it
should also have one or more flexible PoS classes.
b. If a language has rigid DCs only, then it should also have rigid
PoS classes only.
The phrases in italics in the above predictions require some explanation.

Recall from Chapter 2 that PoS systems are called flexible when they include
one or more flexible lexeme class(es). This means that a language with a
flexible PoS system does not need to have flexible PoS classes only. In fact,
Tagalog and Kharia are the only two languages in the sample that do not
have any rigid PoS classes. All other languages with flexible PoS systems
display a mixture of flexible and rigid PoS classes. In contrast, rigid PoS
systems were defined as consisting of rigid PoS classes only (even though
not all rigid systems have a rigid PoS class for every propositional function).
These asymmetric definitions also apply to the predictions in (1) above, as
the phrases in italics are meant to indicate: (1a) makes reference to languages
with flexible PoS systems, and (1b) to languages with rigid PoS systems.
4.3.3 Global matching including differentiation for structural DC type

The predictions in (1) can be made more specific by taking into account the
internal morpho-syntactic characteristics of the DC construction(s) under
study. In particular, this is done by adding a parameter that differentiates
between the three structural types of DCs defined in Chapter 3, section 3.4.
They are summarized once more in Table 4.2 (which is the same as Table 3.1):

number label 1 2
Type 1 B SENT / Ø SENT / Ø + +/–
Type 2 D-SENT SENT / Ø SENT / Ø – +/–
Considering the morpho-syntactic properties of the three DC types in

Table 4.2, we can say that:
(i) Type 1 DCs (Balanced) do not show any formal reflection of de-
categorization (i.e. TAM / Person is expressed as in independent
clauses), nor of re-categorization (arguments, if overt, are SENT, i.e.
expressed as in independent clauses) 66.
(ii) Type 2 DCs (D-SENT) show formal reflections of de-
categorization (i.e. TAM / Person is (partially) lost), but not of
re-categorization (arguments, if overt, are SENT, i.e. expressed as in
independent clauses).
(iii) Type 3 DCs (D-ALT) show formal reflections of both de-
categorization (i.e. TAM / Person is (partially) lost) and re-
categorization (at least one argument is expressed with an ALT
strategy, i.e. different from independent clauses).
A larger amount of formal de- / re-categorization of a DC construction

implies that it has less morpho-syntactic characteristics of a clause and
more of a lexical construction. It is hypothesized is that the more formally
similar a DC construction is to a lexical expression, the more functionally
similar the DC will be to its lexical counterpart. This means that deranked
DCs, i.e. type 2 (D-SENT) and 3 (D-ALT), are expected to show more
functional similarity with lexical categories than balanced DCs, i.e. type 1
(B). Furthermore, within the group of deranked DCs (types 2 and 3) it is
expected that the distributional patterns of more deranked DCs (type 3)
will be functionally more similar to lexical expressions than those of less
deranked DCs (type 2). Thus, we can formulate the two sub-predictions B1
and B2, as in (2a) and (2b):
66
Recall that determiners and / or case-markers are not regarded as features reflecting re-
categorization, because they do not affect the internal structure of the construction on which
they operate.

Prediction B1 / B2:
(2) a. The functional possibilities of deranked DCs (type 2 / 3) are
more similar to the functional possibilities of PoS than those of
balanced DCs (type 1).
b. Within the group of deranked DCs, the functional possibilities
of type 3 DCs are more similar to the functional possibilities of
PoS than those of type 2 DCs.
Note that, even though it has never been tested empirically on a larger
scale, the hypothesis of a functional connection between lexeme classes
and deranked DC constructions is not new. In fact, deranked complement
clauses (nominalizations / infinitives), relative clauses (participial clauses), and
adverbial clauses (converbal constructions) are traditionally characterized
as the syntactically derived clausal counterparts of nouns, adjectives and
adverbs, respectively (Croft 1991, Koptjevskaja-Tamm 1993, Haspelmath
1994, 1995). These connections between PoS and deranked DCs can be
summarized as in Table 4.3 (adapted from: Haspelmath 1995: 3-4) 67.
Forms Functions
Ref Head Ref Mod Pred Mod
PoS noun adjective (manner) adverb
deranked DC nominalization / infinitive participle converb
type construction construction construction
Table 4.3: Functional connection between PoS and deranked DCs
We can now combine predictions B1 and B2 in (2) above with predictions

A1 and A2 in (1) above. For flexible constructions, this yields the four sub-
hypotheses given in (3a-d):
(3) a. If a language has one or more flexible balanced DC(s) of type 1,
then it should also have one or more flexible PoS class(es).
b. If a language has one or more flexible deranked DC(s) of type
2 / 3, then it should also have one or more flexible PoS class(es).
67
Notably, the PoS classes and DCs defined in Chapters 2 and 3 make reference to 4
propositional functions: in Table 4.3 the function of head of a predicate phrase is not taken
into account. In Chapter 6 it will become clear that this functional slot is irrelevant for the
present study to the extent that there are no (or hardly any) DC constructions that express it.

c. If a language has one or more flexible deranked DC(s) of type 2,
d. If a language has one or more flexible deranked DC(s) of type 3,
On the basis of the structural differences between the DC types, it is

expected that the prediction in (3a), involving balanced clauses, is less likely
to receive empirical support than the one in (3b), concerning deranked
clauses. Further, the prediction in (3c), involving less deranked clauses of
type 2, is expected to be less probable than the one in (3d), with concerns
more deranked clauses of type 3.
Note that, due to the asymmetrical definitions of languages with flexible
versus rigid PoS systems, it does not make sense to formulate specific
predictions about implicational relations between the presence of rigid PoS
only and the presence of rigid DCs of a specific behavioural potential type only.
This is because, on the basis of the available typological evidence discussed
in Chapter 3, one would not predict that languages express all subordination
relations with rigid DCs of just a single structural type. Rather, semantically
different types of subordination relations are likely to be expressed by
morpho-syntactically different DC constructions. Therefore, no separate
sub-hypotheses are formulated that make reference to and rigid DCs of
type 1 / 2 / 3 only.
4.3.4 Specific functional matching

I now turn to the second perspective on the relation between PoS and DCs,
which involves the investigation of specific matches between the functional
possibilities of particular PoS classes and DC constructions. We have seen
in Chapters 2 and 3 that there are 10 predicted types of flexible constructions
(lexical and clausal), depending on how many and which propositional
functions they can express. These flexible construction types are listed once
more under (4):
(4) a. Lexical and clausal constructions that can be used in all four
propositional functions: contentives and contentive clauses;
b. Lexical and clausal constructions that can be used in all
propositional functions except the head of a predicate phrase:
non-verbs and multi-functional clauses;

c. Lexical and clausal constructions that can be used as the head
and modifier in a referential phrase: nominals and nominal clauses;
d. Lexical and clausal constructions that can be used as modifiers
in referential and predicate phrases: modifiers and modifier clauses;
e. Lexical and clausal constructions that can be used as the head and
modifier in a predicate phrase: predicatives and predicative clauses;
f. Lexical and clausal constructions that can be used as heads of
referential and predicate phrases: heads and head clauses;
g. Lexical and clausal constructions that can be used as heads of
referential and predicate phrases and as modifiers in referential
phrases: Flex PoS A and Flex clause A.
h. Lexical and clausal constructions that can be used as heads of
referential and predicate phrases and as modifiers in predicate
phrases: Flex PoS B and Flex clause B.
i. Lexical and clausal constructions that can be used as the head of
a referential phrase and as a modifier in a predicate phrase: Flex
PoS C and Flex clause C.
j. Lexical and clausal constructions that can be used as the head of
a predicate phrase and as a modifier in a referential phrase: Flex
PoS D and Flex clause D.
As regards rigid constructions, four construction types were predicted, each

of them specialized for the expression of a single propositional function.
They are listed in (5):
(5) a. Lexical and clausal constructions that are specialized for the
function of head of a predicate phrase: verbs and predicate clauses.
b. Lexical and clausal constructions that are specialized for the
function of head of a referential phrase: nouns and complement
clauses;
c. Lexical and clausal constructions that are specialized for the
function of modifier in a referential phrase: adjectives and
relative clauses;
d. Lexical and clausal constructions that are specialized for the
function of modifier in a predicate phrase: manner adverbs and
adverbial manner clauses.

It is hypothesized that each DC of a specific flexible / rigid type, being a
secondary construction, will have a PoS class of the same flexible / rigid type
as its primary counterpart. This hypothesis is operationalized in the form of
prediction C in (6):
Prediction C:
(6) If a language has a DC construction of a flexible or rigid type X,
then it should also have a PoS class of type X.
Applying Prediction C to each of the specific flexible constructions listed in

(4), we arrive at the set of predictions in (7):
(7) a. If a language has contentive clauses, then it should also have
lexical contentives.
b. If a language has multi-functional clauses, then it should also
have lexical non-verbs.
c. If a language has nominal clauses, then it should also have
lexical nominals.
d. If a language has modifier clauses, then it should also have
lexical modifiers.
e. If a language has predicative clauses, then it should also have
lexical predicatives.
f. If a language has head clauses, then it should also have lexical
heads.
g. If a language has Flex clause A, then it should also have
Flex PoS A.
h. If a language has Flex clause B, then it should also have
Flex PoS B.
i. If a language has Flex clause C, then it should also have
Flex PoS C.
j. If a language has Flex clause D, then it should also have
Flex PoS D.
A parallel list of predictions for the specific rigid constructions listed in (5)
appears in (8):
(8) a. If a language has predicate clauses, then it should also have
lexical verbs.

b. If a language has complement clauses, then it should also have
lexical nouns.
c. If a language has relative clauses, then it should also have lexical
adjectives.
d. If a language has adverbial manner clauses, then it should also
have lexical manner adverbs.
The predictions in (7) and (8) can be further fine-tuned by taking into account
the parameter of structural DC types. This is done in the next section.
4.3.5 Specific matching including differentiation for structural DC type

The predictions in (7) and (8) concerning specific flexible and rigid
construction types are now combined with predictions concerning different
structural DC types. Starting with maximally flexible constructions, i.e.
lexical contentives and contentive clauses, this yields the set of predictions
listed in (9a-d). The expected likelihood that each of the implicational
relations in (9) will actually hold, is the same as for the set of implications in
(3) above. The implication in (9a), concerning balanced DCs, is less likely to
receive empirical support than the one in (9b), concerning deranked DCs.
Further, the implication in (9c), which makes reference to less deranked
DCs of type 2, is less likely to hold than the one in (9d), involving more
deranked DCs of type 3.
(9) a. If a language has balanced contentive clauses of type 1, then it

should also have lexical contentives.
b. If a language has deranked contentive clauses of type 2 / 3, then
it should also have lexical contentives.
c. If a language has deranked contentive clauses of type 2, then it
d. If a language has deranked contentive clauses of type 3, then it
Analogous sets of predictions can be set up for the remaining nine types of
flexible constructions listed in (4) / (7) above, by means of substituting lexical
contentives for non-verbs, nominals, modifiers, etcetera, and substituting
contentive clauses for multi-functional clauses, nominal clauses, etcetera.

Similarly, the parameter of structural DC type is expected to interact with
the predictions concerning different types of rigid constructions: For each of
the rigid constructions listed in (5) / (8) a set of four sub-predictions can be
set up along the lines of (9) above. Taking as an example lexical and clausal
constructions specialized for the function of head of a referential phrase, i.e.
nouns and complement clauses, we get the set of implicational relations in
(10a-d). Again, the expectation is that the implication in (10a) is less likely
to be born out than the one in (10b), and that the implication in (10c) is less
probable than the one in (10d).
(10) a. If a language has balanced complement clauses of type 1, then it

should also have lexical nouns.
b. If a language has deranked complement clauses of type 2 / 3,
then it should also have lexical nouns.
c. If a language has deranked complement clauses of type 2, then
it should also have lexical nouns.
d. If a language has deranked complement clauses of type 3, it
Analogous lists of testable implications can be set up for the other types of
rigid constructions, by means of substituting nouns with verbs, adjectives,
or manner adverbs, and by substituting complement clauses with predicate
clauses, relative clauses, and adverbial manner clauses.
4.3.6 Summary
In sum, the general research question posed in this study is whether there is
a match between the functional possibilities of lexical constructions (PoS)
in a language and clausal constructions (DCs) in that language. This general
question is approached from two converging perspectives: First, global
matches are investigated between flexibility / rigidity as displayed by the PoS
systems and the dependent clause constructions of particular languages.
Second, specific matches are investigated between the functional patterns
of particular types of flexible and rigid PoS classes and DC constructions.
Across both perspectives cuts the additional parameter of structural DC
type, defined in terms of behavioural potential. The specific predictions
formulated above will be tested using a statistical method that is presented
in the following section.

4.4 Method
The predictions formulated in the previous section take the form of classic
Greenbergian implicational universals, namely: (With more than chance
frequency), if a language has structural characteristic A, then it has structural
characteristic B (Greenberg 1963; Cysouw 2005: 564). An implicational
universal can thus be construed as a hypothesis about a particular type of
dependency relation between A and B, such that one value of A (the positive
one) constrains the value of B, whereas the other value of A (the negative
one) does not (Maslova 2003: 103). This means that, in order to claim an
implicational universal, two methodological steps are required, as described
in (11a-b):
(11) a. Establishing a dependency relation between A and B, i.e. a

statistically significant correlation between the co-occurrence in
languages of characteristic A and characteristic B.
b. Establishing an asymmetrical dependency relation between
A and B, such that the distribution of characteristic B among
languages with characteristic A is more strongly skewed than
the distribution of B among languages without A.
Consider first (11a): Establishing a dependency between A and B requires

rejection of the hypothesis of independence, which states that the probability of
any combination of values of A and B is just the product of the probabilities
of these values taken in isolation. Statistical tests can be used to reject this
hypothesis (Maslova 2003: 102).
In the present study, each of the predictions formulated in the previous
section involves 2 binary parameters, one concerning PoS, the other
concerning DCs. Each parameter can have two values: Either the particular
type of PoS or DC is attested in the language ([+PoS], [+DC]), or it is not
attested ([-PoS], [-DC]). For every prediction this yields a 2x2 contingency
table with four cells. As is shown in Table 4.4 below, each of the cells in such
a table represents one of the four possible combinations of the two binary
parameters under investigation:
(i) [+PoS, + DC] = both the PoS and the DC feature are attested;
(ii) [+ PoS, - DC] = the PoS feature is attested, but the DC feature is
not attested;

(iii) [- PoS, + DC] = the PoS feature is not attested, but the DC feature
is attested;
(iv) [- PoS, - DC] = neither the PoS feature nor the DC feature is
attested.
DCs
+ –
PoS + (i) (ii)
– (iii) (iv)
Table 4.4: A 2x2 contingency table
The observed frequencies, i.e. the numbers of languages in the sample that
display a particular feature combination (as mentioned in (i)-(iv) above),
can be compared with the frequencies that would be expected if the co-
occurrence of the PoS and the DC features would be purely coincidental.
If this is not the case, i.e. if there is indeed a dependency relation between
a particular PoS feature and a particular DC feature, then the observed
frequencies in cells (i) and (iv) will be higher than the frequencies that are
expected on the basis of chance, whereas the observed frequencies in cells
(ii) and (iii) will be lower than the expected frequencies. A Fischer’s Exact
test can be used to calculate whether the deviation between observed and
expected frequencies is statistically significant. This test produces a p-value
between 0 and 1, which specifies how likely it is for the observed distribution
to be the result of chance (Cysouw 2003: 91). The critical value used to
identify a statistically significant correlation between two grammatical traits
is p<0.05, meaning that there is a 0.5% chance that the observed frequencies
are coincidental.
As Maslova (2003: 102) notes, if Fischer’s Exact fails to reject the
hypothesis of independence, i.e. if no significant correlation between the
parameters can be established, this does not automatically mean that they
really are independent: it can also be the case that the sample is too small to
reveal the dependency. Following the guidelines of Cohen (1995), a sample
size of 26 is needed to have an 80% chance of detecting a large effect with
a test like Fisher’s Exact. For the detection of a medium effect one needs a
sample of 84, while small effects are detectable only with a sample as large
as 785 cases. This means that the sample used in the present study suffices
for the detection of medium to large effects. However, in the case of small
effects, the sample will be too small for Fischer’s Exact to be able to reject
the hypothesis of independence.

Since Fisher’s Exact is thus highly sensitive to sample size, I will report, in
those cases where Fisher’s Exact is statistically significant, the contingency
coefficient (CC). This CC is a value between 0 and 1, which estimates the
effect size independent of the sample size. A CC value of approximately
0.10 indicates a small effect, around 0.30 is a medium effect, and from 0.45
upwards is a large effect (Everitt 1977).
As mentioned in (11b) above, if Fischer’s Exact yields a p-value
below 0.05, then a second step must be taken, namely the identification
of an asymmetrical dependency relation, such that the positive value of one
parameter (in this case the PoS parameter) constrains the value of the other
one (in this case the DC parameter). Maslova (2003) proposes a statistical
method to establish such an asymmetrical dependency relation 68. The basic
idea is to correlate each of the two parameters, in this case the PoS parameter
and the DC parameter, to a third, derived parameter: PoS = DC, which
refers to the event of the PoS parameter and the DC parameter having the
same value. This third parameter thus contrasts languages that confirm the
correlation, i.e. languages in which PoS and DC have the same value (PoS =
DC: either [+ PoS, + DC] or [- PoS, -DC]), with languages that disconfirm
the correlation, i.e. languages in which PoS and DC do not have the same
value (PoS ≠ DC: either [+ PoS, - DC] or [- PoS, + DC]). This involves two
extra contingency tables, as illustrated with Tables 4.5a and 4.5b:
PoS + PoS –
PoS = DC
PoS ≠ DC
Table 4.5a: Maslova test no. 1: Change DC parameter to PoS=DC
DC + DC –
PoS = DC
PoS ≠ DC
Table 4.5b: Maslova test no. 2: Change PoS parameter to PoS=DC
Submitting the observed frequencies in two extra tables of this kind to

Fischer’s Exact tests may yield three possible results, listed in (12a-c)
68
See also Cysouw (2005) for a brief assessment of this method.

(12) a. A symmetrical dependency: None of the tests shows a significant
interaction, i.e. in both cases p > 0.05.
b. A one-sided asymmetrical dependency: One of the two tests
shows a significant interaction, but not the other, i.e. in one case
p < 0.05, and in the other p > 0.05.
c. A two-sided asymmetrical dependency: Both tests show a
significant interaction, i.e. in both cases p < 0.05.
An example of (12a), a symmetrical dependency, is given in Tables 4.6 and

4.7a-b below. The original distribution appears in Table 4.6 69. Applying
Fischer’s Exact to these frequencies yields a significant correlation, i.e. there
is a dependency relation between the PoS and DC parameters. The two
additional tables are 4.7a and 4.7b, both of which yield a non-significant
p-value. This means that neither the positive not the negative value of the
PoS or the DC parameter imposes a constraint on the event of the two
parameters having the same value.
Table 4.6: A symmetrical dependency; original distribution

DC + DC –
PoS + 12 40
PoS – 60 30
p < 0.05 (significant)
Table 4.7a: Distribution for PoS and PoS = DC (Maslova-test no. 1)

Pos + Pos –
PoS = DC 12 30
PoS ≠ DC 40 60
p < 0.05 (not significant)
Table 4.7b: Distribution for DC and PoS = DC (Maslova-test no. 2)

Pos + Pos –
PoS = DC 12 40
PoS ≠ DC 60 40
69
These ‘observed’ frequencies are just illustrations. They are taken from examples in Maslova
(2003:105-106).

Second, Tables 4.8 and 4.9a-b illustrate the situation in (12b) above, i.e. a one-
sided asymmetrical dependency. Table 4.8 shows the original distribution,
which again yields a significant p-value, so that a dependency relation is
established. Tables 4.9a and 4.9b represent the two additional tests. Table
4.9a yields a significant p-value, while 4.9b does not. More specifically,
Table 4.9a shows that the positive value of the PoS parameter poses a
strong constraint on the event of PoS and DC having the same value. In
contrast, Table 4.8 makes clear that the event of (non-)attestation of the DC
characteristic and the event of the PoS and the DC parameters having the
same value are independent of each other.
Table 4.8: A one-sided asymmetrical dependency; original distribution

DC + DC –
PoS + 90 10
PoS – 35 28

Pos + Pos –
PoS = DC 90 28
PoS ≠ DC 10 35

DC + DC –
PoS = DC 90 28
PoS ≠ DC 35 10
This type of result may be interpreted as a statistical basis to formulate

a weak unidirectional implicational universal (Maslova 2003: 106; Dryer
2003: 111) of the following form:

[+ PoS] 2 [+ DC]
Languages with a positive PoS value significantly more often have a
positive DC value (regardless of whether languages with a negative PoS
value also tend to have a positive DC value).
Finally, Tables 4.10 and 4.11a-b illustrate the third type of result, a two-
sided asymmetrical dependency (see (12c) above). The original distribution
appears in Table 4.10, which again reveals a significant dependency relation.
Moreover, the two extra tests, applied to the figures in Tables 4.11a and 4.11b,
both yield a significant p-value.
Table 4.10: A two-sided asymmetrical dependency; original distribution

DC + DC –
PoS + 40 6
PoS – 50 60

Pos + Pos –
PoS = DC 40 60
PoS ≠ DC 6 50

DC + DC –
PoS = DC 40 60
PoS ≠ DC 50 6
As in the previous case, the positive value of the PoS parameter poses a
constraint on the event of PoS and DC having the same value. The significant
correlation in Table 4.11a reflects this. In addition, the original distribution
in Table 4.10 shows that the distribution of PoS for the negative value of the
DC parameter is clearly more skewed than for the positive value. In other
words, it is not only the case that [+PoS] statistically implies [+ DC],

but also that [-DC] implies [-PoS]. The latter dependency is shown by the
significant correlation in Table 4.11b.
This type of result may be interpreted as a statistical basis to formulate
a strong unidirectional implicational universal (Maslova 2003: 106; Dryer
2003: 111) of the following form:
[+ PoS] 2 [+ DC] and [-DC] 2 [-PoS]

Languages with a positive PoS value significantly more often have a
positive DC value, and the tendency for languages with a positive PoS
value to have a positive DC value is significantly stronger than the
tendency for languages with a negative PoS value to have a positive
DC value.
In sum, the method used in this study involves observing frequencies of co-
occurrence of formal linguistic traits. These frequencies are used to detect
different types of dependency relations between various pairs of formal
parameters, related to PoS classes and DC constructions, respectively, in
terms of the hypotheses formulated in the previous section.
4.5 Outlook
This chapter rounds off the first, theoretical part of the book. In the second
part, the data will be presented. Chapter 5 presents the classification of PoS
classes and PoS systems attested in the languages of the sample. Chapter
6 presents the typology of DC constructions in the sample languages.
In Chapter 7, these two typological data sets are combined to test the
hypotheses formulated in the present chapter. Chapter 8 provides a more
descriptive discussion of the results obtained in Chapter 7, explaining them
from a functionalist perspective. Finally, Chapter 9 presents the study’s
overall conclusions.

Parts of Speech in the Languages
of the Sample
5.1 Introduction
In this chapter I present the classification of parts of speech (PoS) classes
attested in the sample languages, following the definitions developed in
Chapter 2. In particular, for every language in the sample I determine the set
of PoS classes it displays, and define which propositional functions each PoS
class can express without extra structural coding. These data are presented
in section 5.2. Subsequently, in section 5.3, I evaluate the attested PoS classes
and PoS systems in terms of the predictions of the implicational map model
of parts of speech, as presented in Chapter 2.
The remainder of the chapter addresses a variety of issues regarding lexical
classification in the sample languages. These issues have already been touched
upon in the theoretical discussion on PoS typology in Chapter 2, but their
practical implications are now further illustrated with actual language data.
First, in section 5.4, I discuss the generality problem and the subclass problem. I
show how the generality problem is circumvented by disregarding behavioural
potential as a criterion for PoS classes. In addition, I discuss the problem of
fuzzy boundaries between PoS classes. I first illustrate cases of lexeme classes
with variable distributional patterns. Second, I consider different types of
‘restricted’ lexeme classes, i.e. small, closed classes and classes consisting
of derived lexemes. Section 5.5 discusses the application of the typology of
non-verbal predication to the sample languages. Section 5.6 focuses on the
identification of flexible PoS classes. Finally, section 5.7 is a brief conclusion.
131
5.2 Overview of the language data
Table 5.1 below shows the PoS classes that are attested in the languages of
the sample. The languages are listed in the leftmost column; they are ordered
from maximally flexible, via less flexible, to fully rigid. Within the group of
languages with rigid PoS systems, those with four large, open lexeme classes
precede the ones with (an increasing amount of ) restricted and / or lacking
PoS classes. Every PoS class appears in a separate row. The terms for different
types of PoS are abbreviated as follows: C for contentives, non-V for non-
verbs, Pred for predicatives, Nom for nominals, V for verbs, N for nouns, Adj
for adjectives and mAdv for manner adverbs 70. These abbreviations appear
in every column corresponding to a function that can be expressed without
extra structural coding by the PoS class in question 71. The names of the four
propositional functions appear in the top row of the Table: Pred Head for
head of a predicate phrase, Ref Head for head of a referential phrase, Ref
Mod for modifier in a referential phrase, and Pred Mod for modifier in a
predicate phrase. The two rightmost columns are used to add information
about the size and status of a PoS class: When it is a small, closed class, an
S appears in the penultimate column (with the heading Small). When a PoS
class consists of derived members only, a D appear in the rightmost column
(with the heading Derived). All other, unmarked cases involve large, open
PoS classes.
Table 5.1: PoS classes in the sample languages

Language Pred Ref Ref Pred Small Derived
Head Head Mod Mod
Tagalog C C C C
Kharia C C C C
Kambera C C C C
mAdv S
V D
Samoan C C C C
mAdv S
V D
70
Some of the PoS classes predicted in Chapter 2 do not appear in Table 5.1. This is discussed
further in section 5.2.
71
If a non-verbal PoS class has an additional predicative use (with a zero-1 expression
strategy, see Chapter 2, section 2.3.1), then the abbreviation of the particular PoS class appears
also in the column for Head of a Predicate Phrase).

Head Head Mod Mod
Guaraní C C C C
V
Santali C C C C
V
Warao V
non-V non-V non-V
N D
Turkish V
non-V non-V non-V
Mod Mod D
Kayardild Pred Pred
non-V non-V non-V
Adj
mAdv S
Paiwan Pred Pred
N N
Adj Adj S
I. Quechua V
Nom Nom
mAdv S
Ma’di V
Nom Nom
Mod Mod S
mAdv S
mAdv D
Gooniyandi V
Nom Nom
mAdv
Hungarian V
Nom Nom
mAdv
mAdv D
Japanese V
N
Nom Nom
mAdv
Hmong Njua V
N
Mod Mod Mod
Chapter 5 – Parts of Speech in the Languages of the Sample | 133

Head Head Mod Mod
Lango V
N
Mod Mod
mAdv
Ket V
N
Mod Mod
Adj S
Adj D
mAdv S
Koasati V
N
Mod Mod D
Adj S
mAdv D
mAdv S
Itelmen V
N
Mod Mod S
Adj D
mAdv D
Thai V
N
Adj Adj
Mod Mod Mod S
mAdv mAdv D
Basque V
N
Adj
Mod Mod S
mAdv D
Abun V
N
Adj Adj
mAdv
Mod Mod Mod S
Bambara V
N
Adj S
mAdv

Head Head Mod Mod
Georgian V
N
Adj
mAdv
Bukiyip V
N
Adj
mAdv
mAdv S
Abkhaz V
N
Adj
mAdv S
mAdv D
Polish V
N
Adj
mAdv S
mAdv D
Burushaski V
N
Adj
mAdv S
mAdv D
Lavukaleve V
N
Adj
mAdv S
Alamblak V
N
Adj
mAdv S
Pipil V
N
Adj
Wambon V
N
Adj

Head Head Mod Mod
Dhaasanac V
N
Adj
Adj Adj S
Berbice D.C. V
N
Adj Adj
Babungo V
N
Adj S
Adj D
mAdv S
Nama V
N
Adj S
Adj D
mAdv D
Hdi V
N
Adj S
mAdv S
mAdv D
Mandarin C. V
N
Adj S
mAdv S
Tamil V
N
Adj S
mAdv S
Kisi V
N
Adj S
Adj D
Nung V
N
Adj Adj S
Garo V
N
mAdv D
mAdv S

Head Head Mod Mod
Krongo V
N
mAdv S
Hixkaryana V
N
mAdv S
Slave V
N
Nivkh V
N
W.Greenlandic V
N
Nunggubuyu V
N
Tuscarora V
N (S)
5.3 Results in relation to the implicational map of parts of

speech
5.3.1 Introduction
In Chapter 2, three hierarchical relations were proposed as constraints on
cross-linguistic variation in PoS classification: A primary relation predication
reference, a secondary relation head modifier, and a ranking between
these two: ((Predication / Reference) (Head / Modifier)). These dominance
relations were combined into an implicational map model of parts of
speech. On the basis of this model, a number of possible PoS systems were
predicted. In addition, it was hypothesized that, as long as no constraints are
violated, in individual languages the predicted PoS systems can be mixed or
supplemented with additional classes.
In this section, I evaluate the data from the sample languages, as
presented in Table 5.1 above, in terms of the predictions of the implicational
PoS map. I start out with a presentation of the languages that exhibit one of
the predicted PoS systems in its ‘pure’ form. Second, I turn to languages with
mixed or ‘intermediate’ PoS systems. Third, I discuss cases of PoS systems
that were predicted in Chapter 2 but are not attested in the sample. There
are also some potential examples of the reverse situation: PoS systems that

were excluded in Chapter 2 on the basis of constraint violation(s), but that
are nonetheless attested in the sample.
5.3.2 Predicted and attested ‘pure’ systems

In this section I present the languages that exhibit one of the PoS systems
predicted in Chapter 2, without any mixing or additional classes. I start with
flexible systems, followed by rigid systems, in the same order as in Table 5.1.
Tagalog and Kharia both have contentives as their only PoS class 72. This
type of system is represented in (1) (cf. (14) in Chapter 2):
(1)
Head Modifier
Predication Contentives
Reference
Hmong Njua has verbs, nouns, and a class of modifiers that can also be used
predicatively 73. This system is represented in (2) (cf. (16) in Chapter 2):
(2)
Head Modifier
Predication Verb Modifiers
Reference Noun
Gooniyandi and Hungarian have a PoS system with verbs, nominals, and
manner adverbs, as in (3) (cf. (21) in Chapter 2). Note that Hungarian has
both simple and derived manner adverbs.
(3)
Head Modifier
Predication Verb
Reference Nominal
72
In section 5.6 the analysis of flexible PoS systems is discussed in full detail.
73
In Table 5.1 this is indicated by the abbreviation Mod in the column for Pred Head.

Georgian has four large, open classes, each of them specialized for a single
function. This is represented in (4) (cf. (17) in Chapter 2):
(4)
Head Modifier
The PoS system represented in (5) (cf. (18) in Chapter 2) has verbs, nouns,
and adjectives, but no manner adverbs. It is attested in Pipil, Wambon,
Dhaasanac, and Berbice Dutch Creole. Note that these languages differ in
terms of the ability to use adjectives in predicative function with a zero-1
strategy (see Chapter 2, section 2.3.1 and the discussion in section 5.4 below):
In Pipil and Wambon adjectives cannot be used as verbs, while in Berbice
Dutch they can. Dhaasanac has both types: a large class of non-predicative
adjectives, and a small class of predicative ones. In Table 5.1, predicative
adjectives are recognisable by the fact that the abbreviation Adj appears in
the column for modifier in a referential phrase and in the column for head
of a predicate phrase. In contrast, classes of non-predicative adjectives are
indicated with Adj in the column for modifier in a referential phrase only
(cf. notes 71 and 73).
(5)
Head Modifier
Finally, there are languages with the PoS system represented in (6) (cf. (19) in
Chapter 2), with verbs and nouns, but no lexeme classes for the two modifier
functions. This system is attested in Slave, Nivkh, West Greenlandic,
Nunggubuyu, and Tuscarora.

(6)
Head Modifier
Reference Noun –
The PoS system of Tuscarora may alternatively be analyzed as intermediate

between the system in (6) above and the system represented in (7) below (cf.
(20) in Chapter 2).
(7)
Head Modifier
Reference – –
The reason for this is that Tuscarora (like other Iroquoian languages) often
uses verbal constructions where other languages would use nouns. Mithun
describes this phenomenon as follows:
“A striking feature of natural speech in Iroquoian languages […] is the

relative rarity of nouns. It can be attributed in part to noun incorporation, as
in ‘he suit-case-carried’. […] More important are the functions that verbs
can serve. […] They can be used as descriptive labels for entities (objects,
animals, people) and even proper names […].” (Mithun 2000: 412)
An example of such a descriptively used verb is given in (8):
Tuscarora (Mithun-Williams 1976: 30)

(8) Ra-hren-ahs
masc-cut-progr
‘he cuts’ 2 ‘surgeon’
Apparently however, this “relative rarity of nouns” in Tuscarora does not

extend to the extreme situation in which only verbs remain. I have tried to
capture this in Table 5.1 by adding a bracketed S to Tuscarora’s noun class.

Many other languages, besides Tuscarora, also display a PoS system which
combines features of more than one of the systems predicted in Chapter 2.
To these ‘intermediate’ systems I turn in the next section.
5.3.3 Predicted and attested ‘intermediate’ systems

First, there are two languages in the sample that combine a class of
contentives with a large, open class of verbs: Guaraní and Santali. Thus,
the PoS system of these languages can be analyzed as a combination of the
systems represented in (1) and (7) above. Other languages have contentives
in combination with a class of derived verbs. This holds for Kambera and
Samoan. In addition, these languages both display another extra class, namely
a small, closed class of manner adverbs. This is an unexpected pattern to the
extent it involves a head-modifier distinction (derived verbs versus adverbs)
without a full-blown predication-reference distinction (see constraint (13) of
Chapter 2). However, it seems that adverbs in Kambera and Samoan have
many characteristics of function words, as opposed to content words (see
also section 5.4.3.2).
Second, the system in (9) (cf. (15) in Chapter 2) is attested in Warao, but
in combination with a class of derived nouns.
(9)
Head Modifier
Predication Verb
Reference Non-verb
The PoS system of Warao may thus be represented as in Figure 5.1 below.
This case is discussed in more detail in section 5.6.3.1.
Head Modifier
Predication Verbs Non-verbs
Reference Derived nouns
Figure 5.1: The PoS system of Warao
There are two more languages with intermediate PoS systems involving
the system in (9). First, Ma’di combines a PoS system of verbs and non-
verbs with a system of verbs, nominals, and manner adverbs. The latter is
represented in (10) (cf. (21) in Chapter 2):

(10)
Head Modifier
Reference Nominal
Turkish, as discussed in Chapter 2 (see Figure 2.5), also has a PoS system
with verbs and non-verbs (see (9) above), but combines it with a class of
derived modifiers, i.e. lexemes with the distributional possibilties of non-
verbs minus the function of head of a referential phrase.
Imbabura Quechua has an intermediate PoS system that involves a
combination of the system in (10) above, and the one in (11) below (cf. (22) in
Chapter 2), since it has verbs, nominals, and a closed class of manner adverbs.
(11)
Head Modifier
Reference Nominal
Japanese combines the system in (10) above with the system in (12), which
has verbs, nouns, and manner adverbs (cf. (30) in Chapter 2) 74.
(12)
Head Modifier
Reference Noun –
Another type of intermediate PoS system involves a system with verbs,

nouns, and flexible modifiers, in combination with rigid classes of adjectives
and / or manner adverbs. This is attested quite often, namely in Lango, Ket,
Koasati, Itelmen, Thai, Abun, and Basque. There is variation in the size
and status of the flexible and rigid PoS classes that can express the modifier
functions in these PoS systems. Lango, for instance, has a large class of
74
Alternatively, the Japanese system can be described as the system in (10) supplemented with
a class of rigid nouns.The PoS system of Japanese is discussed in more detail in section 5.4.2.1.

flexible modifiers and a large class of rigid manner adverbs, but no rigid
adjectives. Thus, its PoS system can be analyzed as a combination of the
system in (12) above and the one in (13) below:
(13)
Head Modifier
Reference Noun
Ket also has a large class of modifiers, but in combination with small classes
of rigid adjectives and manner adverbs (cf. (4) above). In addition, Ket has
derived adjectives.
Another possibility is that the rigid adjective and / or adverb class(es)
are large, while the flexible modifier class is relatively small. In Abun, for
instance, the classes of adjectives and manner adverbs are larger than the
modifier class. Thai and Basque both have full-blown adjective classes,
combined with (very) restricted classes of modifiers and manner adverbs.
Finally, it may be the case that all three types of PoS classes, i.e. modifiers,
adjectives and adverbs, are small and / or derived. This is attested in Koasati
and Itelmen.
A number of other languages have a fully differentiated PoS system
with four rigid PoS classes, in which either the adjective class or the manner
adverb class or both are small and / or derived. The following languages have a
small and / or a derived class of manner adverbs: Abkhaz, Polish, Burushaski,
Lavukaleve, and Alamblak. This type of PoS system is represented in (14).
It may be regarded as a system intermediate between the pure systems
presented in (4) and (5) above.
(14)
Head Modifier
Predication Verb S / D Manner adverb
Bukiyip has two types of manner adverbs. One type takes verbal inflection,
but cannot appear independently as the head of a predicate phrase. Conrad
and Wogiga (1991) do not give information about the size of this class,

nor about its open or closed status. An example of this type of adverb is
given in (15) below. The second class is described as a small, closed class
of uninflected manner adverbs, consisting of some 13 items (Conrad &
Wogiga 1991: 41).
Bukiyip (Conrad & Wogiga 1991: 41)

(15) Awou w-a-gamu
3pl.fem 3pl.fem.subj-real-well
w-a-dúkemech
3pl.fem.subj-real-understand
‘The women understand well.’
There are also languages in which both adjectives and manner adverbs
constitute small and / or derived classes, as is represented in (16) below. This
is the case in Babungo, Nama, Hdi, Mandarin Chinese, and Tamil.
(16)
Head Modifier
Predication Verb S / D Manner adverb
Reference Noun S / D Adjective
Finally, there are languages in which either the adjective class is small / derived
and the manner adverb class is altogether missing, or in which the adjective
class is missing and the adverb class is small / derived. The former situation
is represented in (17). It is attested in Kisi and Nung. Kisi has both a small
and a derived class of adjectives, neither of which can be used predicatively.
Nung has only a small class of adjectives, which can also be used as verbs.
(17)
Head Modifier
Finally, the reverse situation – a restricted class of manner adverbs and no

adjectives – is represented in (18). This type of system is attested in Garo,
Krongo, and Hixkaryana.

(18)
Head Modifier
Predication Verb S / D manner adverb
Reference Noun –
5.3.4. Predicted but not attested systems

(and attested but not predicted systems)
When considering all the predicted PoS systems in Chapter 2, we find that
the following ones are not attested in any of the sample languages:
(i) PoS systems with flexible predicatives: see (19) and (20) below (and
cf. (23) and (24) in Chapter 2);
(ii) PoS system with flexible heads: see (21) below (and cf. (25) in
Chapter 2);
(iii) PoS systems with a class of flexible lexemes that can be used in all
functions except one of the modifier functions (Flex A and Flex B):
see (22) and (23) below (and cf. (26) and (27) in Chapter 2);
(iv) PoS systems that involve either one or both of the ‘cross-wise’
flexible classes Flex C and Flex D: see (24) and (25) below (and cf.
(28) and (29) in Chapter 2).
(19)
Head Modifier
Reference Nominal
(20)
Head Modifier
Reference – –

(21)
Head Modifier
Predication Head –
Reference –
(22)
Head Modifier
Predication –
Reference Flex A
(23)
Head Modifier
Predication
Reference Flex B –
(24)
Head Modifier
(25)
Head Modifier
Predication Flex D Flex C
Reference Flex C Flex D
I start out with the systems involving predicatives. First, the fact that the
system in (20) is not attested does not appear to be very surprising, since this
system has no lexical strategy to express any function within the referential
domain. Even though constraint (11) in Chapter 2 determines that the
predicative domain ranks higher than the reference domain, this constraint
seems to have more bearing on functional specialization (as opposed
to flexibility) of PoS classes, rather than predicting the presence versus

absence of a lexical strategy in the first place. Apparently, as the discussion
of Tuscarora in the previous section also suggests, languages may make
relatively extensive use of verbal as opposed to nominal constructions, but a
complete lack of lexical strategies to express any function in the referential
domain seems to be a marginal option, if it occurs at all.
In contrast, the system in (19) does have a lexical strategy for the
expression of referential functions: it has predicatives and nominals. In
fact, even though this system is not attested in its ‘pure’ form, there are
two languages in the sample with PoS systems that resemble it to some
extent. One of these languages is Paiwan. It seems that all lexemes in this
language that usually occur as predicative heads (i.e. ‘verbs’) can also be used
as predicate modifiers (‘manner adverbs’), and vice versa. There are only two
items that possibly qualify as lexical adverbs, because they cannot be used
predicatively (Egli 1990: 158, 313). Since only one of these items appears
to be a manner adverb, I do not make reference to a (small) class of manner
adverbs in Paiwan (see Table 5.1). Even though Paiwan thus has a class of
predicatives, its PoS system differs from the system in (19): Paiwan has rigid
nouns and a restricted class of rigid adjectives, whereas the system in (19) has
nominals. Thus, the PoS system of Paiwan can be analyzed as intermediate
between the systems represented in (26)* and (27)* below (cf. the systems in
(15)* and (16)* of Appendix ii). As the asterixes are meant to indicate, these
systems are both excluded by the implicational PoS map, in particular on the
basis of constraint (11a) in Chapter 2, which states that the availability of a
rigid class for heads of referential phrases (nouns) implies the availability of
a rigid class for heads of predicate phrases (verbs).
(26)*
Head Modifier
(27)*
Head Modifier
Reference Noun –

There is, however, an alternative analysis, which hinges on the fact that it
is not easy to distinguish in Paiwan between a predicative head modified
by a simple lexical manner expressions and a serial verb construction, since
the language has very little verbal morphology. Therefore, it is possible that
Paiwan predicatives are in fact rigid verbs (cf. Egli 1990: 312). If this is the
right analysis, then constraint (11a) of Chapter 2 is no longer violated.
The second language in the sample with predicatives is Kayardild.
According to Evans, in this language
“[V]erbals primarily denote actions and processes, but may also provide
adverbial type information about the manner in which these are carried
out. […] A few lexemes only permit the modifier function.” (Evans
1995: 86)
In fact, Evans lists only four lexemes that are restricted to the function of
modifier in a predicate phrase (i.e. rigid manner adverbs), and about these
he remarks that “it is possible that a bigger corpus would see even these used as
main verbs” (Evans 1995: 303-304). Thus, it seems that Kayardild indeed has
predicatives. However, Kayardild’s PoS system also differs in some respects
from the system in (19) above: First, Kayardild has a class of non-verbs,
rather than nominals. In addition, there is a class of rigid manner adverbs,
which are analyzed as a subclass of non-verbs, since, unlike predicatives,
they take nominal inflection (Evans 1995: 227-229). There are thus two
lexical strategies available to express the function of modifier in a predicate
phrase: a ‘verbal’ and ‘nominal’ one. Moreover, Kayardild has a reasonably
large class of rigid adjectives (about 100 items). This is unexpected, since
it involves a specialized class for modification in the referential phrase,
without a specialized class for the corresponding head function (i.e. without
rigid nouns, cf. constraint (12a) of Chapter 2). In Figure 5.2 the PoS system
of Kayardild is schematically represented:
Head Modifier
Predication Predicative Manner adverbs
Reference Non-verbs Adjectives
Figure 5.2: The PoS system of Kayardild

I now turn to the predicted but unattested PoS system in (21) above. Even
though this system does not occur in its pure form, there are some languages
in the sample with a PoS system consisting of rigid nouns and verbs in
combination with a restricted group of lexemes that can apparently be used
as both. The relevant languages are Slave, Nivkh, West Greenlandic, and
possibly Nunggubuyu. The question is, however, whether these apparent
‘heads’ are truly flexible lexemes, or whether they rather involve zero-
derivation / conversion or polysemy. In section 5.6, where I discuss lexical
flexibility in the sample languages, I will argue that the non-compositional
semantic interpretation of the relevant lexemes in verbal and nominal
function suggests that they do not represent cases of true flexibility.
The non-attestation of flexible heads may also be related to the fact that
the propositional function(s) of modification appear not to be relevant in all
languages. Rather, in some languages ‘modifiers’ take the form of either separate
predications or of appositional referential expressions. In such cases the head-
modifier parameter becomes irrrelevant, which means that a hypothetical
class of ‘heads’ would in fact be a class of ‘contentives’, in the sense of a lexeme
class whose members can be used in all (relevant) functions 75.
The issue of irrelevance of propositional function(s) probably plays a
similar role in explaining why the systems in (22) and (23) above are not
attested. It is quite possible that some of the languages classified as having
a single class of contentives, in fact have a PoS class with maximally flexible
lexemes, which can be used in the complete set of relevant functions; a set
that excludes one of the two modifier functions. In Tagalog, for instance, it
seems that the slot for modification in a predicate phrase is irrelevant (see
Hengeveld & Van Lier 2009).
Finally, the systems in (24) and (25) above seem intuitively rather
improbable, even though they cannot be excluded on the basis of the
constraints as formulated in Chapter 2. In particular, these systems involve
one or two flexible lexeme class(es), which combine the opposite values
of both the predication-reference and the head-modifier parameter. In
other words, the relevant flexible classes combine the head function of one
domain with the modifier function of the other domain. The markedness
75
Note that the possibility of irrelevant propositional function(s) in specific languages
suggests that the functional space as defined in Chapter 2 is not universal. This is in line
with the assumptions of the framework of Functional Discourse Grammar, which explicitly
distinguishes between language-specific pragmatic and semantic functions, which belong
to the grammar of a language, and universal conceptualisation, which belongs to general
cognition (Hengeveld & Mackenzie 2008).

of this situation, i.e. the neutralization of maximally divergent functional
distinctions, may explain the non-occurrence of the systems (24) and (25) in
the sample languages.
Of course, in these and all other cases discussed above, it is possible
that the non-attestation of a predicted PoS system is purely coincidental,
and that the relevant system will turn up in an investigation of a larger
sample. In addition, as I have suggested several times above, both the non-
attestation of predicted systems and the attestation of excluded systems may
be due to particular difficulties in the analyses of the available language data.
Finally, it may very well be that the formulation of the restrictions on PoS
systems in Chapter 2 is not yet optimal, and / or that the model should take
into account more functional parameters.
5.3.5 Summary
To summarize, in this section I have related the PoS systems attested in
the languages of the sample to the PoS systems predicted in Chapter 2.
It was shown that most predicted systems are actually attested, either in
their ‘pure’ form or in the form of an ‘intermediate’ system. I have suggested
possible explanations for the cases of predicted but unattested PoS systems.
Apart from the unexpected attestation of rigid adjective and noun classes
in the two languages with predicatives (Paiwan and Kayardild), none of
the systems that were excluded on the basis of the constraints in Chapter
2 (see Appendix ii) have been found in the sample languages. In general,
these results suggest that the implicational map model of PoS developed in
Chapter 2 has a reasonable typological adequacy.
5.4 The generality and the subclass problems in practice
5.4.1 The generality problem

As explained in Chapter 2, the Hengeveldian method for identifying
PoS classes crucially depends on whether or not extra structural coding
(specific function-indicating morpho-syntax) is required to use a member
of a particular lexeme class in a certain propositional function. In contrast,
behavioural potential (the expression of morpho-syntactic categories
belonging to a particular function) does not play a role in the definition of
PoS classes.
In the present study, the practical application of this method has in
some cases resulted in analyses of PoS systems that differ from the ones

proposed in the relevant descriptive sources. In all these cases, the difference
in analysis is of the same nature: I propose a flexible PoS class where the
reference grammar distinguishes two (or more) rigid word classes. The
reverse situation – in which a reference grammar claims flexibility, while
I propose multiple rigid PoS classes – does not occur. Obviously, this
‘lumping’ effect is due to the fact that structural coding is a rather restricted
defining criterion. Most reference grammars identify PoS classes using
the full set of language-specific distributional data, as advocated by Croft
(2001). However, as discussed in Chapter 2, the Croftian method raises
the generality problem: PoS classes identified in this way are not cross-
linguistically comparable. In what follows, I will give examples to show
how the Hengeveldian method gets around the generality problem. At
the same time, these examples illustrate how the results obtained with the
Hengeveldian method may differ from the results of a complete language-
specific distributional analysis.
First, consider Kambera 76. According to Klamer (1998: 91-95), Kambera
has verbs, nouns, and a class of multifunctional lexemes that can be used
as both. Nouns and verbs are distinguished by Klamer on the basis of the
following criteria:
(i) Nouns can occur with articles and with the marker bai / bi, which
expresses appreciation or derogation of (properties of ) the noun.
Noun phrases can be modified by emphatic or demonstrative
pronouns. Nouns also have specific quantifying properties. They
may occur, for instance, with a bare, un-derived numeral and may
have a classifier.
(ii) Verbs can be derived and they can be modified by adverbs (adverbs
do not modify nouns).
These criteria involve language-specific aspects of behavioural potential,

rather than structural coding. Therefore, following the Hengeveldian
approach, these facts are not regarded as evidence for a lexical distinction
between nouns and verb.
Moreover, there is explicit evidence in favour of a flexibility analysis.
First, Klamer (1998: 96) acknowledges the presence of “a considerable number
76
Recall that Kambera is one of the two languages added to the balanced sample (see
Chapter 4), and will therefore not be included in the analyses of Chapter 7.

of multifunctional lexical items” in Kambera. In addition, ‘nouns’ and ‘verbs’
have identical distributional possibilities in terms of the ability to be used
predicatively and referentially. This is illustrated in examples (28) and (29)
below. In example (28a), an action-denoting lexeme is used as the head of a
predicate phrase and takes the inflection belonging to this function, while
in (28b) the same lexeme is used as the head of a referential phrase and
combines with an article. In (29) two object-denoting lexemes are used as
predicative heads 77.
Kambera (Klamer 1998: 105, 107)

(28) a. Jàka manganga-ma-na-yna-i una, …
if steal-emph-3sg.gen-3sg.dat-iter emph.3sg
‘If he does steal again,…’
b. Na ma-kaloru-nya na manganga…
art rel.subj-arrange-3sg.dat art steal
‘Who is engaged in theft, …’
(29) Nina nda tustel-a-ya, senter-ya

deic-3sg neg camera-mod-3sg.acc torch-3sg.acc
‘This is not a camera, it’s a torch.’
Apart from expressing the two head functions, ‘verbs’ and ‘nouns’ in Kambera
can also be used, without structural coding, as modifiers in predicate phrases
and referential phrases. The examples in (30) show an action-denoting
lexeme (a) and an object-denoting lexeme (b) functioning as modifiers in
a referential phrase. Examples (31a) and (31b) show an action-denoting and
an object-denoting lexeme, respectively, in the function of modifier in a
predicate phrase.

(30) a. iyang wàu b. meu rumba
fish smell cat grass
‘smelly fish’ ‘wild cat’
77
These examples show that the semantic interpretation of the lexemes in either function
is fully compositional: an action-denoting lexeme in nominal function denotes ‘the act of
X-ing’, while an object-denoting lexeme in verbal function denotes ‘be X’. I return to this
issue in section 5.6.

(31) a. dedi meti-ma-a-nanya na ina-na
be.born die-emph-mod-3sg.cont art mother-3sg.gen
‘His mother died in labour’ (lit: ‘died while giving birth’)
b. jangga eti
be.tall liver
‘be arrogant’ (lit.: ‘be tall liver-wise’)
In short, these examples show that there is no difference between nouns

and verbs in Kambera, at least not in terms of the definitions developed in
Chapter 2.
Ma’di represents another case in which the Hengeveldian method results
in a different classification of a language’s PoS system than the one proposed in
the reference grammar. Blackings and Fabb (2003) distinguish between nouns
and adjectives in Ma’di, on the basis of the following distributional facts:
(i) Most adjectives are inflected for number (except for colour terms),
while most nouns are not;
(ii) Adjectives always combine with an article when they are used as the
head of a referential phrase, whereas nouns can also appear without one.
However, adjectives and nouns do not differ in terms of their possibility

to be used, without structural coding, as the head and the modifier of a
referential phrase. This is illustrated in (32) and (33). In the first example
a property-denoting lexeme is used as the head of a referential phrase; in
the second example an individual-denoting lexeme is used as a modifier in
a referential phrase. Therefore, I have classified Ma’di as a language with
flexible nominals.
Ma’di (Blackings & Fabb 2003: 106, 304)

(32) ālī rɨ pɨ ɛ¯-tʃá ádƷɨńɨ̄
short(pl) del pl.pron (3)-ve-arrive yesterday
‘The short ones arrived yesterday.’

(33) Màɗí ādrúpi rɨ ɔ¯ɗ ā rʊ̄ ɨ lɛ́-nɨ
person brother(indef) def act refl that-like
kʊ̄
neg(non-pst)
‘A person who is a brother won’t behave like that.’
Note in passing that a determiner is also obligatory in Ma’di when both a

modifier and a head are present in the referential phase. The ungrammatical
example in (34) shows this:
Ma’di (Blackings & Fabb 2003: 302)

(34) *básɨ ɨŋgwɛ¯ ɛ¯-tʃá rá
bus white (3)-ve-arrive aff
‘The / a white bus has certainly arrived.’
The case of Ma’di may be contrasted with a case like Hdi. Frajzyngier and
Shay (2002: 71) use the following two criteria to define the class of adjectives
in Hdi:
(i) Adjectives can modify a noun without any intervening marker

(whereas nouns cannot);
(ii) Adjectives cannot be used as arguments (whereas nouns can).
Example (35a) and the ungrammatical (35b) show that these two criteria
qualify the lexeme xɓ ùzá ‘big-bellied’ as an adjective:
Hdi (Frajzyngier & Shay 2002: 72)

(35) a. ndá n`gh-íyù tá gù xɓ ùzá
assoc see-1sg obj goat big-bellied
‘I saw a big-bellied goat.’
b. *ndá n`gh-íyù tá xɓ ùzá

assoc see-1sg obj big-bellied
‘I saw a big-bellied one.’
Thus, the distinction between nouns and adjectives, as proposed by

Frajzyngier and Shay (2002) is based on functional possibilities and

structural coding only. Therefore, their PoS classification is the same as the
one I arrive at using the Hengeveldian method.
In this section I have illustrated how the criterion of structural coding
allows for the establishment of cross-linguistically comparable PoS classes
and PoS systems. It should be clear that the down-side of this method is that
it may result in lumping together groups of lexemes that would be regarded
as separate classes when the full set of language-specific distributional facts
would be taken into account. This brings us to the related subclass problem,
which is discussed in the next subsection. Further on, in section 5.6, I return
in more detail to the issue of defining flexible PoS classes.
5.4.2 The sub-class problem

In this section I discuss a number of issues related to the sub-class problem.
First, as pointed out in Chapter 2, PoS classes in particular languages
may have fuzzy boundaries in terms of their distributional behaviour.
Such PoS classes display a combination of properties associated with two
other groups of lexemes and / or they display varying behaviour in a single
propositional function. These phenomena are illustrated in section 5.4.2.1.
A second problem concerns the boundary between major, open PoS classes
and various types of more restricted classes. In section 5.4.2.2 I consider
two types of such restricted classes: (i) small, closed lexeme classes, and (ii)
classes of derived lexemes.
5.4.2.1 Fuzzy boundaries

In Chapter 2, I discussed the distinction between nouns and adjectives
in Japanese. This case was adduced by Croft (2001) to illustrate that the
Hengeveldian approach to PoS definition cannot handle the phenomenon of
fuzzy boundaries between language-specific lexeme classes. Since Japanese
is one of the languages of my sample, I consider here once more the relevant
data, which are presented in Figure 5.3. The different groups of lexemes
identified in Croft’s analysis are indicated here with numerical codes (cf.
Figure 2.13 in Chapter 2).

PoS class Function / Coding
Head of Pred Phrase Head of Ref Phrase Mod in Ref Phrase
1 da (COP) Ø no (GEN)
1/2 da (COP) Ø no (GEN) / na (LK)
2 da (COP) Ø na (LK)
2/3 a da (COP) / Ø + Tense Ø na (LK) / Ø + Tense
2/3 b Ø + Tense Ø na (LK) / Ø + Tense
3 Ø + Tense X Ø + Tense
Figure 5.3: The noun-adjective distinction in Japanese (2)
Following the Hengeveldian approach, the data in Figure 5.3 result in the
distinction of three PoS classes in Japanese:
(i) A class of nouns (class number 1): lexemes that are used without
structural coding as the head of a referential phrase. These lexemes
need a copula when used in predicative function, and a genitive
marker no when used as modifier in a referential phrase;
(ii) A class of nominals (class number 2): lexemes that are used without
structural coding as the head of a referential phrase, and with the
linking element na (rather than no) when used as a modifier in a
referential phrase. In predicative function these lexemes need a copula.
(iii) A class of verbs (class number 3): lexemes that are used without
structural coding as the head of a predicate phrase, and that
cannot be used as the head of a referential phrase. When used as a
modifier in a referential phrase, these lexemes remain verbal (i.e. are
relativized): they inflect for tense and take neither the genitive nor
the linking element, as do nouns and nominals, respectively.
The remaining classes, i.e. the ones with the codes 1/2, 2/3, 2/3a and 2/3b in
Figure 5.2 above, are not taken into account as separate classes. This is reflected
in Table 5.2, which repeats the relevant data for Japanese from Table 5.1.
Language Pred Head Ref Head Ref Mod Pred Mod

Japanese V
N
Nom Nom
Table 5.2: Verbs, nouns and nominals in Japanese

In other languages of the sample I have come across comparable cases of
fuzziness. For instance, in Mandarin Chinese the existence of a separate
class of adjectives is still debated. In terms of the Hengeveldian approach
the following facts are relevant: There is a large group of property-denoting
lexemes (i.e. semantic ‘adjectives’) that can function as modifiers in referential
phrases only when they are structurally coded with the marker de, whereas
they can be the heads of predicate phrases without any structural coding.
The modifying construction is illustrated in (36) below. This construction
is exactly the same as the one used for action-denoting lexemes in modifier
position, i.e. relative clauses, as illustrated in example (37). For this reason,
semantic ‘adjectives’ are analyzed as (stative) verbs.
Mandarin Chinese (Li & Thompson 1981: 32, 581)

(36) [Hóng=hong de] hūa
red=red lk flower
‘flowers that are really red’
(37) [Jīntiān yíng de] qián

today win lk money
‘the money that we won today’
However, according to Paul (2005), there is also a relatively small group

of so-called “non-predicative adjectives”. When these property-denoting
lexemes modify a noun, they appear either with or without the linking
element de, as is shown in (38) below. In addition, these lexical items cannot
function as the head of a predicate phrase without structural coding; they
need the copula shi in combination with de, as is illustrated in (39).
Mandarin Chinese (Paul 2005: 760, 759)

(38) ta mai-le yi-ge fang (de) panzi
3sg buy-pfv 1-cl square lk plate
‘He bought a square plate.’
(39) Zhei-ge panzi shi fang de

this-cl plate cop square lk
‘This plate is square.’

On the basis of these distributional facts, I have analyzed Mandarin Chinese
as having (i) a class of verbs: lexemes that can only be used predicatively,
and (ii) a small class of adjectives: lexemes that can (at least sometimes)
function without structural coding as modifiers in referential phrases, and
that cannot be the heads of predicate phrases without a copula. The fact
that the adjective class had ‘fuzzy edges’, in the sense that its members show
variable behaviour in the modifier position, is not taken into account. This
analysis is represented in Table 5.3, which repeats the data for Mandarin
Chinese from Table 5.1:
Language Pred Head Ref Head Ref Mod Pred Mod Small
Mandarine Ch. Verb
Adj S
Table 5.3: Verbs and Adjectives in Mandarin Chinese
In sum, this section further illustrated the practical repercussions of the

restrictive structural coding criterion of the Hengeveldian method. In
particular, fuzzy boundaries between language-specific lexeme groups and
class-internal behavioural variability are ignored, in order to maintain the
cross-linguistic comparability of PoS classes.
5.4.3.2 Restricted classes
Small, closed classes
Lexemes and lexeme classes in languages can be divided (broadly) into

content words and function words. Generally, content word classes are open,
which means that new members can easily be added to them. Also, content
word classes are large; they comprise hundreds or thousands of items.
Function word classes, on the other hand, tend to be closed and small.
It can be difficult, however, to determine the boundary between these
two types of word classes in individual languages. Especially adjectives and / or
(manner) adverbs (or flexible modifier classes, for that matter) are in some
languages more like function words than like content words, in that they
constitute relatively restricted, closed classes (Haspelmath 2001: 16539-42).
The available descriptive sources do not always provide clear indications of
the size and status of adjective, adverb, and / or modifier classes. And even if
such information is available, definitions remain far from absolute.

I already briefly touched upon this point in Chapter 2, in connection with
the adjective class in Alamblak. A comparable case is attested in Abun.
According to Berry and Berry:
“The class of adjectives in Abun hovers on the borderline of being an

open or a closed word class (…). In the corpus a total of 38 adjectives
has been distinguished, but there may well be others, thus it is included
as an open class.” (Berry & Berry 1999: 36)
The number of manner adverbs in Abun is somewhat more limited; Berry

and Berry (1999: 25) identify 25 items. They state that:
“The subclass of manner adverbs is not as ‘open’ as a regular open class

because many adverbial meanings are expressed in other ways, such as
associative prepositional phrases or the use of adjectives without any
special marking.” (Berry & Berry 1999: 37)
Even though manner adverbs are thus not very numerous, Berry and Berry
include them as an open word class in Abun. In addition to adjectives and
adverbs, there is a group of about 12 lexical items in Abun that can function
as both (see also the second quotation above). This group is analyzed as a
small (and probably closed) class of flexible modifiers.
In some other languages even smaller classes are distinguished. Hdi, for
instance, has 8 lexical adjectives (Frajzyngier & Shay 2002: 71). Classes of
manner adverbs may likewise consist of (very) few members. For example,
Tamil is described as having only 3 manner adverbs: mella ‘softly’, molla
‘slowly’, and jalti ‘quickly’ (Asher 1985: 115).
In many other cases, the available descriptive sources do not give specific
numbers or lists of items that belong to classes of adjectives, adverbs, or
modifiers. Rather, such classes are characterized for instance as ‘restricted’
compared to other classes like nouns and verbs, which comprise much larger
amounts of items. Clearly, the bottom line of this section is that at least
some of the decisions to mark a PoS class as ‘small’ (S) in Table 5.1 involved
a certain degree of arbitrariness.

Derived classes
PoS classes that are marked as ‘derived’ (D) in Table 5.1 involve items that are
formed by means of productive derivational processes only 78. Furthermore,
derived PoS classes are mentioned separately only in cases where there
is either no categorically equivalent class of simple items, or where the
categorically equivalent simple items constitute a small, closed class.
For example, if a language has a large, open class of simple nouns and a
productive process to derive nouns, then the latter is not mentioned in the
data. In contrast, if a language has no class of simple adjectives, but does have
a productive process to derive adjectives, then a derived class of adjectives
is included in the data. Also, when a language has but a small, closed class
of adjectives, then the possibility to productively derive adjectives is also
mentioned separately. In the present section I give some examples of derived
PoS classes from the sample languages.
Consider once more Kambera. We already saw that, as far as un-
derived items are concerned, Kambera is analyzed as having contentives
and adverbs. However, the language also has two productive verbalization
processes: (i) pa- prefixation, which derives verbs with causative, permissive,
factitive, resultative, intensive, infinitive, or reciprocal interpretation; and (ii)
-ng suffixation, which forms applicative verbs. Some examples of these two
processes are given in (40) and (41):

(40) ànga ‘foolish / useless’ – pa-ànga ‘confuse, cheat on someone’
ana ‘child’ – pa-ana ‘have children’
dua ‘two’ – pa-dua ‘divide’
ndia ‘no’ (emphatic negator) – pa-ndia ‘deny’
(41) bungga ‘open X’ – bunggahu-ng ‘open X for Y’

riki ‘laugh’ – riki-ng ‘to laugh at / about X’
angu ‘friend’ – angu-ng ‘have X as a friend’
nàmu ‘towards the speaker’ – nàmu-ng ‘move towards the speaker’
78
The notion of ‘productivity’ is of course rather slippery. I have, as much as possible, relied
upon unambiguous statements in reference grammars in deciding whether or not a certain
derivational process is productive, i.e. applicable to all members of the input class(es).

In languages with rigid PoS systems, it is often the case that a class of
derived items exists alongside a small, closed class of underived items with
the same categorial value. Usually, the relevant lexeme classes are adjectives
and / or manner adverbs. Nama, for instance, has a small class of simple
adjectives and two types of productively derived adjectives, formed with the
suffixes -xà (‘attributive’) and -‘o (‘privative’), respectively (Hagman 1979:
60). However, classes of derived lexemes also occur without the presence of
any categorially equivalent class of simple items. Basque, for example, has
derived manner adverbs only. These are productively formed from adjectival
bases with the suffix -ki, as in sendo ‘strong’, sendo-ki ‘strongly’ (Hualde &
Ortiz de Urbina 2003: 193) 79.
A factor that may complicate the identification of classes of productively
derived lexemes is the influence of diachronic processes. First, it may be
the case that a certain proportion of the derived forms becomes lexicalized
and that their base forms no longer occur independently. Consider for
example Babungo: The large majority of the adjectives in this language is
derived from verbs. However, there are also a few adjectives for which no
corresponding verbal form exists. Historically, these forms probably did
exist, but synchronically it seems appropriate to say that Babungo has a
small class of simple adjectives, next to a class of derived adjectives.
The case of manner adverbs in Babungo is similar, except that the
process of lexicalization is apparently more advanced: Many adverbs take the
form of reduplications. Only a few of them, however, can still be identified
as derivations from verbs. For instance, the manner adverb bwāŋnə̄-bwāŋnə̄
‘gently’ is derived from the verb bwāŋnə́ ‘to be well, to be soft’. In contrast,
for the lexeme wèe-wȅe ‘slowly, gently’ no verbal base form is available
(Schaub 1985: 246). In accordance with cases of the second type, Babungo
is synchronically analyzed as having a restricted class of underived manner
adverbs. Furthermore, since the derivational process is no longer productive,
no class of derived manner adverbs is included for Babungo in Table 5.1.
A related potential problem concerns the distinction between lexical
and clausal derivation. In Tamil, for example, it may seem as if there is a
class of derived adjectives, formed from verbal and nominal bases by means
79
In one case, namely Hungarian, I have coded both a class of simple and a class of derived
manner adverbs. In fact, however, the status of these classes is not entirely clear. According
to Kenesei et al. (1998: 222), simple adverbs are ‘numerous’. However, according to Rounds
(2001: 180) most manner adverbs are derived, which seems to suggests that the class of
simple items is in fact relatively restricted.

of the suffixes -a and -aana / -uɭɭa, respectively. However, these derivations
are in fact all dependent clauses: The suffix -a operates on the phrase level,
and the argument(s) of the dependent predicate retain the same expression
as main clause arguments. This is shown in example (42):

(42) [Akkaa taŋkaccikki caata pooʈʈ -a] karanʈi
elder.sister younger.sister-dat rice put:pst-ptc spoon
‘The spoon with which elder sister gave rice to younger sister.’
Derivations from nominal bases with -aana / -uɭɭa are no different: These
suffixes are in fact dependent, participial forms of the verbs aaku ‘to become’
and uɭɭu ‘(existential) be’. This can be seen in example (43):
Tamil (Lehmann 2005: 13)

(43) [aȤak-aan-a] kaɳɳati
beauty-become.pst-ptc mirror
‘beautiful mirror’ (lit. ‘mirror that has become beautiful’)
In sum, what may look like a derived adjective in Tamil is in fact a relative
clause construction.
Before rounding off this section, it is worth mentioning that the notion of
derived PoS classes clashes in some sense with the criterion of structural coding,
since word-class changing derivation can be regarded as an instance of structural
coding in itself. Therefore, it may seem strange to say that the members of a
class of derived lexemes can be used without structural coding in a particular
function. On the other hand, derivation is also a means by which new group of
lexical stems is created, which is especially suited for the expression of a specific
function. This group of stems may in turn form the input of further processes
of morphological or syntactic derivation and structural coding. Moreover, in
the context of the present study, the crucial question is whether derived PoS
classes may serve as functional models for DC constructions, and there seems
to be no principled reason why they may not.
5.5 Non-verbal predication

As explained in Chapter 2, the function of head of a predicate phrase has
an exceptional status within Hengeveld’s PoS theory. This is because this
function is the only possible function of the PoS class defined as ‘verbs’,

while it can be an additional function of PoS classes other than verbs. In
Chapter 2 I discussed the possible strategies that languages may employ
to express non-verbal predication. These strategies were divided into those
with a copula and those without one. The strategies without a copula were
further subdivided into zero-1 and zero-2 strategies. A zero-1 strategy
means that the non-verbal predicate shows the same behavioural potential
as a verbal predicate in the language under analysis. A zero-2 strategy means
that the non-verbal predicate does not express any verbal categories and is
simply juxtaposed to its argument.
The use of a copula strategy counts as structural coding. Therefore, a
non-verbal PoS class that employs this strategy is analyzed as not including
the function of head of a predicate phrase in its range of functional
possibilities. The same holds for non-verbal PoS classes that use a zero-2
strategy: they are also analyzed as non-predicative, since they cannot express
the same categories as verbal predicates. In addition, zero-2 strategies often
occur in alternation with a copula strategy. In such cases, the copula does
not appear when there is no functional need to express verbal features like
TAM distinctions and / or Person. Typically this involves ‘default’ cases, such
as present tense and third person singular. The Hungarian examples (44) (in
present tense) and (45) (in past tense) illustrate the alternation of a zero-2
strategy and a copula strategy.
Hungarian (Kenesei et al. 1998: 62, 78)

(44) A lány diák
the girl student
‘The girl is tall / a student.’
(45) Péter diák volt.

Peter student cop(pst)
‘Peter was a student.’
In short, PoS classes that employ a copula or a zero-2 strategy when used as
the head of a predicate phrase are regarded as non-predicative. In contrast,
PoS classes that use the zero-1 strategy are analyzed as including the function
of head of a predicate phrase in their range of distributional possibilities.
However, several difficulties may arise when classifying the strategies
used for non-verbal predication in individual languages. First, it may be
the case that only part of the behavioural potential of verbal predicates

can be carried over to non-verbal predicates. This happens for instance in
Abkhaz and Hdi. Non-verbal predicates in Abkhaz can take verbal suffixes
that express present tense, past tense, and imperative mood. However, other
verbal suffixes can be added to dynamic stems only. A copula intervenes
when one of the latter suffixes must be expressed on a non-dynamic, non-
verbal predicate. In Hdi nouns do not need a copula when they are used in
predicative function and tense can be expressed on them. However, unlike
verbal predicates, nouns cannot be marked for aspect. The reason for this
discrepancy is that tense is marked by means of independent morphemes,
while aspect marking involves different stem forms of verbs, and these are
not available for nouns. In cases like Abkhaz and Hdi I have opted for the
strongest possible definition of the zero-1 strategy: a non-verbal predicate
is analyzed as employing a zero-1 strategy only if the full set of verbal
behavioural potential can be expressed on the non-verbal predicate.
Second, it can be problematic to apply this criterion for the zero-1
strategy to isolating languages. In these languages, which lack inflection by
definition, it is not possible to decide in favour of a zero-1 strategy (instead
of zero-2) on the basis of the expression of verbal inflection on the non-
verbal predicate. In such cases, there are two possible ways to distinguish
between a zero-1 and a zero-2 strategy:
(i) The possibility to add a copula;

(ii) The possibility to combine the non-verbal predicate with
freestanding particles that express verbal categories such as TAM.
For instance, Paiwan does not have a copula, and non-present tense can
be expressed with non-verbal predicates by means of the freestanding
particles na (for past) and urhi (for future), as is shown in (46) below. This
is interpreted as evidence that there is no difference between verbal and
non-verbal predication in terms of behavioural potential, i.e. that a zero-1
strategy is used in Paiwan.
Paiwan (Egli 1990: 61)

(46) Na / urhi sivitai ti kama
pst / fut soldier foc father
‘Father was / will be a soldier.’

In other isolating languages, such as Hmong Njua, Abun, Thai, and Nung,
nominal predicates do take a copula. Under certain circumstances this copula
can be omitted, which results in a zero-2 strategy. In contrast, adjectival
predicates never take a copula. In addition, there appear to be no principled
restrictions on the usage of grammatical particles with adjectival predicates. In
(47) I give an example from Thai, which shows the expression of imperfective
aspect with an adjectival predicate by means of the particle yùu. On the
basis of these data, Thai is analyzed as having predicative adjectives. Similar
evidence is available for the other isolating languages mentioned above.
Thai (Iwasaki & Ingkaphirom 2005: 155)

(47) aahǎan kamlaŋ rɔ́ɔn yùu
food adv hot ipfv
‘The food is still hot.’
It may be worthwhile at this point to stress the difference between (i) a

situation in which adjectives can be used predicatively with a zero-1 strategy,
and (ii) a situation in which there is no difference between adjectives and
verbs. While the latter case involves bi-directional flexibility, in the former
case there is only a unidirectional identity relation: all adjectives can be
used as verbs, but not all verbs can be used as adjectives without structural
coding 80. Rather, when a verb is used as a modifier in a referential phrase, it
must be relativized, as in example (48), again from Thai:

(48) Khon [thîi duulɛɛ] nî pen pen acaan lə̌
person rel take.care prt cop cop teacher q
‘Is the person who takes care (of the students) a teacher?’
Notably, the use of a verb in the function of modifier in a referential phrase

does not always require an overt relativizer in Thai. Subject relative clauses
that ascribe a general property to their head can also appear without the
relativizer, as shown in (49). Nonetheless, the availability of the relativizer
as a structural coding mechanism, even though it is not always needed, is
interpreted as evidence that there is no complete overlap between verbs and
adjectives in Thai.
80
This amounts to what Stassen (1997) terms “verby adjectives”.

(49) dèk [rian kèŋ]
child study well
‘a child who studies well.’
I will return to the issue of bi-directionality as a criterion for flexible PoS classes
in the course of the following section, which is concerned more generally with
the identification of lexical flexibility in the languages of the sample.
5.6 Identifying flexibility
5.6.1 Introduction
In Chapter 2, I discussed three criteria for lexical flexibility, as proposed by
Evans and Osada (2005): (i) the criterion of semantic compositionality, (ii) the
criterion of exhaustiveness, and (iii) the criterion of equivalent combinatorics.
In this section, I present evidence for flexible PoS classes in the languages
of the sample. As far as the data allow it, I organize the discussion along
the lines of Evans and Osada’s criteria. In section 5.6.2 I apply these criteria
to languages that appear to display the most radical type of flexibility: a
class of contentives that can be used in all propositional functons. In section
5.6.3 I focus on languages that display less flexible lexeme classes, i.e. classes
that can express the functions of head and modifier in referential phrases
(nominals), modifier in predicate and referential phrases (modifiers), or all
three functions (non-verbs).
5.6.2 Languages with contentives

In some languages contentives are the only available lexeme class. This is
the case is Tagalog and Kharia. Other languages combine a large class of
contentives with one or more rigid PoS classes, which can but need not be
small or derived. This is attested in Samoan, Kambera, Santali, and Guaraní.
5.6.2.1 Semantic compositionality

I argued in Chapter 2 (section 2.5.2.2) that flexible languages are
characterized by a mismatch between lexical categorization and syntactic
categorization. Derivational processes pertaining to the former type
of categorization typically involve semantically (and phonologically)
unpredictable interpretations. Syntactic categorization of flexible lexemes, in
contrast, yields compositional semantic meanings. The presumably flexible

languages discussed in Chapter 2 were shown to exhibit the latter type of
categorization, and can thus be analyzed as behaving in accordance with
Evans and Osada’s compositionality criterion. The relevant languages are
Samoan, Tagalog, and Kharia. At the same time, however, these languages
were shown to also display (zero-marked) processes of lexical derivation,
with unpredictable semantic outcomes. In this section, I will further discuss
this situation. In addition, I will show that similar situations are attested in
Kambera and Santali.
For Kambera I have given examples in (28) and (29) above, showing
that the insertion of presumably flexible lexemes into verbal versus nominal
syntactic functions yields compositional semantic interpretations. However,
there are also other types of processes in this language that are accompanied
by more idiosyncratic meaning shifts. As illustrated in (50) below, the types
of shifts are comparable to the ones discussed in Chapter 2. In particular, they
involve meaning such as: ‘the typical cognate object of the action denoted
by the root’, ‘a location associated with the action denoted by the root, ‘the
result of an action’, etcetera. As in Chapter 2, I argue that the examples in
(50) are instances of lexical zero-derivation, whereas examples (28) and (29)
illustrate ‘true’ flexibility.
Kambera (Klamer 1998: 110)

(50) Form Nominal meaning Verbal meaning
hukung ‘law / penalty’ ‘punish’
hindi ‘attic’ ‘dry / smoke’
bàndil ‘rifle’ ‘shoot’
dadu ‘dice’ ‘play dice’
lindi ‘bridge’ ‘cross a bridge’
ramuk ‘pool / swamp’ ‘be in a pool / swamp’
tanda ‘sign’ ‘know’
ludu ‘song’ ‘emit sound / sing’
Like the other flexible languages discussed in Chapter 2, Kambera also has
overtly marked lexical derivations. In fact, these forms are only formally
derived; the process itself is no longer productive. As would be predicted,
formally derived forms can be used in nominal as well as in verbal syntactic
environments. This is illustrated in (51a-b):

(51) a. Ku-manàhal-nya na n-jala-nggu
1sg.nom-regret-3sg.dat art mistake-1sg.gen
‘I regret my sins.’
b. Eha! N-jala-mbu-nggunja-i-ka nú kawài

excl be.wrong-also-1sg.ca-iter-pfv deic just.now
‘Darn! I was mistaken here too.’
On the other hand, Kambera also has derivational processes that produce
verbal output forms, i.e. forms that are no longer flexible, but rather
categorized in terms of phrase structure. I have illustrated these verbalizing
processes in (40) and (41) above.
Recall from Chapter 2 that the flexible language Kharia has the
possibility to insert complex phrases consisting of multiple roots or stems
into verbal as well as nominal syntactic functions (see example (51) in
Chapter 2). Interestingly, complex phrases in Kambera are flexible as well.
Consider for instance example (52), in which the translational equivalent of
an English NP functions as the head of a predicate phrase. The semantic
interpretation is compositional: ‘be X’.

(52) [Tau mayila]-mbu-kai nyimi ná
person poor-also-2pl.a you(pl) deic
‘Moreover, you are also poor people.’
Similar examples can be found in Samoan, as is shown in (53): The complex

phrase atu mata tasi ‘one-eyed bonito’ appears as an object argument with
the possessive form la’u in (53a), and as a predicate with the future tense
particle ole’ā in (53b). Again, the semantic interpretation of the phrase in
these two functional environments is entirely predictable.
Samoan (Mosel 2004: 286, 287)

(53) a. fia maua la’u [atu mata tasi]
want find my bonito eye one
‘I want to find me a one-eyed bonito.’

b. Ia, ‘ ole’ā [atu mata tasi]
well fut bonito eye one
‘Was that going to be the one-eyed bonito?’
Consider now Santali, which resembles the other flexible languages in that
it combines very regular meaning shifts with more idiosyncratic ones. The
former type of shift, which characterizes ‘true’ flexibility, is illustrated in (54)
and (55) below. First, examples (54a-b) show the compositional interpretation
of action-denoting roots in the nominal functions of direct object (zero-
marked for accusative), and dative-marked locative adjunct, respectively:
Santali (Neukom 2001: 17; Rau, forthcoming)

(54) a. Uni-ṭhɛn əḍi ruhɛt’-iɲ jɔm-akat’-ma
that-dat much scold-1sg.subj eat-pfv:act-ind
‘I got scolded badly by him.’ (lit: ‘I ate much scolding from him’)
b. Gapa-dɔ am-geg si-ok’-ʈ hɛn

tomorrow-top you-foc plough-mid-dat
ɖaŋgra-dɔ laga-əgu-kin-me
bullock-top drive-bring-3dat.obj-2sg.subj
‘Tomorrow you shall drive the bullocks to where I am ploughing.’
The examples in (55) illustrate the compositional interpretation of

semantically nominal lexemes in predicative function. In (55a) an individual-
denoting lexeme is used as the head of a predicative phrase. Examples (55b-d)
show that the same possibility exists for proper names, onomatopoeic forms,
and complex NPs, respectively 81.
Santali (Rau, forthcoming; Neukom 2001: 15)

(55) a. adɔ-e raj-en-a
then-3sg.subj king-pst:m-ind
‘So he became king.’
81
See Neukom (2001: 15) for comparable examples with pronouns, numerals, adverbs,
demonstratives, and quantifiers. Similar examples are provided by Peterson (2006: 62ff ) for
Kharia.

b. hɔpɔn-tɛt’-dɔ Ənuə-a-e-a
son-3poss-top Anua-appl-3sg.obj-ind
‘The name of their son was Anua.’ (lit.: Their son was Anua-ed.’)
c. Bar pɛ dhao ãã-y-en-a

two three time-3sg.subj onom-y-pst:m-ind
‘It (the buffalo) groaned two or three times’
In contrast, there are also instances of more idiosyncratic meaning shifts.

Some examples are provided in (56). These cases are interpreted as involving
zero-marked derivation, rather than flexibility.
Santali (Neukom: 2001: 20)

bicər ‘judgement’ ‘consider’
ḍhərwək’ ‘branch’ ‘send a branch’ (to inform someone
of the date of an event)
jɔm ‘food’ ‘eat’
bəhu ‘bride’ ‘take a wife for somebody’
Interestingly, Santali has the same overt lexical derivational process of

nasal-vowel-infixation that was discussed for Kharia in Chapter 2 (see the
examples in (52) of that chapter). However, in Santali the process is no
longer productive. Some examples appear in (57) (note in passing that in
Santali the process has an allomorph, namely -t-vowel):
Santali (Neukom 2001: 59)

(57) Base Lexical derivation
osar ‘to be broad’ – o-no-sar ‘breadth’
sarec’ ‘to remain’ – sa-ta-rec ‘remainder’
ɛhɔp ‘to begin’ – ɛ-tɔ-hɔp ‘beginning’
As in Kharia, these derived forms remain syntactically flexible, even though

their noun-like semantics suggest otherwise. This flexibility is illustrated in
(58), where ɛtɔhɔp ‘beginning’ appears as the head of a predicate phrase:

Santali (Neukom 2001: 60)
(58) oṛak’-iɲ ɛtɔhɔp’-akat’-a
house-1sg.subj beginning-pfv:act-ind
‘I have begun (to build) the house.’
Thus, the process exemplified in (56) is interpreted as the zero-marked

counterpart of overt lexical derivation, as illustrated in (57). If this is the
right analysis, then it is predicted that both the ‘nominal’ and the ‘verbal’
instantiations of the examples in (58) can be used nominal as well as verbal
syntactic environments. Unfortunately, however, the available data are not
conclusive on this point 82.
Finally, as mentioned in section 5.3.4 above, there are some languages in
the sample that seem to combine rigid classes of verbs and nouns with a class of
flexible heads that can be used as both. The languages in which this is attested
are Slave, Nivkh, West Greenlandic, and possibly Nunggubuyu. However, it
appears that in all cases the meaning shifts displayed by the relevant lexemes
are unpredictable, which disqualifies them as instances of flexibility. In (59) I
give some examples of presumably flexible items in Nivkh.
Nivkh (Matissen & Drossard 1998: 62)

kʿu ‘arrow’ ‘shoot’
ŋa ‘beast’ ‘pursue’
kelma ‘step’ ‘step’
kʿər ‘hunger’ ‘starve’
mos ‘porridge’ ‘grind’
əlv ‘roof ’ ‘thatch’
Similarly, Sadock (2003: 79) notes for West Greenlandic that “there is usually
an incompletely predictable meaning relation between homophonous noun / verb
pairs of stems.” Some examples are provided in (60):
82
For Guaraní, there are unfortunately no clear data about the semantic relationship between
nominal and verbal uses of flexible items.

West Greenlandic (Sadock 2003: 4)
iga / igavoq ‘cooking pot’ ‘cook’
imeg / imerpog ‘water’ ‘drink’
kalleq / kallerpoq ‘thunder’ ‘be thundering’
kuuk / kuuppoq ‘river’ ‘flow’
niu / niuvoq ‘leg’ ‘get out of a vehicle’
sianeq / sianerpoq ‘bell’ ‘ring up, telephone’
siku / sikuvo ‘ice’ ‘be frozen over’
Finally, also in Slave there appear to be different types of unpredictable

semantic shifts associated with the nominal and the verbal use of the same
root. Some examples appear in (61)83:
Slave (Rice 1989: 161)

t’éh ‘charcoal’ ‘cook’
tę ‘ice’ ‘freeze’
seeh ‘saliva’ ‘spit’
dzéh ‘gum’ ‘be sticky’
tthįh ‘axe’ ‘chop with an axe’
On the basis of the unpredictable semantics of the meaning shifts illustrated

in (59)-(61), I argue that these are not to be analyzed as instances of flexibility.
In addition, there is a crucial difference between the data in (61)-(63) from
Nivkh, West Greenlandic, and Slave on the one hand, and the lexical zero-
derivations discussed for the flexible languages on the other hand (see
examples (50) and (56)). This difference is due to the separation in flexible
languages between lexical and syntactic categorization. In flexible languages,
such as Samoan, the two meanings pertaining to a particular form reflect
lexical categorization only. Syntactically, i.e. in terms of phrase-structural
possibilities, both zero-derived forms are flexible, i.e. can be inserted in
verbal as well as in a nominal syntactic frames, with compositional semantic
interpretations. In contrast, in rigid languages, such as Nivkh, lexical and
syntactic categorization go hand in hand. Therefore, the lexical processes
83
The case of Nunggubuyu is less clear. As I have no data for this language of the type
discussed for Nivkh, West Greenlandic and Slave in (59)-(61), I postpone the discussion of
Nunggubuyu to the next subsection.

which zero-derive the two unpredictably related meanings of a single root
form also determine the phrase-structural possibilities of their output. In
other words, any lexically categorized item is also syntactically categorized:
it is either a noun or a verb, and can no longer be used in another function
than its defining function, at least not without further measures (see also
Don & Van Lier, forthcoming).
In sum, truly flexible languages can, in accordance with Evans and
Osada’s compositionality criterion, use simple roots, (zero-)derived stems,
and complex phrases in more than one syntactic function without extra
structural coding and with predictable semantic interpretations. The
possibility of semantic compositionality is interpreted as a direct consequence
of the fact that in flexible languages lexical and syntactic categorization do
not coincide as they typically do in rigid languages.
In the next sub-section I address Evans and Osada’s second criterion,
which is concerned with the quantitative measurement of lexical flexibility.
5.6.2.2 Exhaustiveness
According to the criterion of exhaustiveness, ‘true’ flexibility requires that
all members of a specific PoS class be flexible (see Chapter 2, section 2.5.3).
Thus, if a language is claimed to have a class of simple contentives, then
all content items belonging to that class should be usable in all relevant
propositional functions.
I already indicated in Chapter 2 that my descriptive sources for Tagalog,
Kharia, and Samoan are quite unambiguous about the fact that flexibility
in these languages concerns all or almost all content words (depending
on whether the class of contentives is combined with some (restricted)
additional rigid PoS class). First, Himmelmann claims for Tagalog that
“(…)all Tagalog content words (both roots and derived words) are
categorically indistinct, i.e. they may all occur in essentially the same
basic syntactic positions.” (Himmelmann, 2007: 249; emphasis added,
EvL)
For Kharia, Peterson writes the following:
“Almost all underived lexical morphemes in Kharia may be considered

precategorial in the sense that they may appear in referential, attributive
and predicative function, and [….] there is no reason to assume that

this is any different for derived stems.” (Peterson 2006: 83; emphasis
added, EvL)
Finally, for Samoan it is claimed that:
“The categorisation of full words into nouns and verbs is not given
a-priori in the lexicon. […] Derivations are more restricted in their
usage; some derivations can occur only in verbal functions.” (Mosel &
Hovdhaugen 1992: 73)
For the other languages in the sample that I have analyzed as displaying a
class of contentives the data are somewhat less straightforward. Regarding
Kambera, I explained earlier (see section 5.4.1) that Klamer (1998)
distinguishes between nouns and verbs in this languages on the basis of
differences in behavioural potential. However, if structural coding is the
only criterion, then all lexemes in Kambera that are not adverbs or derived
verbs apparently belong to a single class of contentives.
Regarding Guaraní and Santali, I have already mentioned that both
combine a class of contentives with a class of rigid, un-derived verbs.
(Nordhoff 2004: 56, 59 and p.c.; Neukom 2001: 13, 17). According to
Neukom, about one third of all content lexemes in Santali are verbs; the rest
are contentives. For Guarani, no such quantitative information is available.
Finally, there are some indications of the proportion of roots that serve
as the basis of both nominal and verbal zero-derivations in Slave, Nivkh,
West Greenlandic and Nunggubuyu, as discussed above. Even though these
cases are not regarded as cases of true flexibility, their degree of pervasiveness
seems interesting to report. For Nivkh, Matissen and Drossard (1998: 61)
remark that “most root types are fixed with respect to lexical categories”. This
suggests that the class of un-categorized roots is at least restricted as
compared to the classes of rigid nouns and verbs.
For West Greenlandic, Sadock points out that:
“The neat division of stems into two distinct part-of-speech classes

is somewhat complicated by the fact that there are quite a few stems
that occur both as a noun and a verb. Bergsland (1955) estimated that
about 200 out of 1500 roots are noun-verb homophones, the rest being
roughly equally divided between the two classes.” (Sadock 2003: 4-5)

In Slave, category-less roots are converted into verbal or nominal stems
by means of stem formation rules. There are three stem formatives that
create nouns, the most common of which is a null suffix. Verb stems are
composed of an uncategorized root and a suffix that indicates mode and
aspect. This process may involve overt suffixes but can also be zero-marked
(Rice 1989).
Finally, for Nunggubuyu the story is rather less clear. Heath (1984:
152) distinguishes “nominal adjectives” as a subclass of nouns. However, the
difference between the two classes is that nouns cannot be used as predicates
using a zero-1 strategy, while nominal adjectives can. In predicative
function, nominal adjectives take a pronominal prefix, like verbs, while in
referential function they take a nominal class prefix, like ‘regular’ nouns.
In addition, it appears that the propositional function of modification is
irrelevant in Nunggubuyu (see also section 5.3.4); modifiers always take the
form of appositional constructions. In my view, this means that the so-called
nominal adjectives in fact constitute a (relatively restricted) class of roots
that can receive either a nominal or a verbal instantiation.
In sum, as far my descriptive sources allow it, I have shown that most
relevant languages, i.e. those that I have classified as displaying a class of
contentives, satisfy the criterion of exhaustiveness.
5.6.2.3 Equivalent combinatorics

The criterion of equivalent combinatorics states that flexibility must be bi-
directional. This means that the members of a flexible word class should be
combinable in an equal way and to an equal extent with multiple functions.
Notably, Evans and Osada (2005) claim that flexibility in Mundari, the
language to which they apply their criteria, is in fact not bi-directional.
Their argument for this claim is that most action-denoting lexemes in
Mundari must be turned into headless relative clauses in order to be usable
in a nominal functional environment. An example is given in (62):
Mundari (Evans & Osada 2005: 377)

(62) susun-ta-n=iq landa-ja-n-a
dance-progr.or-intr=3sg.subj laugh-incep-intr-ind
‘The one who is dancing has laughed.’
However, it seems that Evans and Osada confuse the semantic denotation
of action-denoting lexemes in nominal function with their distributional

possibilities. A lexeme like susun ‘dance’ in (62) needs to be relativized only
if it is to receive the interpretation of ‘the one who is dancing’. As Evans and
Osada remark themselves, all verbs can indeed be used directly in what they
call “clausal argument positions”, and this yields a compositional semantic
interpretation, in which the relevant lexeme still denotes an action rather
than an object or an individual. Example (63) illustrates this:
Mundari (Evans & Osada 2005: 377)

(63) dub=ko laga-ja-n-a
sit=3pl be.tired.of-incep-intr-ind
‘They are tired of sitting.’
However, in some cases other semantic interpretations are found. An example

would be the root jom, which means ‘to eat’ in predicative function, and
‘food’ (rather than ‘eating’) in referential function. In fact, this pattern is very
reminiscent of the situation in other flexible languages. In particular, I would
argue that jom is a case of conversion: lexical, zero-marked derivation with
an unpredictable semantic outcome. This is in contrast to the case in (63),
which illustrates flexibility. In short, by adhering to the criterion of semantic
compositionality as proposed in the previous subsection, we may also solve this
potentially problematic aspect of the criterion of equivalent combinatorics.
A second aspect of equivalent combinatorics involves the claim that
truly flexible items should be usable to an equal extent in multiple functions.
As already pointed out in Chapter 2 (section 2.5.4), I disagree with Evans
and Osada that each flexible lexeme should appear equally frequently in all
propositional functions. In some of my descriptive sources, this issue of usage
frequency is explicitly addressed. For instance Peterson, in his comment on
Evans and Osada’s (2005) article, argues for Kharia that:
“[I]n fact, the only real restriction on this type of flexibility is what might
be termed semantic compatibility. For example, the loan ʈebal / ʈebul
‘table’ can function as the complement of a predicate with the meaning
‘table’, in the middle voice with the meaning ‘become a table’ or in the
active voice with the meaning ‘turn (something / someone) into a table’.
Needless to say, objects seldom turn into tables, hence the predicative
use of this lexeme is virtually never found in actual conversations.”
(Peterson 2005: 396)

Crucially however, such predicative usage is not excluded for grammatical
reasons. Peterson argues that a similar situation is likely to hold for Mundari,
since speakers of this language accept constructions like the one in (64):
Mundari (Peterson 2005: 400)
(64) Siŋboŋga am-e ʈebal-ked-me-a.

God 2sg-3sg table-act.pst-2sg-ind
‘God turned me into a table.’
In the same vein, Mosel and Hovdhaugen say the following about Samoan:
“Although certain full words seem to be used more as a verb or more

as a noun phrase nucleus for semantic reasons, there are no lexical or
grammatical constraints on why a particular word cannot be used in one
or the other function.” (Mosel & Hovdhaugen 1992: 73)
Regarding Tagalog, Himmelmann makes clear claims about the frequency

with which action-denoting roots are used in nominal function (he terms
this “unaffixed uses”, i.e. without voice marking):
“It should be clearly understood that in general unaffixed uses of action

roots are not in any way exceptional. Instead, they are reasonably
common both in terms of types as well as in terms of tokens. […]
Although there are differences with regard to how frequent and natural
it is for a given action root to occur without affixes, it is clearly the
case that Tagalog action roots quite generally allow for unaffixed uses.”
(Himmelmann 2007: 284)
In sum, the available data strongly suggest that the languages I have
analyzed as flexible indeed satisfy the criterion of equivalent combinatorics,
as far as the possibility to use lexemes in multiple functions is concerned. As
regards usage frequency, there are indications of asymmetry, but this only to
be expected on the basis of semantic-pragmatic markedness.
5.6.3 Non-verbs, nominals, and modifiers

In this sub-section I focus on languages that are characterized by a certain
degree of lexical flexibility regarding the following three functions: (i)
head of a referential phrase, (ii) modifier in a referential phrase, and (iii)

modifier in a predicate phrase. As we have seen in section 5.2, there are some
languages, in which these three functions are expressed by a single PoS class
termed non-verbs. This is the case in Warao, Turkish, Kayardild, and (if
small classes are included) Ma’di. Other languages have a class of nominals
that can fulfil the two functions of head and modifier in a referential phrase.
This is attested in Imbabura Quechua, Ma’di, and Hungarian. Finally, a
number of languages have flexible modifiers, i.e. lexemes that can function
as modifiers in both referential and predicate phrases. Below I discuss data
from languages with non-verbs, nominals, and modifiers in turn.
5.6.3.1 Non-verbs
In Warao, according to Romero-Figeroa (1997: 49) “there is no clear-
cut distinction between nouns and adjectives; Warao nouns may function as
attributives modifying other nouns.” Romero-Figeroa illustrates this claim
with examples involving lexemes that denote abstract properties, such as
yak ‘beauty’ in (65):
Warao (Romero-Figeroa 1997: 50)

(65) Hiaka yak-era auka saba tainisa-n-a-e
garment beauty-augm for she buy-sg-punct-pst
‘She bought a beautiful dress for her daughter.’
Note however, that it is not clear in how far object- and individual-denoting
lexemes can also function without structural coding as modifiers in referential
phrases. Moreover, there is no explicit evidence for bi-directional flexibility.
That is to say, there are no examples in which items like yak are used without
structural coding as heads of referential phrases (even though this possibility
is expected on the basis of the ‘nominal’ translation of the lexeme).
In addition to the fuzzy noun-adjective distinction, the lexical flexibility
in Warao presumably also includes the manner function. Example (66)
below shows that the item yak ‘beauty’ can be used without structural coding
as a modifier in a predicate phrase:
Warao (Romero-Figeroa 1997: 70)

(66) Tai hi saba yak-era ana tan-a-e
it you for goodness-augm neg happen-sg-punct-pst
‘It happened not so well for you.’

On the basis of these data, I have classified Warao as having a class of non-
verbs, but admittedly the evidence is rather patchy.
Turkish has a class of non-verbs as well (Göksel & Kerslake 2005:
49-50). In Chapter 2 this was illustrated with the lexeme güzel ‘beauty’. Note
that, like yak in Warao, Turkish güzel denotes a property concept. Therefore,
it may be questioned in how far object-denoting lexemes have the same
distributional possibilities. According to Göksel and Kerslake however,
indeed many lexemes display noun-adjective flexibility, even though each
lexeme has a ‘primary’ function in which it occurs most frequently for
semantic reasons.
Finally, Kayardild has a large class of lexemes that can be used flexibly as
heads and modifiers in referential phrases. Some examples are given in (67):
Kayardild (Evans 1995: 85, 238)

(67) maku: ‘woman / female’
balumbanda: ‘westerner / western’
wurkura: ‘boy / male’
jambanda: ‘hollow log / hollow’
Such items are also productively used as what Evans (1995) calls ‘secondary
predicates’. I regard secondary predicates as participant-oriented manner
expressions (see Himmelmann & Schultze-Berndt 2005). This means that
in Kayardild the functions of nouns, adjectives, and manner adverbs can
be expressed by a single class of lexemes: non-verbs. Notably, there is also
a small class of rigid manner adverbs in Kayardild (see figure 5.1 above).
Evans interprets these as a subclass of nominals, because they take nominal
morphology, as is illustrated in (68) below. However, he explicitly mentions
the difference in distributional freedom between these ‘manner nominals’ on
the one hand and non-verbs on the other: “The distinguishing feature of this
minor subclass is that it is restricted to one function, whereas the other nominal
classes are versatile.” (Evans 1995: 227; emphasis added, EvL).
Kayardild (Evans 1995: 259)

(68) Ngawarri bukawa-th
thirsty.nom die-act
‘(He) died thirsty.’

5.6.3.2 Nominals
Languages with non-verbs and nominals have in common that they lack a
distinction between nouns and adjectives. As explained in section 2.5.4 of
Chapter 2, this type of flexibility can be difficult to ascertain. In particular,
it is often problematic to establish whether property-denoting lexemes can
be referentially independent, or whether they rather function as modifiers
of understood heads.
Hengeveld (1992: 63) argues for Imbabura Quechua that property-
denoting lexemes can indeed be referentially independent. Example (69)
shows the use of the property-denoting word hatun ‘big’ as the sole element
of a referential phrase.
Imbabura Quechua (Schachter & Shopen 2007: 17)

(69) Rikaška hatun-ta
see.pst-1sg big(pl)-acc
‘I saw the big one(s).’
In contrast, Beck (2002: 146-149; see also Cerrón-Palomino 1987) claims

that Imbabura Quechua does have a lexical noun-adjective distinction. The
evidence Beck advances for this claim is that modifying lexemes with nominal
semantics, such as rumi ‘stone’ in rumi ɲan ‘stone road’, can occur only once
in a referential phrase, whereas modifying lexemes with ‘adjectival’ semantics
are recursive. According to Beck, this makes constructions with ‘nominal’
modifiers look very much like compounds. Support for this analysis is found
in the fact that when such ‘compounds’ function themselves as modifiers in a
referential phrase, the only possible reading is the one where the combination
of the two nouns in the ‘compound’ modifies the head. Thus, in hara čakra
rumi ‘stone from a cornfield’ the combination hara čakra ‘corn field’ modifies
the head rumi ‘stone’. On the other hand, as Beck acknowledges, there is no
phonological evidence for a ‘compound’ analysis. On a more general note,
Beck observes that, on the basis of isolated sentences in a descriptive grammar,
it is quite hard to decide upon the issue of noun-adjective flexibility in any
language. This is obviously true, and in the case of Imbabura Quechua it poses
a serious problem. Nonetheless, for lack of conclusive evidence to the contrary,
I maintain that Imbabura Quechua has a class of flexible nominals.
The second language with nominals is Hungarian. It was already briefly
mentioned in Chapter 2, where I pointed out that Hungarian property-
denoting lexemes can be referentially independent (Moravcsik 2001).

Examples (70a-b) and (71a-b) show that object-denoting and property-
denoting lexemes can both function as heads and modifiers in referential
phrases. Moreover, these usages can be distinguished from appositional
constructions, which are also open to both semantic types of lexemes. This
is illustrated in (72a-b)
Hungarian (Moravcsik 2001: 339)

(70) a. A ház-ak-at látom
the house-pl-acc I.see
‘I see the houses.’
b. A nagy-ok-at látom.
The big-pl-acc I.see
‘I see the big ones.’
(71) a. Ezr a bestia nö-t utálom.

this:acc the beast woman-acc I.hate
‘I hate this beast of a woman.’
b. A kél ház-ak-at látom.

the blue house-pl-acc I.see
‘I see the blue houses.’
(72) a. A ház-at, a szülöhely-em-et, látom

The house-acc the birth.place-sg1-acc I.see
‘I see the house, my birthplace.’
b. A ház-at, a kék-et, látom

The house-acc the blue-acc I.see
‘I see the house, the blue one.’
The third language in the sample with a class of nominals is Gooniyandi.

McGregor states that in this language:
“[N]early all words which can realise the Entity role can also realise the
role of Qualifier: in other words, words which can be referential can
also be qualifying, and may indicate qualities or properties of things.”
(McGregor 1990: 142).

It should be noted that the above quotation makes reference to unidirectional
flexibility only: the use of entity-denoting lexemes (semantic ‘nouns’) in
modifying function. With respect to the use of semantic ‘adjectives’ in
referential function, things are slightly more complicated. As McGregor
notes, for a number of property-denoting lexemes such usage has not yet
been attested. However, according to the author, this is due to pragmatic
markedness and “limitations of the data”, since many semantically similar (i.e.
other property-denoting) lexemes do occur as heads of referential phrases.
Therefore, I conclude with McGregor that there is insufficient evidence to
justify a noun-adjective distinction in Gooniyandi.
Noun-adjective flexibility in Ma’di has already been addressed in section
5.4.1. It seems that in this language property-denoting lexemes can indeed
be referentially independent. Note that, in addition to nominals, Ma’di
has a small, closed class of what appear to be flexible non-verbs. These
are property-denoting lexemes, which can be used as modifiers in both
referential and predicative phrases (and also as heads of referential phrases
if one assumes that there is no lexical noun-adjective distinction). Some
examples are given in (73) 84:

(73) Form Adjectival meaning Adverbial meaning
pélére clean properly
lɔ̄sɔ̄ good well
rīrī quick quickly
tʃé tʃé slow slowly
At this point, it may be worthwhile to consider, for the sake of contrast, some
examples of languages that may seem to display noun-adjective flexibility,
while in fact the absolutive use of adjectives requires the presence of an
understood head in the discourse context. In these languages, evidence for
such an understood head comes from the fact that it triggers agreement
on the absolutive modifier. In Chapter 2, I already gave an example of this
phenomenon from Spanish. Abkhaz is a similar case. Chirikba (2003: 29)
remarks that in this language it may be difficult to distinguish adjectives
from nouns, when the latter are used as the sole element of a referential
84
It is not clear whether these lexemes would remain property-denoting (i.e. become abstract
nouns such as ‘cleanliness’) or would rather be interpreted as an object / individual with the
property denoted by the base form (‘clean one’) when used as the head of a referential phrase.

phrase and combine with determiners and number markers. According
to Hewitt (1979: 226) however, the morphological form of an absolutive
adjective is determined by the unexpressed noun to which it functions as
an attribute. Also in Basque, absolutive use of adjectives is “exceedingly
common”. However, the animacy marking on the adjective is dependent on
the animacy status of its understood head. For instance beltza ‘the black
one’ takes animate morphology when the understood head is a horse, but
inanimate morphology when it is a skirt (Hualde & Ortiz de Urbina 2003:
149-150).
5.6.3.2 Modifiers
I will now discuss what can be regarded as the most marginal form of lexical
flexibility, involving the two modifier functions. As already mentioned, there
are some languages in which flexible modifiers constitute quite large, open
classes, while other languages rather have small or derived modifier classes 85.
As a first example, consider Lango, a language in which modification in
predicate and referential phrases is expressed in the same way: with the linking
particle à and a member of the class of flexible modifiers (Noonan 1992:
181). Notably, this class of modifiers exists alongside a class of rigid manner
adverbs, which take the particle nɪ̂, as illustrated in (74) below. Noonan (1992:
181) explicitly mentions the difference between flexible modifiers and rigid
manner adverbs: “Unlike type 1 [i.e. modifiers, EvL], these forms [i.e. manner
adverbs, EvL] can never have another grammatical function”.
Lango (Noonan 1992: 181)

(74) ŋámô cɛ̀m nɹ̂ mwə̀k- mwə̀k
3sg.subj.chew.hab food prt noisily
‘He chews (his) food noisily.’
85
The derived class of modifiers in Turkish had been mentioned several times already. It is
interesting to note that, although -CA derivations can be used as modifiers in both predicative
and referential phrases, they are always event-oriented (also when used attributively).
Attributive use is only possible with nominal bases (notional nouns or nominalizations) that
describe a result, or can at least take a resultative interpretation. The attributively used -CA
form expresses the manner in which the action has been performed, and stands in opposition
with an non-derived attribute, which simply predicates a property of the resultative noun,
without specifying the orientation of this property:
Akıllı-ca birplan Akıllı birplan

smart-mod a.plan smart a.plan
‘A smartly made plan.’ ‘a smart plan.’ (Schroeder 2004: 201)

Also in Ket, most modifying lexemes belong to the same flexible class.
However, there are a few lexemes that can be used as adjectives only, and
some others which are used exclusively as manner adverbs (Werner 1997:
119-120; Vaida 2004: 40).
In Itelmen, modifying lexemes can be divided into a morphologically
regular class and a morphologically irregular class (Georg & Volodin (1999:
106). The members of the former group have different forms when they
are usd as modifiers in referential versus predicate phrases: In the former
function they take the suffix -lah, and in the latter the suffix -q. This is
illustrated in (75a) and (75b), respectively:
Itelmen (Georg & Volodin 1999: 111, 112)

(75) a. K-k’ol̦-knen newen as-lah ŋejŋ-anke.
inf.iii>come<inf.iii dem high-adj mountain-dat
‘He came to a high mountain.’
b. A t’salaj iseŋ-q sal̦te-s-kinen

but fox quiet-adv follow-prs-3.sg:3.sg.i
‘But the fox was following him quietly.’
However, Itelmen also has a small, closed group of lexemes that are not
morphologically recognizable as adjective or adverb: They have the same
form in both functions. This is illustrated in (76a-b) with the lexeme
miça ‘beautiful’:
Itelmen (Georg & Volodin 1999: 119)

(76) a. Kninə-kit miça mimsx t-ənnksə-çen.
2.sg-caus beautiful woman 1.sg-let.go-3.sg.pat
‘Because of you I have let go of a beautiful woman.’
b. E, çasit miça sun-s-ç

interj now beautifully live-prs-2.sg
‘Oh well, now he is doing well.’
Finally, Thai also has a small class of flexible modifiers, existing alongside
rigid adjectives and rigid (derived) manner adverbs. An example of a flexible
modifier is dii ‘good, well’, which can be used as an modifier in a referential
phrase, as in khon dii ‘good person’, and as a modifier in a predicate phrase,

as in tham dii ‘do well’. (Iwasaki & Ingkaphirom 2005: 92). A similar system
attested in Abun has already been discussed in section 5.2.2.
5.6.4 Summary
In this section I presented data in favour of lexical flexibility in the relevant
languages of the sample. First, I discussed languages with contentives in terms
of Evans and Osada’s three criteria for flexibility. In line with the discussion
in Chapter 2 (section 2.5), I have proposed that the crucial characteristic
of a ‘truly’ flexible language lies in the separation of lexical and syntactic
categorization. This allows for the use of basically all content material in
all syntactic functions, with semantically compositional interpretations. The
only restriction on this flexibility concerns differences in frequency of use,
which can be explained in terms of relative semantic-pragmatic markedness
rather than absolute grammatical constraints.
Second, I discussed types of lexical flexibility with respect to the
functions of head and modifier in a referential phrase, and modifier in a
predicate phrase. In general, it seems that there has been less debate in
the literature – both theoretical and typological – regarding these types of
lexical flexibility. As a result, the kind of argumentation that should support
the identification of non-verbs, nominals, and modifiers in actual languages
is less developed, and sources provide less data that are relevant to the issue.
This also has repercussions for the thoroughness of the discussion of these
issues in the above section.
5.7 Conclusion, outlook

In this chapter I have presented the data on PoS classsification in the
languages of the sample, and evaluated them in light of the predictions and
theoretical issues of Chapter 2.
In the next chapter I turn to the classification of the dependent clause
constructions in the sample languages, within the functional-typological
framework developed in Chapter 3.

Dependent Clauses in the
Languages of the Sample
6.1 Introduction
This chapter presents a classification of the DC constructions attested in the
languages of the sample, in terms of the typology developed in Chapter 3.
First, in section 6.2, the DC constructions of each language are identified on
the basis of their structural coding, i.e. the form of either the subordinating
conjunction or the special marker on the dependent predicate. For every DC
defined in this way I determine which propositional function(s) it can express.
This yields a classification of DCs in terms of several rigid and flexible types.
These attested types are compared with the DC types predicted in Chapter
3 (section 3.2.3). Second, in section 6.3, the DCs identified in section 6.2
are classified according to their internal morpho-syntactic properties, i.e. in
terms of the (non-)expression of verbal and nominal features and argument
encoding. On the basis of these data, every DC construction is assigned
to one of the three structural DC types defined in Chapter 3 (section 3.4).
Finally, in section 6.4, the data presented in 6.2 and 6.3 are integrated into a
single typology. Section 6.5 rounds off with a brief summery.
6.2 DC constructions and propositional functions
6.2.1 Introduction
This section is organized as follows: in 6.2.2 I start out with a prelimimary
discussion, in order to demarcate the type of data that are (and are not)
187
included in the analysis. Following this discussion, all relevant DC types are
presented, together with the propositional functions that they can express.
Subsequently, Section 6.2.3 compares the attested DC types with those
predicted in Chapter 3. Section 6.2.4 is a summary.
6.2.2 Rigid and Flexible DCs: Attested types
6.2.2.1 Preliminary considerations

Before presenting the DCs attested in the languages of the sample and
their distributional patterns, a few remarks are required. They concern
three issues:
(i) The treatment of DCs that are normally used as heads or modifiers
in referential phrases, but also occur within larger constructions that
function in turn as predicate modifiers;
(ii) The definition of adverbial manner clauses, as opposed to other
types of adverbial clauses;
(iii) The distinction between clausal and lexical derivation (see also
Chapter 5, section 5.4.3.2).
I will discuss these issues in turn.
Nominal / adjectival DCs as predicate modifiers
In a number of languages, DCs that are normally used as heads and / or

modifiers in referential phrases (i.e. as complement clauses, relative clauses,
or nominal clauses) also appear within adverbial clauses, specifically in one
of the following construction types:
(i) Marked for oblique case;

(ii) Dependent upon an adposition;
(iii) As a modifier of a nominal head that means something like ‘manner’
or ‘way’.
Such cases are not regarded as instances of flexibility, since they involve
DCs that still function as a referential heads or modifiers, only within larger
constructions that function in turn as a predicate modifiers.

Consider for instance examples (1) and (2) from Tamil. In (1), the clausal
nominalization eŋka ammaa kaapi pooʈar-atu ‘our mother making coffee’
functions as the head of a referential phrase within the postpositional
phrase headed by poola ‘like’. This postpositional phrase has the function
of a predicate modifier. In (2) the same type of nominalization construction
biicile maɳalviiʈu katar-atu ‘building a sandhouse on the beach’ functions
referentially within a case-marked locative phrase.
Tamil (Asher 1982: 48, 21)

(1) [[Eŋka ammaa kaappi pooʈar-atu] poola-vee]
our mother coffee put:prs-nmlz like-emph
en mandevi pooʈaraa
my wife put-prs-3sg.fem
‘My wife makes coffee just as our mother makes it.’
(2) Koȥanteŋka [[biic-ile maɳalviiʈu kattar-atu]-ile]

child-pl beach-loc sand-house build.prs-nmlz-loc
curucuruppaa iruntaaŋka
busy be.pst.3pl
‘The children were busy building a sand-house on the beach.’
It is not always obvious how to distinguish between, on the one hand, a

nominalization with an oblique case marker or an adposition, and, on the
other hand, a dedicated adverbial clause construction 86. The same holds
for the difference between relative clauses modifying a nominal head and
dedicated adverbial clause constructions.
As regards the former case, in order to decide whether something is an
oblique case marker or a (bound) adverbial marker, it should be determined
whether or not the morpheme in question is part of a larger case marking
paradigm. The locative marker in the Tamil example in (2) above, for
instance, can be identified as a case marker because it is part of a paradigm
of eight other case forms (Asher 1982: 103). For adpositional constructions,
the relevant question is whether the marker is restricted to coding adverbial
manner clauses, or whether it can also be used productively with other
construction types, in particular with lexical nouns or NPs. The postposition
86
Adpositional or oblique case-forms of nominalizations can become independent from their
original paradigm. This diachronic development is a common scenario for the genesis of rigid
converb constructions (Haspelmath 1995: 17).
Chapter 6 – Dependent Clauses in the Languages of the Sample | 189

poola in (1) above does occur with ‘regular’ NPs and is therefore not analyzed
as (part of ) a dedicated adverbial conjunction (Asher 1982: 104).
In cases involving DC constructions that look like relative clauses, it
must be determined whether or not the extra coding element is a lexical
(nominal) head. Consider for instance Abun. In this language, adverbial
manner clauses are marked by sa gato, as illustrated in (3). The second part
of this complex marker (gato) has the same form as the relativizer, as can be
seen in example (4). However, the first part (sa) is a conjunction rather than
a lexical noun (Berry & Berry 1999: 158). Therefore, sa gato is analyzed as a
specialized complex marker of rigid adverbial manner clauses, rather than as
an instance of a relative clause construction modifying a nominal head sa.
Abun (Berry & Berry 1999: 158, 146):

(3) An da ben mó sarewo an yo ben
3Ssg actual do exist however 3sg neg do
kete bado yo teker [sa gato nyim
too.much maybe neg too.much adv earlier
ne nde re].
det neg pfv
‘Although she does [these things] she does not do [them] very
much, I mean, not like [she did them] before.’
(4) Men mu gu ye [gato man siri su men

1pl go kill person rel do wrong with 1pl
bi nggon].
poss woman
‘We will go and kill the person who committed adultery with our
(clans) woman.’
Basque displays a similar construction, but here the choice between a

relative and an adverbial construction is somewhat less straightforward.
Consider examples (5a-b); they show a DC construction marked with the
subordinating conjunction -(e)n, which can be used as a complement clause
(5a) and as a relative clause (5b). In example (6) the same DC appears within
a larger construction, which involves the extra element bezala ‘like’ and
functions as a predicate modifier:

Basque (Hualde & Ortiz de Urbina 2003: 646, 764, 722):
(5) a. Entzun dut [Amaiaren neba hil d-en]-a
b. [Pellok ekarri du-en] dirua galdu dut

Peter.erg bring aux-rel money.det lose aux
‘I lost the money that Peter brought.’
(6) [Zeuk esan didaz-en bezala] egin dut lana

you.emph say aux-adv do aux job
‘I did my job the way you told me.’
Adverbial constructions of the type illustrated in (6) also occur with other
forms than bezala, to express different meanings such as ‘until’ and ‘though’.
According to Hualde and Ortiz de Urbina (2003: 721), these other forms
are either “true postpositions” or “lexicalized inflected nouns”. These analyses
would characterize the construction in (6) as either a complement clause
depending on a postposition, or a relative clause modifying a nominal head
bezala, but not as a specialized adverbial clause construction. However,
it seems that bezala occurs exclusively with clausal complements. For
comparative constructions with simple nouns different (although related)
forms are used, such as bezain, as shown in (7):

(7) Koldo bezain azkarra zara
Koldo as.much.as fast.det are
‘You are as fast as Koldo.’
Moreover, when bezala does occur with a simple noun or pronoun, it must
take additional structural coding in the form of the attributive suffix ko:

(8) Zü bezala-ko polliterarik
you like-attr beautiful.abl
‘from [someone] as beautiful as you’

These distributional facts argue in favour of analyzing the construction
[clause + -en bezala] as a dedicated adverbial manner construction
Another Basque DC is marked by -ela, again a bound conjunction. This
DC can be used as a complement clause, as in example (9), and without
further measures as an adverbial manner clause, as illustrated in (10). Thus,
this case, unlike the previous ones, is indeed an instance of flexibility. Note
however, that there is a dialectal variant of (10), which involves the partitive
case marker -rik, as shown in (11), and as such does not involve flexibility.
Rather, it is a rigid complement clause construction, which takes oblique
case when expressing an adjunct.
Basque (Hualde & Ortiz de Urbina 2003: 635, 712, 713)

(9) batzuek uste dute [hauk oro kazeten eta
some.erg think aux these all journals.gen and
kazeta-egileen egitekoak dir-ela]
journal-makers.gen duties are-comp
‘Some think that all these are duties of journals and journalists.’
(10) [[Zer egin ez neki-ela] geratu nintzen

what do not knew-adv stay aux
‘I stood there not knowing what to do.’
(11) [[Jaikitzen da, jauzi egiten du-ela]-rik]

rise.ipfv aux, jump do.ipf aux-comp-part
‘(S)he gets up, jumping.’
In short, there are languages that express adverbial clauses by means of

a larger construction, in which a DC functions as a referential head or
modifier. Since these cases are not analyzed as instances of flexibility, there
is no reflection of the ‘adverbial use’ of these DCs in Table 6.1 below 87.
Finally, some languages express adverbial manner clauses through
reduplication of an infinitival construction. In my sample this strategy is
attested in Kharia and Imbabura Quechua. Example (12a) illustrates the
use of the Kharia infinitival construction as a complement clause (without
reduplication), and (12b) shows the same construction, with reduplication,
87
Nevertheless, I have included this information in Appendix III, where I mention, for
instance, that the Tamil atu-nominalization construction (see (1)-(2) above) can function
adverbially in combination with a postposition or an oblique case marker.

used as an adverbial clause. A similar pair of examples from Imbabura
Quechua appears in (13a-b). The reduplication is regarded as an iconically
motivated formal reflection of the fact that the SoA denoted by the DCs
stretches out temporally over the SoA denoted by the matrix clause, rather
than as additional structural coding. This means that the examples in (12a-b)
and (13a-b) are analyzed as instances of flexibility.
Kharia (Peterson 2006: 259, 249)

(12) a. [iɲ u ikuɁɖ sundar kontheɁɖ=ki=te
1sg this very beautiful bird=pl=obl
satay=na] um=iɲ lam=te
torment=inf neg=1sg want-act.prs
‘I don’t want to torment these beautiful birds.’
b. …lekin [lam=na lamna] souɁb=te ikuɖ jughay

but search=inf rdp all=obl very much
ɖaɁ-piyas laɁ=ki.
water-thirst emot=m.pst
‘But searching and searching, [they] all became very thirsty.’
Imbabura Quechua (Cole 1982: 40, 62)

(13) a. ñuka-ka [shuj ali wagra-ta-mi randi-y]-ta muna-ni
i-top one good cow-acc-val buy-inf-acc want-1sg
‘I want to buy a good cow.’
b. [Kanda-y kanday] shamu-rka-ni

sing-inf rdp come-pst-1
‘I came singing.’
Delineating manner clauses
A number of the DCs that are listed in Table 6.1 as adverbial manner clauses
also have ‘non-manner’ adverbial uses: They are used for instance as purpose
clauses or as cause clauses. Since this study is restricted to adverbial manner
clauses, such non-manner adverbial functions are not taken into account.
For the same reason, rigid DCs that express non-manner adverbial functions
only are not included in Table 6.1.

However, in many cases it is not straightforward to determine what
constitutes a ‘true’ manner clause, as opposed to other adverbial clause types
with related semantics. In a number of typological studies sets of adverbial
meanings have been proposed, sometimes in the form of semantic maps.
In such studies, the following types of adverbial meanings are typically
mentioned in close relation to manner: simultaneity, instrumentality,
accompanying circumstance, and comparison (König 1995, Kortmann
1998, Hengeveld 1998, Himmelmann & Schulze-Berndt 2005, Thompson
et al. 2007) 88.
In the present study, I make an (admittedly fairly rough) distinction
between two major types of adverbial clauses that are semantically close
to manner clauses. I will call these similative clauses and simultaneity
clauses, respectively. The former type has comparative meaning; it expresses
similarity between the SoA of the dependent clause with some other action
or psychological state expressed by the matrix clause. An example of a
similative construction is given in (14):
Pipil (Campbell 1985: 131; 289)

(14) Ah, ni-pa:xa:lua [ke:n-aken taha ti-ki:s-tuk ti-pa:xa:lua]
Oh I-walk just.as you you-leave-pfv you-walk
‘Oh, I’m taking a walk just like you have come out to take a walk.’
Simultaneity clauses are strictly speaking temporal clauses; they describe

the two linked SoAs as temporally overlapping. As such, ‘simultaneity’ is
meant here to include ‘accompanying circumstance’. In many languages,
simultaneity clauses are used by semantic extension to express that the
dependent SoA further specifies the manner in which the action expressed
by the matrix clause is carried out. An example of a simultaneity clause in
Babungo is given in (15) 89:
Babungo (Schaub 1985: 220)

(15) ŋwə́ bwéy [kɨ̀ shə́ŋ ŋwáa]
he sleep-impf sim rest-ipfv body
‘He was asleep, resting his body.’
88
This is not meant to be an exhaustive list. The terms largely speak for themselves; the exact
definitions proposed for each of them differ from one author to the next, but they are not
crucial for the present discussion.
89
Instrumental clauses will be regarded as a sub-type of adverbial manner clauses.

Similative and simultaneity clauses are included in the data whenever I could
not find evidence for a DC construction with ‘true’ manner semantics, and
whenever the boundary between manner clauses and manner-like clauses
was not clear-cut 90.
Clausal versus lexical derivation
It can be problematic to decide whether a construction is truly an instance of

clausal derivation, as opposed to lexical derivation. Consider for instance the
Ket infinitive and the Basque perfective participle constructions, illustrated
in (16) and (17), respectively. Since in these constructions no overt arguments
are expressed, it is not clear whether they should be analyzed as relative
clauses or rather as lexical adjectives derived from verbs.
Ket (Vajda 2004: 79)

(16) bʌ̀ ɣ ságdì
find.inf boot
‘A boot that is found’

(17) gizon ikas-i-a
man learn-pfv.ptc-det
‘a learned man.’
In the cases of Ket infinitives and Basque perfective participles, the

distinction between lexical versus clausal derivation is directly relevant to the
question whether these DCs are flexible or rigid, since both constructions
can also be used as complement clauses, in which function they do take
overt argument(s). This is shown in examples (18) and (19) 91:
Ket (Vajda 2004: 78):

(18) [ə́k-nà qús-bèt ] bínùt
2pl-anim.pl.gen tent-making.inf it.ended
‘We finished making the tent.’ (Lit: ‘Our tent-making ended.’)
90
In Appendix iii, a remark on semantics is added to similative and simultaneity adverbial
clauses.

Basque (Hualde & Ortiz de Urbina 2003: 668):
(19) Damu dut [zu irain-du]-a
regret aux you offend-pfv.nmlz-det
‘I regret having offended you.’
Of course, it is quite possible that the same suffix is used for both a lexical and
a clausal derivational process. This is even to be expected from a diachronic
point of view. However, this does not solve the problem of determining the
synchronic status of the relevant suffix. For the Basque case there is some
evidence in favour of a lexical analysis for (17): In this language, relative
clauses precede the head noun, while adjectives follow it (Hualde & Ortiz
de Urbina 2003: 791). Since the ‘participle’ in (17) is postnominal, this can be
taken as evidence that it is a derived adjective, rather than a relative clause.
Consequently, the perfective nominalization in (19) is analyzed as a rigid
DC, which can only be used in the function of head of a referential phrase.
In Ket, however, there is no evidence against analyzing the construction in
(16) as a relative clause. Therefore, the Ket infinitive construction is analyzed
as a flexible DC.
6.2.2.2 Language data

Table 6.1 displays the set of relevant DC constructions in the language of
the sample, and their possibilities to express each of the four propositional
functions. The languages are listed in the leftmost column, in the same
order as in Table 5.1 of the previous chapter. The DC constructions appear
in separate rows in the second column and are identified by their structural
coding 92. The four remaining columns represent the four propositional
functions of DCs: Head of a predicate phrase (Pred Head), head of a
referential phrase (Ref Head), modifier in a referential phrase (Ref Mod),
and modifier in a predicate phrase (Pred Mod). A plus sign in any of these
four columns indicates that the relevant DC can express that particular
propositional function. An empty cell indicates that the DC cannot express
91
The fact that in the Ket example both arguments have ALT coding still makes it difficult to
decide whether this is a deranked DC or a regular derived NP. In Malchukov (2006: 989), the
Ket infinitive construction is characterized as exemplifying the final stage of de-verbalization,
since it not only loses the inflectional slots of polysynthetic independent verb forms, but
cannot take derivational categories pertaining to valency and aspect (causative, resultative
etc.) either.
92
In Appendix iii, all DC constructions are listed and illustrated in the same order in which
they appear in Table 6.1. In this Appendix, the DCs are also identifiable through their
structural coding.

the propositional function in question. In a few cases the evidence was not
conclusive, which is indicated with question marks in the relevant cells.
Table 6.1: DCs in the sample languages and their possibilties concerning the expression of
propositional functions
Language Structural Propositional Functions
Coding Pred Head Ref Head Ref Mod Pred Mod
Tagalog pag- ? + + +
na/-ng + + +
Kharia RDP/Ø + + + +
-na + + +
-na-wala +
-al +
-ker etc. +
-ga + RDP +
-ta + RDP +
no +
gam-kon +
je etc. +
a/i etc. +
Ø +
Kambera pa- + +
ma- +
Ø (NMLZ) +
wà +
Samoan =ga +
Ø (NMLZ) +
ona/ina +
-e +
Ø + +
Guaraní há(gwe)/Ø +
va +
vo +
Santali Ø (no INDIC) + +
-kate +
CORREL +
mente +

Warao -kitane +
kotai +
Turkish -DIK/-(y)AcAK + +
-mAK +
-mA +
-An +
-(y)ArAk +
-(y)A…-(y)A +
ki + +
diye +
Ø +
Kayardild -n- + + +
-Thirri-n +
-n-garrba +
-ntha + +
Ø +
Paiwan tu(a)/tjai +
a + +
a parhu +
(-)in- -an + a +
I. Quechua -j/-shka/-na + +
-ngapaj +
-chun +
-y + +
-shpa +
Ma’di -lɛ̄ + +
-dƷɔ + +
-ka +
-rɛ̄ /-ɓá +
Ø +
-zɨ̄ + sɨ̀ +

Gooniyandi -woo +
-wadda +
-mawoo +
-bari +
-ya/ +
-gowaaya
Ø (+RSP) + + +
Hungarian -ni +
-ás/-és +
-ó +
-ótt +
-andó/-endó +
-vá/-vé +
-vén +
(azt) hogy (etc.) +
úgy (a-)hogy +
REL.PRON +
Japanese no/mono etc. +
Ø (RSP) +
-te/-de/-ite +
-i/-Ø +
Hmong qhov +
Njua kuam/(has)tas +
Ø +
kws +
Lango -(kk)ɔ̀ +
nî +
à-mɛ̀ +
Ket Ø (bare INF) + +
Ø + +
ɛta qɔr’a +
ásqà +
-PROSEC +
REL.PRON +
-s/-bes +

Itelmen INF (various +
forms)
Ø +
min +
quatz +
Koasati NMLZ (various +
forms)
-:sáya +
-:yólli +
-:ka +
-:kítta +
-laho:li:sáya +
-n +
-k +
-t +
Thai Ø + +
thiî + +
wâa +
Basque -t(z)e +
-tu/-du/-i/-Ø +
-en + +
-en bezala +
-(e)la + (+)
bait + +
Abun do/Ø +
gato +
sa gato +
Bambara ka +
-le/-ne +
-min(u)/ +
-mun(u)
-tò +

Georgian -a +
m- (-a-)(-el/ +
-al))
-ul/-il/m- -ar/ +
-al
sa- (-el/-al/r) +
rom + +
REL.PRON +
ra +
rogorc +
Bukiyip Ø +
(ú)li +
bwidouk -(u) +
mu
Abkhaz -ra +
-NFIN +
REL.PRON +
-NFIN
-ş- + -NFIN +
Polish -nie +
INF (various +
forms)
-c +
-any etc. +
(PST.PASS) PTC +
(various forms)
PRS.PL-c- +
że +
REL.PRON +
jak (gdy)by +
Burushaski -(á)as + +
-im/-um/-am +
n-STEM-(a)n +
ke/ki + + +
sén-/ét- +

Lavukaleve -e/-i +
PERSON.SUFF +
Alamblak -nef +
-(kfë)t +
ɨnd +
Pipil ka(h) + +
ne +
ke +
ke:n-aken +
ADV +
Wambon -e +
-a + o +
Ø +
-mo/-o +
ka +
Dhaasanac DET(+DEM) + +
-ɲ/-an +
Berbice fu/fi/Ø +
Dutch bifi/dati/Ø +
Wh/Ø +
Babungo lāa +
fáŋ/yúu + +
kɨ̀(i)/Ø +
Nama !xáis-à +
(no INDIC)
Ø (no INDIC) +
(RSP)
-se/!’aa/tsii/ +
Ø (no INDIC)
Hdi ká +
tá + +
tà +
tà + NMLZ +
-a +

Mandarin Ø + + +
de + +
-zhe +
Tamil -atu +
-(kk)a +
-a +
Kisi (m)àà/Ø +
Ø +
CL +
Nung Ø + + +
(tị-vạ) (+DEM) +
(+FOC)
bạ t +
Garo in-e +
-a +
-a-ni +
-na +
-kan/-kan-a +
-gip-a +
-e/-e-min/-e-r +
Krongo àní tíŋ +
-(t) +
m-/n- + +
Hixkaryana nɨ/-thɨ-/ +
-hɨto + rɨ
Slave nį/Ø + +
gha/gú +
i/sį́i/líi +
gháré +
Nivkh -vut/-vur +
Ø +
NFIN +
r/-t-ř//-n +

Greenlandic -niq +
-ta/-sa +
PTC mood + +
CONT mood + +
6.2.3 Discussion
6.2.3.1 Attested and predicted types
The data in Table 6.1 show that most of the attested DCs represent predicted
types. First, as far as rigid DCs are concerned, there are complement clauses,
relative clauses, and adverbial manner clauses. The functional distribution
of these DC types is represented in (19), (20), and (21), respectively (cf. (9),
(10) and (11) in Chapter 3):
(19)
Head Modifier
Predication
Reference Complement clause
(20)
Head Modifier
Predication
Reference Relative clause
(21)
Head Modifier
Predication Manner adverbial clause
Reference
I will illustrate each of these rigid DC types with examples from the sample
languages. Example (22) is a rigid complement clause from Kisi; this
construction can be used in the function of head of a referential phrase only.

Kisi (Childs 1995: 280)
(22) ò cà kó bɛ́ɛ́ [màà tè ò cǎŋ pɛ́ ò fùlá lènǐŋ]
he see only indeed comp if he cry if he go.out inside
‘He really thinks that if he cries he will be free.’
Note that the behavioural potential of DCs does not play a role in the
typology yet. Thus, rigid deranked nominalization constructions, such as in
(23) from Turkish, also count as complement clauses.

iste-n-iyor-muş
‘It seems they want everyone to tell a story.’
Example (24) is a rigid relative clause construction from Hmong Njua:
Hmong Njua (Harriehausen 1990: 141)

(24) Tug txiv neej [kws kuv saib] sab
cl man rel 1sg see big
‘The man that I saw was tall.’
Relative clauses may also involve deranked participle constructions, as in

example (25) from Hungarian:
Hungarian (Kenesei et al. 1998: 45)

(25) [A könyv-et a fiú-nak gyorsan olvas-ó]
the book-acc the boy-dat fast read-act.prs.ptc
lány itt van
girl here is
‘Here is the girl who reads the book to the boy fast.’
A rigid adverbial manner clause construction appears in (26), from Guaraní.
Guaraní (Gregores & Suárez 1967: 180)

(26) H-asẽ́ [o-sẽ́ vo]
she-cry she-go.out adv
‘She goes out crying.’

Rigid adverbial manner clauses may also be deranked converbal constructions,
such as the one in (27) from Polish:
Polish (Bielec 1998: 71)

(27) [Sɫuchają-c muzyki,] ubraɫem siȩ.
listen-conv music dress.I (my)self
‘Listening to music, I got dressed.’
Turning to flexible DC types, the following predicted types are attested:

multi-functional clauses, modifier clauses, nominal clauses, and flexible
clauses of type C (which can be used as complement clauses and adverbial
manner clauses). The functional distributions of these DC types are
represented in (28), (29), (30), and (31), respectively (cf. (13), (14), (15) and
(20) in Chapter 3):
(28)
Head Modifier
Predication
Reference Multi-functional clause
(29)
Head Modifier
Predication Modifier clause
Reference

(30)
Head Modifier
Predication
Reference Nominal clause
(31)
Head Modifier
Predication Flex clause C
Reference Flex clause C
Below I will again illustrate each of these DC types with examples from
the sample languages. Multifunctional clauses are attested in the isolating
languages Mandarin Chinese and Nung. In both cases the DC is zero-
marked. Examples (32a-c) are from Mandarin, and show that the DC can
be used in the functions of head of a referential phrase (32a), modifier in a
referential phrase (32b), and modifier in a predicate phrase (32c) 93.
Mandarin Chinese (Li & Thompson 1981: 599, 611, 597)

(32) a. Wŏ pànwàng [nĭ kuài yidiăn bìyè]
I hope you soon a.little graduate
‘I hope you’ll graduate a bit sooner.’
b. tā yŏu yi-ge mèimei [hĕn xĭhuān kàn diànyĭng]

3sg exist one-cl younger.sister very like see movie
‘S / he has a younger sister who like to see movies.’
c. Tāmen [yòng shŏu] chī-fàn

they use hand eat-food
‘They eat food using their hands.’
Deranked multi-functional clauses are attested in Kayardild. So-called

‘nominalizations’ in this language can be used as the head and modifier of a
referential phrase, and as a modifier in a predicate phrase, as is illustrated in
(33a-c) below. Note that the structural coding of the multi-functional clause,
93
Burushaski appears to have an overtly marked, balanced multi-functional DC, borrowed
from Persian: This ke / ki construction is claimed to be usable as a complement, relative and
adverbial clause, but my sources provide no clear examples (see Tikkanen 1995: 498).

the -n- marker, is glossed according to the function of the DC in the relevant
example. Thus, in (33a) -n- is glossed as a nominalizer, in (33b) as a participle,
and in (33c) as a converb. A similar procedure is followed for the glossing
of structural coding in other types of flexible clauses, including balanced
clauses. In the latter, the subordinating conjunction will be glossed as a
complementizer (COMP), a relativizer (REL), or an adverbial subordinator
(ADV), according to the function performed by the DC in the sentence (cf.
for instance examples (9) and (10) above). This is to avoid tedious terminology
such as ‘dependent predicate marker of a multi-functional balanced clause
construction’, and opaque corresponding glosses. Note however, that the
disadvantage of this procedure is that the gloss of an isolated example does
not reveal whether the DC is a rigid or a flexible construction.
Kayardild (Evans 1995: 476, 474,

(33) a. Ngada kurri-ja [niwan-ji budii-n-marri]
1sg.nom see-act 3sg.poss-mloc run-nmlz-priv
‘I saw that he was not running.’
b. Nga-ku-l-da [wirr-n-ku] dangka-wu kurri-ju

1-inc-pl-nom dance-ptc-mprop man-mprop see-pot
‘We will watch the dancing man.’
c. [Bilaangka-nurru kari-i-n-da] ngada warra-j

blanket-assoc cover-m-conv-nom 1sg.nom go-act
‘I went along, covering myself in a blanket.’
Flexible modifier clauses are attested in Babungo, Mandarin Chinese, and

Krongo. The balanced modifier construction in Babungo is illustrated below.
In (34a) the DC functions as a modifier in a referential phrase, and in (34b)
as a modifier in a predicate phrase.
Babungo (Schaub: 1985: 32, 39)

(34) a. mə̀ yè wə́ ntɨ ə́ [fáŋ (ŋwə́) shàw ngú yê]
I see.pfv person that rel he steal.pfv fowl your
‘I have seen that man who has stolen your fowl.’

b. ŋwə́ sɔ̀ sāy [fáŋ nshú wī nàysə́ tɨ́ ŋwə́]
she plant.pfv corn adv mother her tell.pfv to her
‘She has planted the corn like her mother told her.’
The Krongo modifier clause construction is illustrated in (35a-b). This is

a deranked construction to the extent that it does not express agreement
and takes a possessive subject. Tense, aspect, and valency / voice marking are
retained, however.
Krongo (Reh 1985: 256, 333)

(35) a. n-úllà àʔàŋ kí-ǹt-àndiŋ
1 / 2-ipfv:love I loc-sg-clothes
[n-úufò-ŋ kò-nìimò kàti]
ptc:neut-ipfv:sew-tr poss-mother my
‘I love the dress that my mother is sewing.’
b. n-áa t-ánkwà-ànì
conn-neut-cop ink-go.round-dtr
[n-úrùná-ŋ úuní kànáày]
conv:neut-ipfv:pay.attention.to footprints poss:3.pl
‘She goes round, paying attention to their footprints.’
Nominal clauses, i.e. DCs that can be used as the head and the modifier in a
referential phrase, occur quite frequently. An example of a balanced nominal
clause construction in Basque appears in (36a-b):
Basque (Hualde & Ortiz de Urbina 2003: 646, 764)

(36) a. Banekien [Mikel berandu etoriko z-en]-a
ba-knew Mikel late arrive.fut aux-comp-det
‘I knew that Mikel would arrive late.’

Deranked nominal clauses are attested for instance in Turkish, as shown in

example (37a-b). Note that the difference in form of the special marker on
the predicate in the two clauses is due to vowel harmony.

Turkish (Göksel & Kerslake 2005: 423, 442)
(37) a. [Orhan-ɪn bir şey yap-ma-yacağ-ı]
Orhan-gen anything do-neg-nmlz-3sg.poss
belliydi.
it.was.obvious
‘It was obvious that Orhan wouldn’t do / wasn’t going to do
anything.’
b. [Fatma-‘nın yarın gör-eceğ-i] film

‘the film that Fatma is going to / will be seeing tomorrow’
Finally, a few cases are attested of flexible DC clauses of type C, which

can be used as complement clauses and as adverbial manner clauses (see
(32) above). One case is the Basque -ela construction, which was already
illustrated in (9) and (10) above. Recall that this flexible construction has a
dialectal variant in which the adverbial use is accompanied by an extra case-
marker -rik (see (11) above) 94.
Another case of a flexible clause of type C is attested in West Greenlandic.
It is called the contemporative or conjunctive mood construction and it is
formed with the mood-marker -lu / -llu, followed by a person marker. This is
illustrated in (38a-b).
West Greenlandic (Fortescue 1984: 40; 55)

(38) a. isumaqatigiip-pugut [maani ata-ssa-lluta]
agree-1pl.ind here stay.on-fut-1pl.cont
‘We agreed to stay on here.’
b. [Irnir-lunga] isir-punga
do.hurriedly-1sg.cont go.in-1sg.ind
‘I entered quickly / in a hurry.’
Even though flexible DC constructions that can be used as complement

clauses and as adverbial manner clauses were predicted in Chapter 3, their
particular distributional pattern seems unexpected. The reason for this is
94
In addition, Hualde & Ortiz de Urbina (2003: 712) quote sources claiming that the marker
-ela used for adverbial manner clauses is not exactly homophonous with the complementizer
-ela, since they have a different accentual pattern.

that this DC type is flexible between the head function of one domain
(reference) with the modifier function of the other domain (predication). As
such, it merges two functions that are maximally different in terms of both
the predication-reference parameter and the head-modifier parameter 95.
This marked situation is apparently reflected by the infrequent attestation
of the relevant DC type.
On the other hand, a possible connection between complement clauses
and adverbial clauses lies in the fact that they function as arguments and
adjuncts, respectively. Complementizers often develop out of adpositions
or case markers, which may already have been used for the marking of
both direct object arguments and (certain types of ) adjuncts (Cristofaro
1998). This flexibility may carry over to the coding of clausal constructions
expressing these two functions, i.e. to complement clauses and (some types
of ) adverbial clauses.
In this subsection I presented all DC types attested in the sample,
according to their functional distribution. These attested types do not cover
all the types predicted in Chapter 3, section 3.2.3. In the next sub-section
I discuss the predicted but unattested DC types.
6.2.3.2 Predicted but not attested DC-types: The problem of

predicate clauses
The following predicted DC types were not attested in any of the sample
languages: predicate clauses, contentive clauses, predicative clauses, head
clauses, and flexible clauses of types A, B, and D 96. These DC types are
presented in (39)-(45) below (cf. (8), (12), (16), (17), (18), (19), and (21) in
Chapter 3):
(39)
Head Modifier
Predication Predicate clause
Reference
95
cf. Chapter 5, section 4.3.4, on the non-attestation of PoS classes with this distributional
pattern.

(40)
Head Modifier
Predication Contentive clause
Reference
(41)
Head Modifier
Predication Predicative clause
Reference
(42)
Head Modifier
Predication
Reference Head clause
(43)
Head Modifier
Predication –
Reference Flex clause A
(44)
Head Modifier
Predication
Reference Flex clause B –
(45)
Head Modifier
Predication Flex clause D
Reference Flex clause D
96
In fact, there appears to be one instance of a contentive clause construction, attested in
Kharia, which I will discuss shortly.

Notably, all these DC-types involve the functional slot for head of a predicate
phrase. As explained in Chapter 2 (section 2.3.1) and Chapter 5 (section 5.5),
this function has a special status in Hengeveld’s theory of PoS classification.
In particular, the function of head of a predicate phrase is the only function
of verbs, while it is a possible additional function of lexeme classes that are
not verbs. Such a non-verbal lexeme class is regarded as having an addition
predicative function only when its members display the same behavioural
potential as verbal predicates (i.e. when a zero-1 strategy is employed).
In order to preserve the parallel between PoS classes and DC
constructions, the latter are subjected to the same criterion as the former:
DC constructions (which are normally used in one or more of the functions
of head and modifier in a referential phrase and modifier in a predicate
phrase) are regarded as having an additional predicative function only when
they employ a zero-1 strategy in that function.
Hypothetically, to use a zero-1 strategy with a DC in predicative function
would mean that verbal features take scope over the clausal construction as a
whole. In the case of a balanced, finite DC (with verbal morphology already
present on the dependent predicate), this would involve double expression of
verbal categories, i.e. a kind of ‘stacking’ of verbal inflection. In the case of a
deranked clause it would mean that constructions with typically non-finite
predicates would express verbal features. In both cases one could imagine that
verbal features somehow appear at the periphery of the DC construction.
However, none of these scenarios is actually attested in the sample languages.
Rather, the use of a DC construction in predicative function triggers either
a copula strategy or a zero-2 strategy. Consider for example the English
infinitival construction in (46), which in predicative function needs a copula:
(46) My plan is [to take you on a trip to the beach]
In short, DC constructions with non-predicative defining uses do not

employ the zero-1 strategy when used predicatively. One reason for this
may be the structural problems involved in inflecting a DC in the same
same way as a lexical verb. This may explain why flexible DC constructions
involving the function of head of a predicate phrase (as represented in (40)-
(45) above) are not attested. It may also explain the non-occurrence of DCs
that are specialized for the function of head of a predicate phrase, i.e. rigid
predicate clause constructions (cf. (39) above).

In addition to this formal restriction, DCs (in particular deranked ones)
appear to be conceptually marked with respect to the function of predicative
head: It has been claimed that he SoAs expressed by DCs lack an autonomous
cognitive profile, and this arguably makes them unlikely candicates for the
expression of independent predications (Langacker 1987; Cristofaro 2003,
cf. Chapter 3).
In view of this last point, it is interesting to consider in more detail
the contentive clause type (represented in (40) above). As noted above, this
clause type, although predicted in Chapter 3, is practically non-existent
in the sample languages. However, there is in fact one DC construction
that comes close to this type, since it can be used in all four functions, and
apparently employs a zero-1 strategy in predicative function. This is the
Kharia DC construction with so-called freestanding forms. These are bare,
non-finite dependent predicate forms, sometimes with stem reduplication 97.
When a freestanding form construction is used in predicative function, it
takes clitics for voice / tense and person, just like any lexical predicate. As
such, this DC construction seems to meet the criterion for a zero-1 strategy.
However, the use of voice marking on freestanding form constructions is
restricted, compared to lexical predicates: Freestanding form clauses can
only take the middle voice, and they always have a habitual interpretation.
Compare the example in (47a), showing a freestanding form construction in
the function of predicative head, with the example in (47b), its counterpart
with a lexical predicate.

(47) a. iɲ ɖaɁ biɁɖ-biɁɖ=ki=ɲ
1sg water pour.out-rdp=m.pst=1sg
‘I used to pour water out.’ (i.e. that was my job)
b. iɲ ɖaɁ biʈh=oɁj
1sg water pour.out:act=pst.1sg
‘I poured water out.’
It seems that the special behaviour of freestanding form clauses in this

respect is motivated in cognitive-semantic terms. As Peterson puts it:
97
They are called masdars in Peterson (forthcoming).

“I would argue that the habitual interpretation of such forms in Kharia
results from the fact that [they are] intimately connected to a depiction
of the event without explicit reference to its internal temporal structure.
If such a form is nevertheless marked as a finite predicate, what results
is a habitual situation, not an activity or event in the usual sense [...].”
(Peterson 2006: 74)
The case of Kharia freestanding froms may be compared to so-called

gerund constructions in Tagalog. Like Kharia, Tagalog has a class of
lexical contentives. Unlike Kharia, however, it has no phrasal clitics. The
Tagalog gerund construction is a deranked DC that can be flexibly used in
all propositional functions. However, in the function of predicative head it
does not employ a zero-1 strategy: Lexical predicates in Tagalog appear in
sentence-initial position and are marked for voice (Himmelmann 2007). A
predicatively used gerund construction also stands in first position, but it is
not marked for voice, as can be seen in example (48).

(48) [pag-lu-lutoɁ ng pagkain] ang trabaho niyá
ger-rdp-cook gen food spec work 3sg.poss
‘His / her job is cooking food.’
Thus, the Tagalog gerund construction should be analyzed as employing

a zero-2 strategy in predicative function. However, it seems that there is a
functional reason for the lack of voice-marking on gerund constructions. In
particular, Tagalog voice-markers orient the action denoted by the predicate
towards one of the participants of that action. Gerunds, in contrast, “refer
to actions or states without orienting them towards one of the participants”
(Himmelmann 2005: 372). In other words, gerund marking and voice-
marking are mutually exclusive in functional terms. In fact, there is a
systematic co-variation between the allomorphs of voice markers and the
allomorphs of the gerund marker. For instance a form taking -um- as the
active voice marker, takes pag- as a gerund marker, whereas a form taking
mag- as the active voice marker, takes pag- plus reduplication for gerund
formation (see Schachter & Otanes 1972: 160-161 for the full paradigms).
In general, it seems that there are both structural and cognitive-semantic
factors that inhibit the use of DCs as predicative heads. This explains the
non-attestation of the predicted flexible DC types that involve the function

of predicative head, as well as the non-attestation of rigid predicate clauses.
In addition, it means that the function of head of a predicate phrase is in
practice largely irrelevant to the purpose of the present study, namely the
comparison of the distributional patterns of PoS and DCs with respect to
propositional functions. Whereas the function of head of a predicate phrase
appears to be the most central function in terms of PoS classfication, it is the
most marginal one for DCs.
6.2.4 Summary
To summarize, in this section I have shown that all DC types predicted in
Chapter 3 are attested in the languages of the sample, except those involving
the functional slot for head of a predicate phrase. This latter finding was
explained in terms of the structural incompatibility of DC constructions
with the zero-1 strategy, and their cognitive-semantic markedness with
respect to the function of independent predication.
6.3 DC constructions and behavioural potential
6.3.1 Introduction
This section describes the second step towards the DC typology as developed
in Chapter 3, namely the categorization of every DC construction (as listed
in Table 6.1 of the previous section) in terms of its internal morpho-syntactic
properties. I start out with a brief discussion of the formal parameters that
are taken into account. In section 6.3.2 I then present the basic data as
regards the (non-)expression of verbal categories and nominal categories in
the DCs, and the realization of their arguments. Subsequently, in section
6.3.3, these data are used to assign every DC to one of the three structural
DC types defined in Chapter 3 (section 3.4). In section 6.3.4 the two datasets
of 6.3.2 and 6.3.3 are combined: The internal formal properties of all DCs
are presented in the form of three separate tables corresponding to the three
structural DC types.

6.3.2 Basic Data
6.3.2.1 Formal parameters
Verbal and nominal categories
Data on the (non)expression of following verbal categories are included in

Table 6.2:
– voice and valency marking

– tense marking
– aspect marking
– mood marking
– person marking (including agreement 98 and cross-reference)
– illocutionary force (indicative or declarative marking)
The nominal categories included in Table 6.2 are:
– determiner expression
– case / adposition marking
Number and gender / class marking are not systematically taken into account,
because these categories are very infrequently attested in the DCs of the
sample languages 99.
For each DC it is determined which verbal and nominal categories are
expressed (indicated with a ‘+’ in Table 6.2) or not expressed (indicated with
a ‘–’ in Table 2). An empty cell means either that the category is not relevant
in the language in question, or that my sources provided no information
about the (non-)expression of the category. The appearance of ‘+ / -’ indicates
one of the following situations:
(i) The expression of the relevant category is possible but not

obligatory;
98
Recall from Chapter 3 that agreement refers to subject agreement, whereas object
agreement is documented in the column for voice-valency marking.
99
In Appendix iii I do include information about these categories, whenever relevant and
available. This holds also for number / gender / class agreement on relative clause constructions.

(ii) Some distinctions of the relevant category are retained while others
are lost (as compared with the range of distinctions available in
independent clauses);
(iii) The category is expressed in some but not all of the propositional
function(s) and / or specific subordination relations that can be
expressed by the DC in question.
Below I will provide examples of each of these three situations. The first one
is applicable to complement clauses in Guaraní. Examples (49a-b) show that
this DC construction may appear either with or without a definite article.
Guaraní (Gregores & Suárez 1967:158)

(49) a. Rey-anú [šé še-rasɨ̀ há]
you-hear I I-be.sick comp
‘You heard that I was sick.’
b. ai-kwaá la [n o-ù moɁã́ i há]

I-know art neg he-go mod neg comp
‘I know that he does not intend to go.’
The second situation occurs in Hixkaryana. Nominalization constructions

in this language retain three out of the seven tense distinctions that can be
expressed in independent clauses (Derbyshire 1979: 25).
A clear example of the situation mentioned under (iii) above is
found in Santali. In this language DCs without the indicative marker -a
exhibit variation in the expression of TAM and person, depending on the
propositional function in which they are used, as well as the particular
semantic and syntactic subordination relation that they express. When the
construction functions as a complement clause in subject function, it shows
no subject marking, while middle voice markers and TAM can be expressed.
In the function of object complement clauses, it lacks subject and tense
marking, but object marking is retained. Only complements of perception
predicates can express subject marking as well as all TAM distinctions.
Finally, when the DC functions as a relative clause, subject marking is lost,
while all TAM distinctions may be expressed.
Note further that in some cells of Table 6.2 a plus sign appears between
brackets: ‘(+)’. When referring to a verbal feature, this indicates that the
relevant category is expressed by means of special dependent forms, either

because the structural coding of the construction fuses with the expression of
a TAM value, or because there is a special dependent / subjunctive paradigm
for the expression of TAM and / or person marking. In Chapter 3 I gave some
examples of these situations, as attested in Imbabura Quechua, Georgian,
and Abkhaz (see examples (6), (7), and (47) of Chapter 3).
As regards nominal categories, a ‘(+)’ appears only in the column for
case. This occurs when a DC is marked for oblique case when it functions
as an adjunct, while in core argument function (i.e. as a subject or object
complement) and / or in the function of referential modifier (i.e. as a relative
clause), the case-marker is a zero-morpheme. Of course, the presence of a
zero-morpheme is assumed only when it fits into a larger paradigm of overt
case markers.
The coding of the arguments
As explained in Chapter 3, the parameter of argument coding has three

possible values:
(i) An argument is expressed as it would be in an independent clause

(SENT);
(ii) An argument is expressed in a different way than it would be in an
independent clause (ALT);
(iii) An argument is not expressed (Ø) 100.
A number of DC constructions display some variation in the realization

of their argument(s). Despite this variation, such DCs are viewed as single
constructions, as long as their structural coding remains unchanged. The
selection of an argument coding strategy may depend on the propositional
function in which the relevant DC appears (when it is a flexible construction)
and / or on the specific semantic / syntactic characteristics of the subordination
relation that it expresses. In section 6.3.3 I will discuss the repercussions of
variable argument coding for the classification of DC constructions in terms
of structural types.
100
For examples of each of the parameter values, see Chapter 3, section 3.4.

Keeping the remarks of the previous subsection in mind, consider the data
in Table 6.2 below. As in Table 6.1 above, the languages are listed in the
leftmost column. The DCs appear in the second column and are named
after their structural coding. The other columns display the behavioural
potential of every DC in terms of the three formal parameters:
(i) (non-)expression of verbal categories;

(ii) (non-)expression of nominal categories;
(iii) coding of the argument(s).
Verbal categories are abbreviated as follows: VV for voice and valency markers;
T, A, and M for tense, aspect and mood marking, respectively; P for person
marking, and IF for illocutionary force. Nominal categories are abbreviated
as DET for determiners and CASE for case markers and adpositions.
Arguments are simply labelled 1 and 2 for first and second argument.
Table 6.2: Basic data on behavioural potential of all DC constructions

Language Structural Behavioural potential
coding Verbal categories Nominal Argument
categories expression
w t a m p if det case 1 2
Tagalog pag- – – + ALT SENT
na/-ng + + + SENT SENT

Kharia RDP/Ø +/- – – (+) ALT SENT
/ALT
-na – – – (+) ALT/Ø SENT
-na-wala – – – Ø SENT
-al – – – ALT SENT
-ker etc. – – – Ø ALT
-ga – – – Ø SENT
-ta + RDP – – – Ø SENT
no + + + – SENT SENT
gam-kon + + + – SENT SENT
je etc. + + + SENT SENT
a/i etc. + + + SENT SENT
Ø + + +/- Ø/SENT Ø/SENT
Kambera pa- + – – +/- ALT/Ø Ø/SENT
ma- – – – + Ø SENT
Ø (NMLZ) +/- +/- + ALT SENT
wà + + + SENT SENT
Samoan =ga – – – + + ALT SENT
/ALT
DET – – – + + ALT SENT
(NMLZ)
ona/ina – – – – – SENT SENT
-e + + + + Ø/SENT Ø/SENT
Ø + + + SENT SENT
Guaraní há(gwe) + + +/- SENT SENT
/Ø
va + + Ø/SENT Ø/SENT
vo + + SENT SENT
Santali Ø (no +/- +/- +/- +/- +/- – (+) Ø/SENT Ø/SENT
INDIC)
-kate + – – – – Ø SENT
CORREL + + + + +/- +/- SENT SENT
mente + + + + + + SENT SENT

Warao -kitane – – Ø SENT
kotai + ? Ø/SENT Ø/SENT
Quechua -j/-shka/ (+) + – +/- Ø/SENT Ø/SENT
-na
-ngapaj – + – – Ø SENT
-chun – + – – SENT SENT
-y – – – +/- Ø SENT
-shpa – – – – Ø SENT
Turkish -DIK/ (+) – – – +/- ALT SENT
-(y)AcAK
-mAK – – – – +/- Ø SENT
-mA – – – – + ALT SENT
-An – – – – – Ø SENT
-(y)ArAk – – – – – Ø SENT
-(y)A… – – – – – Ø SENT
-(y)A
ki + + + + – Ø/SENT SENT
diye + + + + – SENT SENT
Ø + + + + – SENT SENT
Kayardild -n- – – – ALT/Ø ALT
-Thirri-n – – – ALT SENT
-n-garrba – – – Ø/SENT Ø/SENT
-ntha + + + SENT SENT
Ø + + + Ø/SENT SENT
Paiwan tu(a)/tjai + + SENT SENT
a + + Ø/SENT Ø/SENT
a parhu + + SENT SENT
(-)in- […]- – (+) Ø Ø
an + a

Ma'di -lɛ̄ ALT/Ø Ø/SENT
-dƷɔ ALT/Ø SENT
-ka ALT/Ø SENT
-rɛ̄/-ɓá Ø SENT
Ø SENT SENT
-zɨ̄ + sɨ̀ – Ø SENT
Gooni- -woo – + – – + Ø SENT
yandi -wadda – + – – – Ø SENT
-mawoo – + – – – Ø SENT
-bari – + – – – Ø SENT
-ya/ – + – – – Ø SENT
-gowaaya
Ø (+RSP) + + + + – Ø/SENT Ø/SENT
Hun- -ni + – + – +/- – ALT/Ø SENT
garian -ás/-és – – – – + ALT ALT
-ó (+) (+) – – Ø SENT
-ótt (+) (+) – – ALT/ Ø
Ø
-andó/ – (+) – – Ø/SENT Ø/SENT
-endó
-vá/-vé – (+) – – Ø SENT
-vén – (+) – – Ø SENT
hogy + + + + +/- SENT SENT
(etc.)
úgy (a-) + + + + + SENT SENT
hogy
REL.PRON + + + + + SENT SENT
Japanese no/mono + + ALT/ SENT
etc. SENT
Ø (RSP) + ALT/ Ø/SENT
SENT/Ø
-te/-de/ – Ø SENT
-ite
-i/-Ø – Ø SENT

Hmong qhov SENT SENT
Njua kuam/ SENT SENT
(has)tas
Ø SENT SENT
kws Ø/SENT Ø/SENT
Lango -(kk)ɔ̀ + – – ALT/Ø SENT
nî + + + SENT SENT
à-mɛ̀ + + Ø/SENT Ø/SENT
Ket Ø (bare – – – +/- ALT/Ø ALT
INF)
Ø + + + SENT SENT
ɛta qɔr’a + + + SENT SENT
ásqà + + + SENT SENT
-PROSEC + + + SENT SENT
REL.PRON + + + Ø/SENT Ø/SENT
-s/-bes + + + SENT SENT
Itelmen INF – – – – – Ø SENT
(various
forms)
Ø + + + + + SENT SENT
min + + + + + Ø/SENT Ø/SENT
quatz + + + + + SENT SENT
Koasati NMLZ + – – – – – – Ø SENT
(various
forms)
-:sáya (+) – – (+) – + Ø/SENT Ø/SENT
-:yólli – (+) – (+) – + Ø/SENT Ø/SENT
-:ka (+) – – (+) – + Ø/SENT Ø/SENT
-:kítta – (+) – (+) – + Ø/SENT Ø/SENT
-laho: (+) – – (+) – + Ø/SENT Ø/SENT
li:sáya
-n – – – – – – SENT SENT
-k – – – – – – Ø SENT
-t – – – – – – Ø SENT

Thai Ø Ø/SENT Ø/SENT
thiî Ø/SENT Ø/SENT
wâa SENT SENT
Basque -t(z)e – – – + SENT/Ø SENT/Ø
-tu/-du/ – – – +/- (+) SENT/Ø SENT/Ø
-i/- Ø
-en + + + +/- SENT/Ø SENT/Ø
-en bezala + + + – – SENT SENT
-(e)la + + + – (+) SENT/Ø SENT
bait + + + – Ø/SENT Ø/SENT
Abun do/Ø Ø/SENT SENT
gato Ø/SENT Ø/SENT
sa gato SENT SENT
Bambara ka SENT SENT
-le/-ne – – – Ø Ø
-min(u)/ Ø/SENT Ø/SENT
-mun(u)
-tò – – – SENT SENT
Georgian -a – + – – + ALT ALT
m- (-a-) – – – + Ø ALT
(-el/-al))
-ul/-il/m- – – – + ALT/Ø Ø
-ar/-al
sa- (-el/ + Ø Ø
-al/r)
rom + + + + – Ø/SENT Ø/SENT
ra + + + + – Ø/SENT SENT
rogorc + + + + – SENT SENT
Bukiyip Ø + + Ø/SENT SENT
(ú)li + + SENT SENT
bwidouk + + SENT SENT
-(u)mu

Abkhaz -ra – – – – – ALT ALT
-NFIN (+) (+) (+) (+) (+) SENT SENT
REL.PRON (+) (+) (+) (+) (+) Ø/SENT Ø/SENT
+ -NFIN
-ş-+ (+) (+) (+) (+) (+) SENT SENT
-NFIN
Polish -nie – – – – + ALT/Ø ALT/
SENT
INF – – – – – Ø SENT
(various
forms)
-c (+) (+) – – + Ø SENT
-any etc. (+) (+) – – – Ø Ø
(PST. (+) (+) – – – Ø Ø
PASS) PTC
(various
forms)
PRS.PL-c- – – – – Ø SENT
że + + + – SENT SENT
REL.PRON + + + + Ø/SENT Ø/SENT
jak (gdy) + + + SENT SENT
by
Buru- -(á)as + – + – – +/- Ø/SENT SENT
shaski -im/-um/ + – + – – (+) SENT SENT
-am
n-STEM + – + – – Ø SENT
-(a)n
ke/ki + + + + + SENT SENT
sén-/ét- + + + + + SENT SENT
Lavu- -e/-i – – – +/- +/- +/- Ø/SENT SENT
kaleve PERSON. + + + + + SENT SENT
SUFF
Alamblak -nef – – ALT/ ALT/
SENT SENT
-(kfë)t – – ALT ALT
ɨnd/Ø + + – Ø/SENT Ø/SENT

Pipil ka(h) + + SENT SENT
ne + + SENT SENT
ke + + SENT SENT
ke:n-aken + + SENT SENT
ADV + + SENT SENT
Wambon -e + + + + Ø/SENT SENT
-a (+ o) + + + + Ø/SENT Ø/SENT
Ø – – – Ø SENT
-mo/-o + – – – Ø SENT
ka + + + SENT SENT
Dhaa- DET + + + Ø/SENT Ø/SENT
sanac (+DEM)
-ɲ/-an – – + ALT/Ø ALT
Berbice fu/fi/Ø – – – Ø SENT
Dutch bifi/ + + + SENT SENT
dati/Ø
Wh/Ø + + + Ø/SENT Ø/SENT
Babungo lāa + SENT SENT
fáŋ/yúu + Ø/SENT Ø/SENT
kɨ̀(i)/Ø – Ø SENT
Nama !xáis-a +/- + – SENT SENT
(no
INDIC)
Ø (no + + – Ø/SENT Ø/SENT
INDIC)
-se/!’aa/ ? – – Ø SENT
tsii/Ø (no
INDIC)
Hdi ká (+) SENT SENT
tá – ALT SENT
tà (+) + SENT Ø
tà + (+) – Ø SENT
NMLZ
-a (+) + Ø SENT

Mandarin Ø Ø/SENT SENT
de Ø SENT
-zhe Ø SENT
Tamil -atu + +/- +/- – + SENT SENT
-(kk)a + – – Ø SENT
-a + + + – Ø/SENT Ø/SENT
Kisi (m)àà/Ø + + + SENT SENT
Ø – – – Ø SENT
CL + + + Ø/SENT Ø/SENT
Nung Ø Ø/SENT SENT
(tị-vạ) Ø/SENT Ø/SENT
(+DEM)
(+FOC)
bạt SENT SENT
Garo in-e + + SENT SENT
-a – – + ALT SENT
-a-ni – – + Ø Ø
-na – – – Ø SENT
-kan/ – – – Ø SENT
-kan-a
-gip-a – – ALT/Ø SENT/
Ø
-e/-e- – – – Ø SENT
min/-e-r
Krongo àní tíŋ + + + + + – SENT SENT
-(t) + + + – – +/- ALT/Ø SENT
m-/n- + + +/- – – ALT/Ø SENT
Hixkar- nɨ/-thɨ-/ +/- – – ALT ALT
yana -hɨto + rɨ
Slave nį/Ø + + + + + – – SENT SENT
gha/gú + + + + + – – SENT SENT
i/sį́i/líi + + + + + – – Ø/SENT Ø/SENT
gháré + + + + + – – SENT SENT

Nivkh -vut/-vur + + + + – SENT SENT
Ø + + + + Ø/SENT SENT
NFIN + + + – Ø/SENT Ø/SENT
r/-t-ř//-n – + + +/- – Ø SENT
Green- -niq + – – + ALT ALT/Ø
landic -ta/-sa (+) – – – + ALT Ø/SENT
PTC mood (+) (+) (+) (+) (+) Ø/SENT Ø/SENT
CONT + (+) (+) (+) (+) Ø SENT
mood
6.3.3 A three-way typology of DCs according to behavioural potential
Based on the behavioural potential of all DCs presented in Table 6.2 above,
these constructions can now be assigned to one of the three structural
types defined in Chapter 3. The properties of the three types are briefly
repeated below:
• Type 1: Balanced constructions (abbreviated as B)

- All TAM distinctions and Person are expressed as in independent
clauses;
- All overt arguments are coded as in independent clauses (SENT-
SENT);
- Determiners (DET) and/or case markers/adpositions (CASE) can
but need not be expressed.
• T ype 2: Deranked DCs with SENT arguments (Abbreviated as

D-SENT)
- (Partial) non-expression of TAM distinctions and / or Person
marking;
- All overt arguments are expressed as in independent clauses (SENT-
SENT);
- DET / CASE can but need not be expressed.

• Type 3: Deranked DCs with at least one alternatively coded argument
(Abbreviated as D-ALT)
- (Partial) non-expression of TAM distinctions and / or Person
marking;
- One argument is expressed with an ALT strategy; the other one, if
available / overt, is either SENT or ALT as well.
- DET / CASE are expressed (if the language has these categories).
Table 6.3 summarizes the formal characteristics of the three DC types

(this table is the same as Table 3.1 and Table 4.2):

number label 1 2
Type 1 B SENT / Ø SENT / Ø + +/-
Type 2 D-SENT SENT / Ø SENT / Ø – +/-
Before presenting the classification of the DCs in the sample languages (see
Table 6.5, in section 6.3.3.3) some preliminary considerations are required.
6.3.3.2 Preliminary considerations
Variable argument coding
As mentioned in the previous section, and as can be seen in Table 6.2,

many DC constructions display some variation in the coding of their
arguments, depending on their propositional function and / or on the specific
semantic / syntactic subordination relation that they express. Such cases of
variable argument coding come in several types.
First, there are complement constructions that can take the form of
either an infinitive or a nominalization. That is to say, the first argument of
such DCs is either unexpressed under co-referentiality or, when overt, it is
expressed with an ALT strategy. The second argument (if present) remains
SENT in both cases. This means that, in terms of the typology in Table 6.3
above, such a DC would classify sometimes as a type 2 construction (when
the first argument is Ø) and sometimes as a type 3 construction (when the
first argument is ALT). However, since the first argument is ALT whenever

it is overtly expressed, these DCs will be assigned to the group of type 3
(D-ALT) constructions. The examples in (50) illustrate this for the -ka
construction in Ma’di, which has an un-expressed first argument in (50a); a
possessive first argument in (50b); and an object argument that is expressed
as in an independent clause in (50c):
Ma’di (Blackings & Fabb 2003: 22, 21, 213)

(50) a. má ɓì [mū-kā] kʊ̄rʊ̀
1sg try (n)-go-nmlz neg(pst)
‘I have not tried / did not try to go / going.’
b. Má ndrɛ̄ [àrɨ́áŋgwá rɨ ̀ Ɂà ɛ̄-ɗɛ́-kā] rá

1sg see bird def poss (n)-ve-fall-nmlz aff
‘I saw the bird’s falling / fall.’
c. ɔ́-lɛ̀ [tʊ́bà sɛ̄-kā] ɗɨ Ɂ́ ā kʊ̄

ind-want cigarette (n)-smoke-nmlz here neg
‘Smoking is not permitted here.’
Note that this type of DC stands in contrast to constructions that are

invariably infinitival. In the latter construction type the first argument always
remains unexpressed, and therefore it is classified as type 2 (D-SENT).
A second type of DC with variable argument coding involves deranked
relative clause constructions with a gap strategy: When the first argument
of such a DC is gapped, the second argument (if present) remains SENT.
However, when the second argument is gapped, the first argument is
expressed with an ALT strategy. As in the previous case, since the first
argument gets ALT coding whenever it is expressed overtly, this type of
DCs is classified as type 3 (D-ALT). An example from Garo is given in (51),
which shows the alternation of ALT expression (51a) and zero-expression
(gapping) of the subject argument (51b):
Garo (Burling 2004: 301, 299)

(51) a. [me͘ chik-ni skang-o den͘ -gip-a] a͘ -bol
women-gen previously-loc cut-ptc firewood
‘firewood that the women chopped previously’

b. [nok-o pi͘ -sa-ko nik-gip-a] me-tra
‘the woman who saw the child at the house’
Note that there are also deranked relative clauses with gapping that do not
have the possibility of ALT coding of the first argument. Rather, whenever
an argument is overt (i.e. not gapped) in such DCs, it gets SENT expression.
In addition, when the relativized item is another argument than the subject
or object, both core arguments are sententially expressed. Therefore, these
DCs are classified as type 2: D-SENT 101. This construction type is attested
in Burushaski and Tamil. Examples from Tamil appear in (52), which show
a relative clause with a gapped subject (52a), a gapped object (52b), and a
gapped instrumental adjunct (52c):

(52) a. [vaɳɳaane aticc-a] taccan
carpenter-acc beat.pst-ptc washerman
‘the washerman who beat the carpenter’
b. [taccan aticc-a] vaɳɳaan

carpenter beat.pst-ptc washerman
‘the washerman whom the carpenter beat’
c. [akkaa taŋkaccikki caata pooʈʈ-a] karanʈi

elder.sister younger.sister-dat rice put.pst-ptc spoon
‘The spoon with which elder sister gave rice to younger sister’
A third case of variable argument coding is found with certain types of

nominal clauses, i.e. DCs that can be used as complement clauses and relative
clauses. In the former function, the first argument of the DC must remain
unexpressed (because of co-referentiality with the matrix clause subject),
while the second argument is SENT. In the latter function, however, this
type of construction expresses object relative clauses and employs a gap
strategy. The first argument then gets ALT coding and the second remains
unexpressed. Thus, there are two possible argument realizations depending
101
Of course, the same classification applies to relative clause construction in which the
subject is obligatorily the relativized and gapped argument (and in which the second
argument is SENT).

on the propositional function in which the DC is used: Ø-SENT in
referential function and ALT-Ø in modifier function. Since the zero-
coding of the subject argument is obligatory in referential function, this
construction type receives a double classification in terms of behavioural
potential: type 2 (D-SENT) / type 3 (D-ALT). There are only three such
cases attested, namely the pa-construction in Kambera, which is illustrated
in (53a-b), and the lɛ́- and dƷɔ́-constructions in Ma’di, illustrated in (54a-b)
and (55a-b), respectively. In all pairs of examples the first one illustrates the
complement clause function and the second the relative clause function
of the DC. Note that these DC constructions, even though they receive
a double classification in terms of behavioural potential, are nonetheless
regarded as flexible constructions, since their structural coding is the same
in both functions.

(53) a. Ta-pakiring [pa-tinu-nya na lau]
1pl.nom-start comp-weave-3sg.dat art sarong
haromu
tomorrow
We will start to weave the sarong tomorrow.’
b. na kalembi na [pa-kei wà-nggu-nya]

art shirt art rel-buy use-1sg.gen-3sg.dat
‘the shirt that I bought’
Ma’di (Blackings & Fabb 2003: 202, 22, 207, 206)

(54) a. Má lɛ̀-ā [èɓí ˋ ɲā-lɛ́] rá
1sg (n)want-obj fish n-eat-nmlz aff
‘I certainly want to eat fish.’
b. àràbɨ́à [ɔ́pɨ́ Ɂà dƷɨ̄-lɛ́] rɨ̀ pá nā ādī rá.

car Opi poss (n)-take-ptc def leg afr deflate aff
‘The car which Opi took certainly has a flat tyre.’
(55) a. ɔ́pɨ́ ɛ̄ɗɔ́ [ˋsī-dƷɔ́] rá

Opi start n-build-nmlz aff
‘Opi has certainly started to build it.’

b. àdƷú [má-à ˋɗï̄ -Ʒɔ́] rɨ̀ Ɂɨ̄ ēgwè dì
spear 1sg-poss n-build-ptc def foc lose com
‘The spear with which I killed it is lost.’
There is one more flexible DC construction, in Krongo, with a double

classification Type 2 / 3, but it is a flexible modifier clause construction,
rather than a nominal clause construction. It has already been illustrated
in (35a-b) above. These examples show that the construction has an ALT-
coded subject when it is used as a relative clause (35a), while in adverbial
function (35b) the subject always remains unexpressed, because it must be
co-referential with the subject of the matrix clause.
An isolated case of double classification concerns Japanese
nominalizations. These constructions are balanced to the extent that
they retain tense marking, but deranked in the sense that they can, albeit
optionally, take a possessive subject argument, as is shown in (56). Because
of this contradictory evidence, the construction is classified as type 1 / 3
(B / D-ALT).
Japanese (Lombardi Vallauri 1997: 497; Alpatov & Podlesskaya 1995: 468)
(56) [Ano hito ga / no hon o kai-ta koto]
that person nom / gen book acc write-past nmlz
ga yoku sira-re-te iru
nom well know-pass-ger be
‘It is well known that that person wrote a book.’
Finally, consider the so-called ‘participial mood’ construction in West

Greenlandic. As Fortescue (1984: 49) notes, this construction is “fully
inflectible” when used as a complement clause, while it lacks person marking
when used as a relative clause, as is illustrated in (57a) and (57b), respectively.
Although this DC is thus not fully finite in both functions, it is nevertheless
classified as a type 1 construction.
West-Greenlandic (Fortescue 1984: 36, 49)

(57) a. Ilisima-vaa [urni-ssa-giga]
know-3sg.3sg.ind come.to-fut-1sg.3sg.ptc
‘He1 knew I would come to him2.’

b. Niviarsiaq [kalaallisut ilinnia-lir-suq]
girl Greenlandic learn-begin-intr.ptc
‘the / a girl who has begun learning Greenlandic.’
The balancing / deranking distinction in languages with little or no

verbal inflection.
Not surprisingly, there are some cases in which the decision to classify a
specific DC as a balanced or a deranked construction is empirically rather
vacuous, as a result of the (near) absence of verbal inflectional categories in
certain languages. This holds most obviously for isolating languages. For
instance in Thai, the construction in (58) is classified as balanced by default;
since there are no verbal inflectional categories that can be lost (and since
no nominal categories are acquired), there is no empirical basis on which to
classify the construction as deranked.

(58) Khon [thîi duulɛɛ] nî pen pen acaan lə̌
person rel take.care prt cop cop teacher interr
‘Is the person who takes care [of the students] a teacher?’
A similar situation obtains in other languages with little verbal morphology.

Consider for instance Ma’di: As regards inflectional morphology, verbs in
this language can only take a low-tone prefix, which in independent clauses
expresses non-past tense. Dependent predicates, which are marked with one
of the subordinating suffixes kā, lɛ́, rɛ̄, ɓá, or dʒɔ́, take a homophonous prefix.
However, according to Blackings and Fabb (2003: 192) it is not really clear
whether this is the same prefix as in independent clauses. An argument
against identity of the two forms is that in dependent clauses the prefix
is compatible with any tense interpretation. Moreover, when the subject
in DCs with the subordinating suffixes lɛ́, dʒɔ́, and kā is overt, then it is
coded as a possessor. On the basis of this evidence I classify these Ma’di
constructions as deranked.
Keeping the above considerations in mind, the three-way DC classification

on the basis of behavioural potential is represented in Table 6.4:

Table 6.4: Three-way classification of DCs
Language Structural coding DC type
Tagalog pag- 3 D-ALT
na / -ng 1 B
Kharia RDP / Ø 3 D-ALT
-na 3 D-ALT
-na-wala 2 D-SENT
-al 3 D-ALT
-ker etc. 3 Ø-ALT
-ga 2 D-SENT
-ta + RDP 2 D-SENT
no 1 B
gam-kon 1 B
je etc. 1 B
a / i etc. 1 B
Ø 1 B
Kambera pa- 2/3 D-SENT / D-ALT
ma- 2 D-SENT
Ø (NMLZ) 3 D-ALT
wà 1 B
Samoan =ga 3 D-ALT
Ø (NMLZ) 3 D-ALT
ona / ina 2 D-SENT
-e 1 B
Ø 1 B
Guaraní há(gwe) / Ø 1 B
va 1 B
vo 1 B
Santali Ø (no INDIC) 2 D-SENT
-kate 2 D-SENT
CORREL 1 B
mente 1 B
Warao -kitane 2 D-SENT
kotai 1 B

Imbabura -j / -shka / -na 2 D-SENT
Quechua -ngapaj 2 D-SENT
-chun 2 D-SENT
-y 2 D-SENT
-shpa 2 D-SENT
Turkish -DIK / -(y)AcAK 3 D-ALT
-mAK 2 D-SENT
-mA 3 D-ALT
-An 2 D-SENT
-(y)ArAk 2 D-SENT
-(y)A…-(y)A 2 D-SENT
ki 1 B
diye 1 B
Ø 1 B
Kayardild -n- 3 D-ALT
-Thirri-n 3 D-ALT
-n-garrba 2 D-SENT
-ntha 1 B
Ø 1 B
Paiwan tu(a) / tjai 1 B
a 1 B
a parhu 1 B
(-)in- -an + a 2 D-SENT
Ma’di -lɛ̄ 2/3 D-SENT / D-ALT
-dƷɔ 2/3 D-SENT / D-ALT
-ka 3 D-ALT
-rɛ̄ / -ɓá 2 D-SENT
Ø 1 B
-zɨ̄ + sɨ̀ 2 D-SENT
Gooniyandi -woo 2 D-SENT
-wadda 2 D-SENT
-mawoo 2 D-SENT
-bari 2 D-SENT
-ya / -gowaaya 2 D-SENT
Ø (+RSP) 1 B

Hungarian -ni 3 D-ALT
-ás / -és 3 D-ALT
-ó 2 D-SENT
-ótt 3 D-ALT
-andó / -endó 2 D-SENT
-vá / -vé 2 D-SENT
-vén 2 D-SENT
(azt) hogy (etc.) 1 B
ugy (a-)hogy 1 B
REL.PRON 1 B
Japanese no / mono etc. 1/3 B / D-ALT
Ø (RSP) 1 B / D-ALT
-te / -de / -ite 2 D-SENT
-i / -Ø 2 D-SENT
Hmong Njua qhov 1 B
kuam / (has)tas 1 B
Ø 1 B
kws 1 B
Lango -(kk)ɔ̀ 3 D-ALT
nî 1 B
à-mɛ̀ 1 B
Ket Ø (bare INF) 3 D-ALT
Ø 1 B
ɛta qɔr’a 1 B
ásqà 1 B
-PROSEC 1 B
REL.PRON 1 B
-s / -bes 1 B
Itelmen INF (various forms) 2 D-SENT
Ø 1 B
min 1 B
quatz 1 B

Koasati NMLZ (various forms) 2 D-SENT
-:sáya 2 D-SENT
-:yólli 2 D-SENT
-:ka 2 D-SENT
-:kítta 2 D-SENT
-laho:li:sáya 2 D-SENT
-n 2 D-SENT
-k 2 D-SENT
-t 2 D-SENT
Thai Ø 1 B
thiî 1 B
wâa 1 B
Basque -t(z)e 2 D-SENT
-tu / -du / -i / -Ø 2 D SENT
-en 1 B
-en bezala 1 B
-(e)la 1 B
bait 1 B
Abun do / Ø 1 B
gato 1 B
sa gato 1 B
Bambara ka 1 B
-le / -ne 2 D-SENT
-min(u) / -mun(u) 1 B
-tò 2 D-SENT
Georgian -a 3 D-ALT
m- (-a-)(-el / -al)) 3 D-ALT
-ul / -il / m- -ar / -al 3 D-ALT
sa- (-el / -al / r) 2 D-SENT
rom 1 B
REL.PRON 1 B
ra 1 B
rogorc 1 B
Bukiyip Ø 1 B
(ú)li 1 B
bwidouk -(u)mu 1 B

Abkhaz -ra 3 D-ALT
-NFIN 1 B
REL.PRON + -NFIN 1 B
-ş- + -NFIN 1 B
Polish -nie 3 D-ALT
INF (various forms) 2 D-SENT
-c 2 D-SENT
-any etc. 2 D-SENT
(PST.PASS) PTC (various forms) 2 D-SENT
PRS.PL-c- 2 D-SENT
że 1 B
REL.PRON 1 B
jak (gdy)by 1 B
Burushaski -(á)as 2 D-SENT
-im / -um / -am 2 D-SENT
n-STEM-(a)n 2 D-SENT
ke / ki 1 B
sén- / ét- 1 B
Lavukaleve -e / -i 2 D-SENT
PERSON.SUFF 1 B
Alamblak -nef 3 D-ALT
-(kfë)t 3 D-ALT
ɨnd / Ø 2 D-SENT
Pipil ka(h) 1 B
ne 1 B
ke 1 B
ke:n-aken 1 B
ADV 1 B
Wambon -e 1 B
-a (+ o) 1 B
Ø 2 D-SENT
-mo / -o 2 D-SENT
ka 1 B
Dhaasanac DET(+DEM) 1 B
-ɲ / -an 3 D-ALT

Berbice Dutch fu / fi / Ø 2 D-SENT
Creole bifi / dati / Ø 1 B
Wh / Ø 1 B
Babungo lāa 1 B
fáŋ / yúu 1 B
kɨ̀(i) / Ø 2 D-SENT
Nama !xáis-à (no INDIC) 1 B
Ø (no INDIC) 1 B
-se / !’aa / tsii (no INDIC) 2 D-SENT
Hdi ká 1 B
tá 3 D-ALT
tà 1 B
tà + NMLZ 2 D-SENT
-a 1 B
Mandarin Ø 1 B
Chinese de 1 B
-zhe 2 D-SENT
Tamil -atu 2 D-SENT
-(kk)a 2 D-SENT
-a 2 D-SENT
Kisi (m)àà / Ø 1 B
Ø 2 D-SENT
CL 1 B
Nung Ø 1 B
(tị-vạ) (+DEM) (+FOC) 1 B
bạt 1 B
Garo in-e 1 B
-a 3 D-ALT
-a-ni 2 D-SENT
-na 2 D-SENT
-kan / -kan-a 2 D-SENT
-gip-a 3 D-ALT
-e / -e-min / -e-r 2 D-SENT
Krongo àní tíŋ 1 B
-(t) 3 D-ALT
m- / n- 2/3 D-SENT / D-ALT

Hixkaryana nɨ / -thɨ- / -hɨto + rɨ 3 D-ALT
Slave nį / Ø 1 B
gha / gú 1 B
i / sį́i / líi 1 B
gháré 1 B
Nivkh -vut / -vur 1 B
Ø 1 B
NFIN 1 B
r / -t-ř / / -n 2 D-SENT
West -niq 3 D-ALT
Greenlandic -ta / -sa 3 D-ALT
PTC mood 1 B
CONT mood 1 B
6.3.4 Behavioural potential of different structural DC types

In this section, the data in Table 6.2 on the behavioural potential of the
DCs in the sample languages are re-ordered in the form of three separate
tables, each of which corresponds to one of the three structural DC types. In
particular, Table 6.5 presents the behavioural potential of all type 1 balanced
DCs; Table 6.6 presents the behavioural potential of all type 2 D-SENT
DCs; and Table 6.7 presents the behavioural potential of all type 3 D-ALT
DCs. Note that those DCs that received double classifications in Table 6.4
above, appear in two of the three tables below.
Table 6.5: Behavioural potential of Type 1 balanced DCs

Coding Verbal categories Nominal Argument
Tagalog na/-ng + + SENT SENT
Kharia no + + + – SENT SENT
gam-kon + + + – SENT SENT
je etc. + + + SENT SENT
a/i etc. + + + SENT SENT
Ø + + +/- Ø/SENT Ø/SENT
Kambera wà + + + SENT SENT

Samoan -e + + + + Ø/SENT Ø/SENT
Ø + + + SENT SENT
Guaraní há(gwe)/ + + +/- SENT SENT
Ø
va + + Ø/SENT Ø/SENT
vo + + SENT SENT
Santali CORREL + + + + +/- +/- SENT SENT
mente + + + + + + SENT SENT
Warao kotai + ? Ø/SENT Ø/SENT
Turkish ki + + + + – Ø/SENT SENT
diye + + + + – SENT SENT
Ø + + + + – SENT SENT
Kayardild -ntha + + + SENT SENT
Ø + + + Ø/SENT SENT
Paiwan tu(a)/tjai + + SENT SENT
a + + Ø/SENT Ø/SENT
a parhu + + SENT SENT
Ma'di Ø SENT SENT
Goon- Ø (+RSP) + + + + – Ø/SENT Ø/SENT
iyandi
Hun- hogy (etc.) + + + + +/- SENT SENT
garian REL.PRON + + + + + SENT SENT
ugy (a-) SENT SENT
hogy
etc. SENT
Ø (RSP) + SENT/Ø SENT/Ø
Hmong qhov SENT SENT
Njua kuam/ SENT SENT
(has)tas
Ø SENT SENT
kws Ø/SENT Ø/SENT
Lango nî + + + SENT SENT
à-mɛ̀ + + Ø/SENT Ø/SENT

Ket Ø + + + SENT SENT
ɛta qɔr’a + + + SENT SENT
ásqà + + + SENT SENT
-PROSEC + + + SENT SENT
REL.PRON + + + Ø/SENT Ø/SENT
-s/-bes + + + SENT SENT
Itelmen Ø + + + + + SENT SENT
min + + + + + Ø/SENT Ø/SENT
quatz + + + + + SENT SENT
Thai Ø Ø/SENT Ø/SENT
thiî Ø/SENT Ø/SENT
wâa SENT SENT
Basque -en + + + +/- Ø/SENT Ø/SENT
-en bezala + + + SENT SENT
-(e)la + + + – (+) Ø/SENT SENT
bait + + + – Ø/SENT Ø/SENT
Abun do/Ø Ø/SENT SENT
gato Ø/SENT Ø/SENT
sa gato SENT SENT
Bambara ka SENT SENT
-min(u)/ Ø/SENT Ø/SENT
-mun(u)
Georgian rom + + + + – Ø/SENT Ø/SENT
ra + + + + – Ø/SENT SENT
rogorc + + + + – SENT SENT
Bukiyip Ø + + Ø/SENT SENT
(ú)li + + SENT SENT
bwidouk + + SENT SENT
-(u)mu
Abkhaz -NFIN + + + + + SENT SENT
REL.PRON + + + + + Ø/SENT Ø/SENT
+ -NFIN
-ş- + NFIN + + + + + SENT SENT

Polish że + + + – SENT SENT
REL.PRON + + + + Ø/SENT Ø/SENT
jak (gdy) + + + SENT SENT
by
Burus- ke/ki + + + + + SENT SENT
haski sén-/ét- + + + + + SENT SENT
Lavu- PERSON. + + + + + SENT SENT
kaleve SUFF
Pipil ka(h) + + SENT SENT
ne + + SENT SENT
ke + + SENT SENT
ke:n-aken + + SENT SENT
ADV + + SENT SENT
Wambon -e + + + + Ø/SENT SENT
-a (+ o) + + + + Ø/SENT Ø/SENT
ka + + + SENT SENT
Dhaa- DET + + + Ø/SENT Ø/SENT
sanac (+DEM)
Berbice bifi/dati/ + + + SENT SENT
Dutch Ø
Wh/Ø + + + Ø/SENT Ø/SENT
Babungo lāa + SENT SENT
fáŋ/yúu + Ø/SENT Ø/SENT
Nama !xáis-à (no + + – SENT SENT
INDIC)
Ø (no + + – Ø/SENT Ø/SENT
INDIC)
Hdi ká (+) SENT SENT
tà (+) + SENT Ø
-a (+) + Ø SENT
Mandarin Ø Ø/SENT SENT
Chinese de Ø SENT
Kisi (m)àà/Ø + + + SENT SENT
CL + + + Ø/SENT Ø/SENT

Nung Ø Ø/SENT SENT
(tị-vạ) Ø/SENT Ø/SENT
(+DEM)
(+FOC)
bạt SENT SENT
Garo in-e + + SENT SENT
Krongo àní tíŋ + + + + + – SENT SENT
Slave nį/Ø + + + + + – – SENT SENT
gha/gú + + + + + – – SENT SENT
i/sį́i/líi + + + + + – – SENT/Ø SENT/Ø
gháré + + + + + – – SENT SENT
Nivkh -vut/-vur + + + + – SENT SENT
Ø + + + + Ø/SENT SENT
NFIN + + + – Ø/SENT Ø/SENT
West PTC mood (+) (+) (+) (+) (+) SENT/Ø SENT/Ø
Green- CONT + (+) (+) (+) (+) Ø SENT
landic mood
Table 6.6: Behavioural potential of type 2 D-SENT DCs

Kharia -na-wala – – – Ø SENT
-ga – – – Ø SENT
-ta + RDP – – – Ø SENT
Kambera pa- + – – +/- Ø/ALT SENT/Ø
ma- – – – + Ø SENT
Samoan ona/ina – – – – – SENT SENT
Santali Ø (no +/- +/- +/- +/- +/- – (+) Ø/SENT Ø/SENT
INDIC)
-kate + – – – – Ø SENT
Warao -kitane – – Ø SENT

Quechua -j/-shka/ (+) + – +/- Ø/SENT Ø/SENT
-na
-chun – + – – SENT SENT
-ngapaj – + – – Ø SENT
-y – – – +/- Ø SENT
-shpa – – – – Ø SENT
Kayardild -n-garrba – – – SENT SENT
Paiwan (-)in- -an – (+) Ø Ø
+a
Turkish -mAK – – – – +/- Ø SENT
-An – – – – – Ø SENT
-(y)ArAk – – – – – Ø SENT
-(y)A… – – – – – Ø SENT
-(y)A
Ma'di -lɛ̄ Ø/ALT Ø/SENT
-dƷɔ Ø/ALT SENT
-rɛ̄/-ɓá Ø SENT
-zɨ̄ + sɨ̀ – Ø SENT
Gooni- -woo – + – – + Ø SENT
yandi -wadda – + – – – Ø SENT
-mawoo – + – – – Ø SENT
-bari – + – – – Ø SENT
-ya/ – + – – – Ø SENT
-gowaaya
Hun- -ó (+) (+) – – Ø SENT
garian -andó/ – (+) – – Ø/SENT Ø/SENT
-endó
-vá/-vé – (+) – – Ø SENT
-vén – (+) – – Ø SENT
Japanese -te/-de/ – Ø SENT
-ite
-i/-Ø – Ø SENT
Itelmen INF – – – – – Ø SENT
(various
forms)

Koasati NMLZ + – – – – – – Ø SENT
(various
forms)
-:sáya (+) – – (+) – + Ø/SENT Ø/SENT
-:yólli – (+) – (+) – + Ø/SENT Ø/SENT
-:ka (+) – – (+) – + Ø/SENT Ø/SENT
-:kítta – (+) – (+) – + Ø/SENT Ø/SENT
-laho:li: (+) – – (+) – + Ø/SENT Ø/SENT
sáya
-n – – – – – – SENT SENT
-k – – – – – – Ø SENT
-t – – – – – – Ø SENT
Basque -t(z)e – – – + SENT SENT
-tu/-du/ – – – +/- (+) Ø/SENT Ø/SENT
-i/-Ø
Bambara -le/-ne – – – Ø/SENT
-tò – – – SENT SENT
Georgian sa- (-el/ + Ø Ø
-al/r)
Polish INF – – – – – Ø SENT
(various
forms)
-c (+) (+) – – + Ø SENT
-any etc. (+) (+) – – – Ø Ø
(PST. (+) (+) – – – Ø Ø
PASS) PTC
(various
forms)
PRS.PL-c- – – – – Ø SENT
Buru– -(á)as + – + – – +/- Ø/SENT SENT
shaski -im/-um/ + – + – – (+) SENT SENT
-am
n-STEM + – + – – Ø SENT
-(a)n
Lavu- -e/-i – – – – +/- +/- Ø/SENT SENT
kaleve

Alamblak ɨnd/Ø + + – Ø/SENT Ø/SENT
Wambon Ø – – – Ø SENT
-mo/-o + – – – Ø SENT
Berbice fu/fi/Ø – – – Ø SENT
Dutch
Babungo kɨ̀(i)/Ø – Ø SENT
Nama -se/!’aa/ +/- +/- – Ø SENT
tsii/
Hdi tà + (+) – Ø SENT
NMLZ
Mandarin -zhe Ø SENT
C.
Tamil -atu + +/- +/- – + SENT SENT
-(kk)a + – – Ø SENT
-a + + + – Ø/SENT Ø/SENT
Kisi Ø – – – Ø SENT
Garo -a-ni – – + Ø Ø
-na – – – Ø SENT
-kan/ – – – Ø SENT
-kan-a
-e/-e-min – – – Ø SENT
/-e-r
Krongo m-/n- + + +/- – – Ø/ALT SENT
Nivkh r/-t-ř//-n – + + +/- – Ø SENT

Table 6.7: Behavioural potential of type 3 D-ALT DCs
Tagalog pag- – – + ALT SENT
Kharia RDP/Ø +/- – – + ALT SENT
/ALT
-na – – – + ALT SENT
-al – – – ALT SENT
-ker etc. – – – Ø ALT
Kambera pa- + – – +/- (+) Ø/ALT SENT/Ø
Ø (NMLZ) +/- +/- + ALT SENT
Samoan =ga – – – + + ALT SENT/
ALT
Ø (NMLZ) – – – + + ALT SENT
Turkish -DIK/ (+) – – – +/- ALT SENT
-(y)AcAK
-mA – – – – + ALT SENT
Kayardild -n- – – – ALT ALT
-Thirri-n – – – ALT SENT
Ma'di -lɛ̄ Ø/ALT SENT/Ø
(OBL)
-dƷɔ Ø/ALT SENT
-ka ALT SENT
Hun- -ás/-és – – – – + ALT ALT
garian -ni + – + – +/- – Ø/ALT SENT
-ótt (+) (+) – – ALT Ø
etc. SENT
Ø (RSP) + ALT/ SENT/Ø
SENT/Ø
Lango -(kk)ɔ̀ + – – Ø/ALT SENT
Ket Ø (bare – – – +/- Ø/ALT ALT
INF)

Georgian -a – + – – + ALT ALT
m- (-a-) – – – + Ø ALT
(-el/-al))
-ul/-il/m- – – – + ALT/Ø Ø/ALT
-ar/-al
Abkhaz -ra – – – – – ALT ALT
Polish -nie – – – – + Ø/ALT ALT/
SENT
Alamblak -nef – – ALT/ ALT/
SENT SENT
-(kfë)t – – ALT ALT
Dhaa- -ɲ/-an – – + Ø/ALT ALT
sanac
Hdi tá – ALT SENT
Garo -a – – + ALT SENT
-gip-a – – ALT/Ø SENT/Ø
Krongo -(t) + + + – – (+) ALT/Ø SENT
m-/n- + + +/- – – ALT/Ø SENT
Hixka- nɨ/-thɨ-/ +/- – – ALT ALT
ryana -hɨto + rɨ
W. Green- -niq + – – + ALT ALT
landic -ta/-sa (+) – – – + Ø/ALT Ø/SENT
6.4 An integrated DC typology: Functional possibilities of

three structural DC types
In this section, the typologies developed in sections 6.2 and 6.3 are integrated.
Below, the functional possibilities of the DCs in the sample languages are
presented in three separate tables, one for each of the structural DC types.
The functional distribution of type 1 balanced clauses is presented in Table
6.8; the functional distribution of type 2 D-SENT DCs in Table 6.9; and
the functional distribution of type 3 D-ALT DCs in Table 6.10.

Table 6.8: Functional possibilities of type 1 balanced DCs
Language Structural Coding Functional possibilities
Pred Ref Ref Pred
Head Head Mod Mod
Tagalog na/-ng + + +
Kharia no +
gam-kon +
je etc. +
a/i etc. +
Ø +
Kambera wà +
Samoan -e +
Ø + +
Guaraní há(gew)/Ø +
va +
vo +
Santali CORREL +
mente +
Warao kotai +
Turkish ki + +
diye +
Ø +
Kayardild -ntha + +
Ø +
Paiwan tu(a)/tjai +
a + +
a parhu +
Ma’di Ø +
Gooniyandi Ø (+RSP) + + +
Hungarian hogy (etc.) +
REL.PRON +
úgy (-a-hogy) +
Ø (RSP) +
Hmong Njua qhov +
kuam/(has)tas +
Ø +
kws +

Pred Ref Ref Pred
Head Head Mod Mod
Lango nî +
à-mɛ̀ +
Ket Ø + +
ɛta qɔr’a +
ásqà +
-PROSEC
REL.PRON +
-s/-bes +
Itelmen Ø +
min +
quatz +
Thai Ø + +
thiî + +
wâa +
Basque -en + +
-en bezala +
-(e)la + (+)
bait + +
Abun do/Ø +
gato +
sa gato +
Bambara ka +
-min(u)/-mun(u) +
Georgian rom + +
REL.PRON +
ra +
rogorc +
Bukiyip Ø +
(ú)li +
bwidouk -(u)mu +
Polish że +
REL.PRON +
jak (gdy)by +

Pred Ref Ref Pred
Head Head Mod Mod
Abkhaz -NFIN +
REL.PRON + -NFIN +
-ş- + -NFIN +
Burushaski ke/ki + + +
sén-/ét- +
Lavukaleve PERSON.SUFF +
Pipil ka(h) + +
ne +
ke +
ke:n-aken +
ADV +
Wambon -e +
-a (+ o) +
ka +
Dhaasanac DET(+DEM) + +
Berbice Dutch bifi/dati/Ø +
Wh/Ø +
Babungo lāa +
fáŋ/yúu + +
Nama !xáis-à (no INDIC) +
Ø (no INDIC) (RSP) +
Mandarin Chinese Ø + + +
de + +
Kisi (m)àà/Ø +
CL +
Nung Ø + + +
(tị-vạ) (+DEM) (+FOC) +
bạ t +
Garo in-e +
Krongo àní tíŋ +
Slave nį/Ø + +
gha/gú +
i/sį́i/líi +
gháré +

Pred Ref Ref Pred
Head Head Mod Mod
Nivkh -vut/-vur +
Ø +
NFIN +
West Greenlandic PTC mood + +
CONT mood + +
Table 6.9: Functional possibilities of type 2 D-SENT DCs

Pred Ref Ref Pred
Head Head Mod Mod
Kharia -na-wala +
-ga +
-ta + RDP +
Kambera pa- + +
ma- +
Samoan ona/ina +
Santali Ø (no INDIC) + +
-kate +
Warao -kitane +
Quechua -j/-shka/-na + +
-chun +
-ngapaj +
-y + RDP
-shpa +
Kayardild -n-garrba +
Paiwan (-)in- -an + a +
Turkish -mAK +
-An +
-(y)ArAk +
-(y)A…-(y)A +
Ma'di -lɛ̄ + +
-dƷɔ + +
-rɛ̄/-ɓá +
-zɨ̄ + sɨ̀ +

Pred Ref Ref Pred
Head Head Mod Mod
Gooniyandi -woo +
-wadda +
-mawoo +
-bari +
-ya/-gowaaya +
Hungarian -ó +
-andó/-endó +
-vá/-vé +
-vén +
Japanese -te/-de/-ite +
-i/-Ø +
Itelmen INF (various forms) +
Koasati NMLZ (various forms) +
-:sáya +
-:yólli +
-:ka +
-:kítta +
-laho:li:sáya +
-n +
-k +
-t +
Basque -t(z)e +
-tu/-du/-i/- Ø +
Bambara -le/-ne +
-tò +
Georgian sa- (-el/-al/r) +
Polish INF (various forms) +
-c +
-any etc. +
(PST.PASS) PTC +
(various forms)
PRS.PL-c- +
Burushaski -á(as) + +
-im/-um/-am +
n-STEM-(a)n +

Pred Ref Ref Pred
Head Head Mod Mod
Lavukaleve -e/-i +
Alamblak ɨnd/Ø +
Wambon Ø +
-mo/-o +
Berbice Dutch fu/fi/Ø +
Babungo kɨ̀(i)/Ø +
Nama -se/!’aa/tsii no INDIC +
Hdi tà + NMLZ +
Mandarin C. -zhe +
Tamil -atu +
-(kk)a +
-a +
Kisi Ø +
Garo -a-ni +
-na +
-kan/-kan-a +
-e/-e-min/-e-r +
Krongo -(t) +
m-/n- + +
Nivkh r/-t-ř//-n +
Table 6.10: Functional possibilities of type 3 D-ALT DCs

Pred Ref Ref Pred
Head Head Mod Mod
Tagalog pag- ? + + +
Kharia RDP/Ø + + + +
-na + + +
-al +
-ker etc. +
Kambera pa- + +
Ø (NMLZ) +
Samoan =ga +
Ø (NMLZ) +

Pred Ref Ref Pred
Head Head Mod Mod
Turkish -DIK/-(y)AcAK + +
-mA +
Kayardild -Thirri-n +
-n- + + +
Ma'di -lɛ̄ + +
-dƷɔ + +
-ka +
Hungarian -ás/-és +
-ni +
-ótt +
Ø (RSP) +
Lango -(kk)ɔ̀ +
Ket Ø (bare INF) + +
Georgian -a +
m- (-a-)(-el/-al)) +
-ul/-il/m- -ar/-al +
Abkhaz -ra +
Polish -nie +
Alamblak -nef +
-(kfë)t +
Dhaasanac -ɲ/-an +
Hdi tá + +
Garo -a +
-gip-a +
Krongo -(t) +
m-/n- + +
West Greenlandic -niq +
-ta/-sa +

6.5 Summary, outlook
In this chapter, the dependent clause constructions of the sample languages
were identified on the basis of their structural coding and classified in two
ways: according to their functional possibilities (section 6.2) and according
to their behavioural potential (section 6.3). These two typologies were then
presented in an integrated fashion in section 6.4.
In Chapter 7, the DC typology presented in the present chapter and the
PoS typology presented in Chapter 5 are combined in order to investigate
dependency relations between the functional possibilities of PoS classes and
of different structural types of DC constructions in the sample languages.

Results
7.1 Introduction
This chapter presents the results for the hypotheses formulated in Chapter
4. First, in section 7.2, I discuss the results for the most general hypothesis,
namely that there is a global match between flexibility versus rigidity in the
domains of PoS and DCs. This prediction is tested first considering DCs
as a single group of constructions. Subsequently, the same prediction is
assessed while distinguishing between the three structural DC types (see
Chapter 3 section 3.4 and Chapter 6, section 6.3.3).
Second, in section 7.3, the functional relation between PoS and DCs
is investigated in terms of specific matches, i.e. one-to-one matches between
particular types of PoS classes and DC constructions, as regards the set
of propositional functions that they can express. As in the case of global
matches, the predictions concerning specific functional matches will be
tested first considering DCs as an undifferentiated group, and then looking
at the three structural DC types separately.
As explained in Chapter 4, section 4.4, the results are presented
as frequency counts in 2x2 contingency tables, which are submitted to
Fischer’s Exact tests in order to identify dependency relations between pairs
of parameters. One parameter of such a pair relates to PoS, and the other to
DCs. Whenever a significant dependency is found, its nature will be further
specified using the method proposed by Maslova (2003).
261
It is shown that significant dependency relations exist between flexible
rather than rigid PoS and DCs; between very flexible rather than less flexible
PoS and DCs; and between PoS and deranked DCs, as opposed to balanced
DCs. These dependencies are asymmetrical in nature, and can as such be
interpreted as statistical bases for implicational universals.
Section 7.4 provides a brief summary of the overall results. The language
data that pertain to the analyses presented in the present chapter are
summarized in a set of tables in the final section (7.5).
7.2 Global functional matches:

Flexibility versus rigidity in the PoS and DC domains
7.2.1 Global matching without differentiating for structural DC-types
In this section I test the hypothesis that there is a global match between
the functional characteristics of the PoS system of a language and its
DC constructions, in terms of flexibility versus rigidity. In Chapter 4 this
hypothesis was operationalized in the form of a two-fold prediction A1 / A2,
repeated here in (1a-1b):
Predictions A1 / A2:
(1) a. If a particular language has one or more flexible DCs, then it
should also have one or more flexible PoS classes.
b. If a language has rigid DCs only, then it should also have rigid
PoS classes only.
7.2.1.2 Global match for flexible constructions

Starting with the results for flexible constructions, Table 7.1 below presents
the frequency counts for Prediction A1. This table (and similar ones in the
remainder of this chapter) should be read as follows: The two grammatical
parameters under investigation appear in the leftmost column and in
the top row, respectively. In this case, i.e. in Table 7.1, these parameters
concern the presence / absence of one or more flexible PoS class(es) and the
presence / absence of one or more flexible DC construction(s). Thus, each of
these parameters represents a variable with two possible values: Either the
relevant trait is attested in the language (+), or it is not attested (-). Together,
these two binary parameters yield four possible value combinations: +/ +, + / -,

- / +, and - / -. The numbers of languages in the sample that exhibit each of
these four value combinations are presented in the four shaded cells in Table
7.1. Each of these shaded cells belongs to a set of two cells, the upper (shaded)
one of which represents the observed frequency, and the lower (transparent)
one the frequency that would be expected if the pattern of co-occurrence of
the two grammatical traits in question were purely coincidental. If, however,
the distribution is not co-incidental, then the observed frequencies in the
two darker shaded cells are higher than the expected frequencies, whereas the
observed frequencies in the lighter shaded cells are lower then the expected
ones. As explained in Chapter 4, Fischer’s Exact yields a p-value that specifies
how likely it is for the observed distribution to be the result of chance; the
critical value is p<0.05 (see Cysouw 2003: 91). In Table 7.1, the observed and
expected counts deviate in the predicted way, but not very much; p = 0.225
(2-sided), which means that there is no significant correlation.
Table 7.1: Frequencies for languages with / without (one or more) flexible PoS class(es) and
DC(s)
Flex DCs (total)
+ – total
Flex PoS + observed freq 10 7 17
expected freq 8 9
– observed freq 11 18 29
expected freq 13 16
total 21 25 46
It should be noted that the total number of 17 languages with (one or

more) flexible PoS class(es) in Table 7.1 includes those languages that have
a large, open and / or a productively derived class of flexible lexemes. When
looking back at the data in Chapter 5 (Table 5.1), we see that this means that
all languages with a small class of modifiers (Itelmen, Thai, Basque, and
Abun) are counted as languages without flexible PoS, while all languages
that are listed above Itelmen in Table 5.1 are counted as languages with
one or more flexible PoS class(es). Note further that Nunggubuyu and
Tuscarora are excluded from all analyses of the present chapter, since they
don’t seem to have any DC constructions. Rather, these languages express
all subordination relations by means of separate, independent clauses.
Therefore, no comparison can be made between the distributional patterns
of the PoS classes in these languages and any DC construction. The language
Chapter 7 – Results | 263

data on which Table 7.1 above is based can be found in Tables 7.33 and 7.34 in
section 7.5 at the end of this chapter.
7.2.1.3 Global match for rigid constructions

Consider now Table 7.2 below, which shows the results for prediction A2
(see (1b) above), concerning the correlation between rigid PoS only and rigid
DCs only.
As in the case of flexible constructions, no significant correlation was
found: Fisher’s exact yields p = 0.552 (2-sided). The relevant language data
can again be found in Tables 7.33 and 7.34 of section 7.5.
Table 7.2: Frequencies for language with / without rigid PoS only and rigid DCs only
Rig DCs (total)
+ – total
Rig PoS + observed freq 15 14 29
expected freq 14 15
expected freq 8 9
total 22 24 46
7.2.1.4 Summary
In its most general form, the hypothesis of a global match between flexibility
versus rigidity in the domains of PoS and DCs is rejected. In the following
section this hypothesis is refined by taking into account the parameter of
structural DC type.
7.2.2 The parameter of structural DC type
In this section I introduce the parameter of the internal structural properties
of DCs. The prediction, as formulated in Chapter 4, is that the more formally
similar a DC construction is to a lexical expression, the more functionally
similar it will be to this lexical counterpart. In other words, it is expected
that functional connections will be stronger between PoS and deranked DCs
(type 2,3) than between PoS and balanced DCs (type 1). In addition, within
the group of deranked DCs, it is expected that more deranked DCs (type 3,
D-ALT) will show more functional similarity with PoS than less deranked

DCs (type 2, D-SENT). In Chapter 4 this was operationalized in the form
of the two predictions B1 and B2, which are repeated in (2a-b) below:
Prediction B1 / B2:
(2) a. The functional possibilities of deranked DCs (type 2 / 3) are
more similar to the functional possibilities of PoS than those of
balanced DCs (type 1).
b. Within the group of deranked DCs, the functional possibilities
of type 3 DCs are more similar to the functional possibilities of
PoS than those of type 2 DCs
These predictions will now be combined with the predictions concerning global
matches between flexible versus rigid PoS systems and DC constructions.
7.2.2.2 Global match of flexible constructions, differentiating for

structural DC type
Applying predictions B1 and B2 to flexible PoS and DC systems yields a set
of four hypotheses, which are given in (3a-d) below (cf. (3) of Chapter 4).
It is expected that the prediction in (3a) is less likely to hold true than the
one in (3b). Furthermore, the prediction in (3c) is less likely to be confirmed
than the one in (3d).
(3) a. If a language has one or more flexible balanced DC(s) of type 1,
b. If a language has one or more flexible deranked DC(s) of type
2 / 3, then it should also have one or more flexible PoS class(es).
c. If a language has one or more flexible deranked DC(s) of type 2,
d. If a language has one or more flexible deranked DC(s) of type 3,
The distributions pertaining to the predictions in (3a-d) are presented in

Tables 7.3, 7.4, 7.6, and 7.7 below. The language data can be found in Tables
7.33 and 7.34 of section 7.5.
First, the frequencies relevant for the prediction in (3a), concerning
balanced DCs, appear in Table 7.3. This table shows that there is hardly
any difference between the observed and the expected dsitribution; Fischer’s
exact yields p = 0.755 (2-sided).

Table 7.3: Frequencies for languages with / without flexible PoS and flexible DCs of type 1
Flex DCs type 1 (balanced)
+ – total
expected freq 6 11
expected freq 11 18
total 17 29 46
Second, the frequencies pertaining to the prediction in (3b), concerning

deranked DCs, are presented in Table 7.4. As expected, this distribution
does yield a significant correlation: p = 0.025 (2-sided). The contingency
coefficient (CC) is 0.339, which suggests a moderately strong effect (see
Chapter 4, section 4.4).
Table 7.4: Frequencies for languages with / without flexible PoS and flexible DCs of type
2/3
Flex DCs type 2 / 3 (deranked)
+ – total
expected freq 4 13
expected freq 6 23
total 10 36 46
Following Maslova (2003; and see Chapter 4, section 4.4), the PoS parameter
and the DC parameters are each separately correlated to the derived
parameter PoS = DC (i.e. the parameter of PoS and DC having the same
or different values), in order to assess the nature of the dependency relation
identified in Table 7.4. This yields Tables 7.5a and 7.5b:
Table 7.5a: Maslova test no. 1: change DC parameter to PoS=DC

Flex Pos + Flex Pos –
PoS = DC 7 26
PoS DC 10 3

Flex DC 2 / 3 + Flex DC 2 / 3 –
PoS = DC 7 26
PoS DC 3 10
p = 1 (not significant)
Tables 7.5a and 7.5b indicate a one-sided asymmetrical dependency between

flexible PoS and deranked flexible DCs. In particular, Table 7.5a shows that
the negative value of the PoS parameter constrains the event of the PoS and
DC parameters having the same value 102. In contrast, Table 7.5b shows that
the value for the DC parameter and the event of the PoS and DC parameters
having the same value are completely independent of each other.
This outcome is interpreted as the statistical counterpart of the
typical situation encountered with implicational universals, namely a 2x2
contingency table with exactly one (nearly) empty cell. In the case at hand,
this is the cell with the value combination [-Flex PoS, + Flex DC type 2 / 3].
The observed frequency in this cell (3) is both low and significantly lower
than the expected frequency (6). This constitutes statistical evidence for the
universal in (4):
(4) [- Flex PoS] 2 [- Flex DC 2 / 3]

Languages without any flexible PoS class(es) also lack flexible
deranked DC(s) of type 2 / 3 (independent of whether languages with
flexible PoS class(es) have flexible deranked DC(s) of type 2 / 3).
Relating this result to the prediction in (3b), we can say that indeed the
availability of flexible DC constructions of type 2/ 3 in a language is dependent
on the availability of at least one flexible PoS class(es) in that language.
I turn now to the predictions in (3c) and (3d) above, differentiating within
the group of deranked flexible DC between DCs of type 2 (D-SENT) and
DCs of type 3 (D-ALT). First, the frequencies pertaining to the prediction
in (3c), concerning type 2 DCs, appear in Table 7.6. They do not show any
correlation: the observed and expected counts are identical, so that Fischer’s
Exact yields p =1.
102
Note that the pattern in Table 7.5a pattern differs from the one in Table 4.9a of Chapter 4
(the hypothetical illustration of a one-sided asymmetrical dependency relation), to the extent
that in the latter case it is the positive rather than the negative value of the PoS parameter for
which the distribution of flexible DCs is skewed.

Table 7.6: Frequencies for languages with / without flex PoS and flex DCs of type 2
Flex DCs type 2 (D-SENT)
+ – total
expected freq 1 16
expected freq 1 28
total 2 44 46
At first sight, this result is unexpected, especially in view of the universal

in (4). However, this finding can be explained by the fact that there are
hardly any flexible DCs of type 2 attested in the sample. That is to say, when
comparing Table 7.6 with Table 7.4, it can be seen that only 2 out of the
total 10 flexible deranked DCs involve a construction of type 2; all others
are of type 3. Therefore, it is not surprising that we do not find a significant
correlation when testing the prediction in (3c).
On the other hand, and as expected, we do find a significant correlation
when testing the prediction in (3d), which is concerned with flexible DCs
of type 3. The relevant frequencies are presented in Table 7.7 below, which
differs only very slightly from Table 7.4 above. They yield a significant
p-value of 0.038. The CC value is 0.340, which indicates a moderately
strong effect.
Table 7.7: Frequencies for languages with / without flex PoS and flex DCs of type 3
Flex DCs type 3 (D-ALT)
+ – total
expected freq 3 14
expected freq 5 24
total 8 38 46
When applying Maslova’s method to these data, again a one-sided asymmetrical

dependency is revealed, as the data in Table 7.8a and 7.8b show:

Flex Pos + Flex Pos –
PoS = DC 6 27
PoS DC 11 2
p = 0.000 (significant)

Flex DC3 + Flex DC3 –
PoS = DC 6 27
PoS DC 2 11
Table 7.8a shows that the negative value of the PoS parameter constrains the
event of the PoS and DC parameters having the same value, while Table 7.8b
shows that the DC parameter and the PoS=DC parameter do not interact.
This is interpreted as evidence for the universal in (5), which is in fact a more
precise version of (4) above.
(5) [- Flex PoS] 2 [- Flex DC 3]

Languages without any flexible PoS class(es) also lack flexible
deranked DC(s) of type 3 (independent of whether languages with
flexible PoS class(es) have flexible DC(s) of type 3).
In sum, these data show that taking into account structural DC type
reveals a dependency relation between flexible PoS and flexible deranked
DCs. In particular, flexible deranked DCs hardly ever occur in languages
without any flexible PoS class(es). However, this does not mean that
all languages with a flexible PoS class also have a flexible deranked
DC construction.
7.2.2.3 Global match for rigid constructions, differentiating for

structural DC type
In this section I turn back to prediction A2 about rigid constructions (see (1b)
above), and combine it with the parameter of structural DC type (see B1 / B2
in (2a, b) above). Recall the definitions of languages with flexible versus rigid
constructions: Whereas flexible PoS and DC systems are defined as those
that have at least one flexible construction, rigid PoS and DC systems consist
of rigid constructions only. It was explained that, due to this asymmetry,

it does not make sense to formulate specific predictions about correlations
between languages with rigid PoS only and rigid DCs of a specific structural
type only. This is because there is no reasonable basis on which to expect
languages to have only a single structural type of rigid DCs. On the contrary,
it is expected that semantically different types of subordination relations will
be expressed by DC constructions with different internal morpho-syntactic
properties. Indeed, as can be seen in Table 34.7 in section 7.5, there are only
a few languages in the sample that have both rigid PoS only and rigid DCs
only, and in which all DCs are of a single structural type. The languages
that display such a pattern are Abun, Bukiyip, Tamil, and Hixkaryana. All
other languages with rigid PoS only and rigid DCs only (11 in total) show
a mixture of two or three structurally different DC types.
Nevertheless, we may investigate which structural DC types occur
in groups of languages with different values for the parameters of rigid
PoS only and rigid DCs only. Specifically interesting are languages with
unexpected value combinations, i.e. languages that do not show a match
between the functional possibilities of PoS and DCs. Such cases may involve
(i) languages with one or more flexible PoS class(es) and with rigid DCs
only, and (ii) languages with rigid PoS only and with one or more flexible
DC construction(s).
Consider once more the data in Table 7.2, repeated below as Table 7.2’ for
convenience, which presents the frequencies for languages with and without
rigid PoS and DCs only (without differentiating for structural DC type).
Table 7.2’: Frequencies for language with / without rigid PoS only and rigid DCs only
Rig DCs (total)
+ – total
Rig PoS + observed freq 15 14 29
expected freq 14 15
expected freq 8 9
total 22 24 46
Languages with flexible PoS systems and rigid DCs only
I start with a discussion of those languages that display the unexpected

feature combination mentioned under (i) above: one or more flexible PoS

class(es) and rigid DCs only ([- Rig PoS only; + Rig DCs only]). Table 7.2’
shows that there are 7 languages of this type. They are: Guaraní, Warao,
Hungarian, Japanese, Hmong Njua, Lango, and Koasati. Considering Table
7.33 in section 7.5, we see that in some of these cases the rigid DCs in question
are balanced (type 1). Even though the attestation of these constructions
contradicts the prediction of a global mapping of rigid PoS and DCs, this
result is relatively un-surprising, since balanced DCs were predicted to be
least likely to show functional similarity to PoS.
What is apparently more surprising is the attestation of a number of
deranked (type 2 or 3) rigid DC constructions in the group of languages with
flexible PoS systems. However, in almost all of these cases the rigid deranked
DC can be regarded as the clausal counterpart of a rigid PoS class (possibly
a derived or a small class) that forms part of a system that also includes
one or more flexible PoS class(es). Therefore, these cases in fact confirm the
prediction that PoS and (deranked) DCs exhibit parallel functional patterns.
In what follows, I will discuss in more detail the cases of languages with
flexible PoS systems and rigid (balanced or deranked) DCs only.
Guaraní and Warao are both languages with a class of very flexible
lexemes: they have contentives and non-verbs, respectively. Guaraní does
not have any relevant extra rigid PoS classes to which its rigid DCs can be
linked. However, all rigid DCs in this language are balanced (type 1), and as
such not very likely to behave functionally like PoS. Warao, in contrast, has a
balanced rigid relative clause construction and a deranked rigid complement
construction of type 2. While the attestation of the former may again be
explained in terms of lack of formal (and therefore functional) similarity
with PoS, the presence of the latter DC can be related to the a class of
derived nouns in Warao. This class of nouns was not taken into the account
in the original analysis, since – in order to avoid double values – classes
of simple, un-derived items are given prominence over classes of derived
items in cases where both are available to express a particular propositional
function (in this case the function of head of a referential phrase).
The other languages with flexible PoS systems and rigid DCs only
(Hungarian, Japanese, Hmong Nua, Lango, and Koasati) have either a
class of nominals or a class of modifiers. Hungarian, for instance, has lexical
nominals and, contrary to the predictions, rigid nominalizations and participle
constructions of types 2 and 3. A possible explanation for this counterexample
may again be sought in the realm of derivational morphology. For one thing,
the form -ás / -és, which is used to mark Hungarian rigid nominalizations of

type 3, can also be used for the lexical derivation of action nouns, as in: olt
‘extinguish’ 2 olt-as ‘extinguishing’. There is thus a functional match between
the deranked DC construction and a derived PoS class. Interestingly, one of
the rigid participial constructions in Hungarian, namely the active (present)
participle form in -ó / -ő, is also productively used to derive lexical nouns,
as for example in: alvas ‘read’ 2 olvas-ó ‘reader’, and ebédel ‘have dinner’
2 ebédl-ő ‘dining room’. Thus, even though the participial construction is
not flexible, its marker also appears in the function of head of a referential
phrase, as a marker of lexically derived nouns. In addition to rigid deranked
nominalizations and participles, Hungarian has three balanced rigid DCs:
complement clauses, relative clauses, and adverbial manner clauses. The
functional rigidity of the former two constructions does not match with the
flexibility of lexical nominals, but their balanced form makes them unlikely
candidates to exhibit such a match in the first place.
Japanese also has nominals, but unlike Hungarian it has a class of simple
rigid nouns as well. Therefore, the rigid complement clauses attested in
Japanese indeed have a lexical counterpart, and as such do not constitute a
counterexample103. Lango is a language with two rigid complement clause
constructions: a deranked one (of type 3) and a balanced one. These DCs are
again not to be regarded counterexamples, since Lango has a class of rigid
simple nouns, which serves as the lexical counterpart for the rigid complement
clauses. In addition to these constructions, Lango has rigid balanced relative
clauses, which do not match with its flexible lexical modifiers. Hmong Njua,
similarly, has flexible lexical modifiers and rigid balanced relative clauses. In
both cases, the fact that the DC construction is balanced may explain the
lack of functional similarity with the relevant PoS class.
Finally, there is one language, Koasati, the PoS system of which is
classified as flexible because it contains a derived class of modifiers, but
which is otherwise fully rigid, i.e. displays open classes of nouns and verbs,
a small class of adjectives, and small and derived classes of manner adverbs.
This means that the nominalizations, participial and converbal constructions
attested in Koasati, all of which are deranked (type 2), have the expected
lexical counterparts.
In sum, there is only one truly problematic case, i.e. one language with
a rigid, deranked DC for which no rigid (derived) lexical counterpart is
103
Recall from Chapter 6 that the classification of Japanese nominalizations in terms of
structural type is not evident: They were classified as 1 / 3, because they retain tense, but also
have the possibility of expressing the subject as a possessor.

available: Hungarian has rigid deranked participle constructions of type 2
and 3, without (simple or derived) rigid adjectives.
Languages with flexible PoS systems and a mixture of flexible and rigid DCs
Table 7.2’ shows that, apart from the group of 7 languages with at least one
flexible PoS class and with rigid DCs only, there is a group of 10 languages
with at least one flexible PoS that do not have rigid DCs only ([- Rig
PoS only; - Rig DCs only]). In principle, this latter value combination is
expected. Notably however, 9 out of the 10 relevant languages have one or
more rigid (deranked) DCs alongside the expected flexible DCs. In other
words, these 9 languages with flexible PoS systems have a mixed system of
flexible and rigid DCs.
Consider for instance a language like Ma’di: It has lexical nominals and,
as predicted, flexible deranked nominal clauses. The latter construction is
illustrated in (6a-b):
Ma’di (Blackings & Fabb 2003: 202, 22)

(6) a. Má lɛ̀-ā [èɓí ˋ ɲā-lɛ́] rá
1sg (n)want-obj fish n-eat-nmlz AFF
‘I certainly want to eat fish.’
b. àràbɨ́à [ɔ́pɨ́ Ɂà dƷɨ̄-lɛ́] rɨ̀ pá nā ādī.

car Opi poss (n)-take-ptc def leg afr deflate
‘The car which Opi took has a flat tyre.’
However, in modifier function the lɛ́-construction is used for object relative

clauses only. For subject relative clauses, a different rigid deranked relative
clause construction (of type 2) is used, as is illustrated in (7):

(7) ágɔ́́ àm-à dƷɔ́ ̀ nɨ ̄ ˋsī-ɓá rɨ̀
man [1pl-poss house spec pron n-build-ptc(pl)] def
‘One of the men who built / are building our house.’
Similarly, Turkish has flexible lexical non-verbs and flexible deranked

nominal clauses of type 3. The latter are illustrated in (8a-8b):

(8) a. (ben) [Ahmed-in öl-düg-ün]-ü duy-du-m
‘I heard that Ahmed died.’
b. [adam-ın git-tiğ-i] okul

‘the school that the man goes / went to’
However, Turkish also has a number of rigid deranked DCs. Examples

(9) and (10) show the infinitival construction in -mAK (type 2), and the
nominalization in -mA (type 3), respectively. Both constructions can be
used in the function of head of a referential phrase only. Example (11)
illustrates the rigid participle construction marked by -An (type 2), which
is used exclusively as a modifier in a referential phrase, expressing subject
and possessive relative clauses. Finally, example (12) shows the rigid converb
construction in -(y)ArAk (also of type 2), which can express only the function
of modifier in a predicate phrase.
Turkish (Kornfilt 1997: 51, 73; Göksel & Kerslake 2005: 420, 440)
(9) [Lütfen pencere-yi aç-mağ]-ı unut-ma
please window-acc open-inf-acc forget-neg
‘Please, don’t forget to open the window!’

iste-n-iyor-muş
‘It seems they want [everyone to tell a story].’
(11) [ögretmen ol-an] Haydar

‘Haydar, who is a teacher’
(12) Ben [etraf-ım-a bak-arak] yür-ür-üm

I around-1sg-dat look-conv walk-aor-1sg
‘I walk looking around (myself ).’

In short, almost all languages with a flexible PoS system (i.e. with at least
one flexible PoS class) have at least one rigid deranked DC construction;
sometimes in combination with flexible deranked DC(s), sometimes not.
In addition, the examples from Turkish point to yet another interesting
issue, namely the fact that flexible deranked DCs in languages with flexible
PoS sometimes do not show exactly the same distributional pattern as their
lexical counterparts, but are rather somewhat less flexible than the relevant
PoS class. In particular, Turkish nominal clauses can express two out of
the three functions that can be expressed by the lexical non-verbs in this
language104. In Chapter 8 I will return to this issue, and suggest that there
are good functional reasons for deranked DCs to be less flexible than PoS
classes in the same language.
Languages with rigid PoS systems and with flexible DCs
Having discussed languages with flexible PoS systems and rigid (as well as
flexible) DCs, I will now consider the second group of languages with an
unexpected combination of parameter values: rigid PoS only and one or
more flexible DC construction(s) ([+ Rig PoS only; - Rig DCs only]). Table
7.2’ above shows that half of the total amount of languages with rigid PoS
only (14 out of 29) do not have rigid DCs only. However, when looking at
Table 7.33 in section 7.5, we see that the majority of these cases (11 out of 14)
involve balanced DCs 105. This suggests again that a lack of formal similarity
ties in with a lack of functional similarity between DCs and PoS. In addition,
there may be other factors at work. First, the relevant flexible balanced DCs
often have the possibility to use some extra morpho-syntactic means (i.e.
in addition to their ‘regular’ structural coding) in order to disambiguate the
multiple functions in which they can appear (Hengeveld & Van Lier 2008).
Second, diachronic developments also seem to play a role. These issues will
be discussed further in Chapter 8.
104
In combination with a postposition gibi (‘like’), nominal clauses in -DIK can be used as
similative clauses (cf. Chapter 6, section 6.2.2), as illustrated in (i). Note however, that this
is not a case of flexibility since -DIK clauses can only function as predicate modifiers in
combination with a postposition (see Chapter 6, section 6.2.2.1).
(i)
Pastayı [ [anne-m-in analat-tɪğ-ɪ] gibi ] yapmaya çalıştım
the cake mother-1sg.poss-gen describe-nmlz-3sg.poss like I tried to make
‘I tried to make the cake [as my mother had described].’ (Göksel & Kerslake 2005: 477)
105
There is one language with rigid PoS, Burushaski, which has both balanced (type 1) and
deranked (type 2) flexible DCs.

Apart from flexible balanced DCs, there are also three unexpected cases of
flexible deranked DC constructions in languages with rigid PoS only (cf. the
observed frequency of the ‘nearly empty cell’ in Table 7.4). The relevant cases are
attested in Burushaski, Hdi, and Krongo, and I will discuss them one by one.
First, Burushaski has rigid PoS classes to express the functions of head
and modifier in a referential phrase, i.e. nouns and adjectives, in combination
with a flexible deranked nominal clause of type 2 that can express both these
functions. The flexible DC is illustrated in (13a-b).
Burushaski (Anderson 2002: 545, Berger 1998a: 171)

(13) a. buṭ muškíl bilá [góo-ltir-as]
very difficult be.IV 2-show-inf
‘It is very difficult to show (it) to you.’
b. [Chá-aṭe oóo-rvγ-as] hvk

post-superess neg-sit-ptc dog
‘a dog which doesn’t sit at its post.’
I have not been able to find a plausible explanation for this counterexample,
except that there are some indications that the lexical distinction between
nouns and adjectives in Burushaski is also not particularly clear-cut.
Lorimer (1935: 102), for instance, characterizes the distinction as “messy”
and observes that “nouns borrowed from other languages are in many cases used
as adjectives, which seems to show a slowness to appreciate the distinction between
noun and adjective”. Some examples of such flexible items would be zo’r
‘power / powerful’, s̆ərʊm ‘shame / ashamed’, and xʌtər ‘danger / dangerous’ (cf.
Berger 1998a: 78). The Burushaski dictionary (Berger 1998b) also provides
some indications of flexible items, such as aasáan ‘easy / easiness’, and ajóono
‘strange / stranger’. Although these data obviously do not suffice as a full
explanation of the flexibility displayed by Burushaski -as clauses, at least
they put their unexpected distributional pattern into some perspective.
Second, Hdi has a deranked (type 3) nominal clause construction, which
is marked by tá. Notably, this marker is not exclusively used for structural
coding of DCs; it also combines with non-clausal constituents and as such
can have two functions: Either it marks a lexical object, or it is a so-called
‘comment marker’ in a focus construction. When introducing a complement
clause, tá is glossed as having the former function, i.e. that of an object
marker. This is shown in example (14a). In contrast, as can be seen in (14b),

tá is glossed as a comment marker when marking a relative clause. Thus,
the flexibility of this DC construction seems to be due to the general multi-
functionality of tá, the morpheme used for its structural coding.
Hdi (Frajzyngier & Shay 2002: 480, 406)

(14) a. Sí tà ɗv-áy-xə̀n [tá hliy-á-mú]
pst ipfv want-po-3pl obj leave-gen-1pl.incl
‘They wanted us to leave.’ (lit. They wanted our leaving.)
b. ghùrúm [tá lá-ghw-í ndá mà xàɗík]

hole comm go-d:so-ref assoc in ground
‘a hole that went deep into the ground’
In Krongo, finally, the flexible deranked DC is a modifier clause; it can be used as

a modifier of either a referential or a predicative head. This contruction has been
exemplified in Chapter 6 (see example (35a-b). Crucially, Krongo does not have
any (flexible or rigid) lexical strategy to express the two modifier functions. This
means that it is not possible to compare the distributional pattern of the Krongo
modifier clause to that of any PoS class(es). Therefore, this counterexample will
be excluded in the analyses of section 7.3, where we will be concerned with
specific matches between individual PoS classes and DC constructions.
In short, there are only very few cases of deranked flexible DCs in
languages with rigid PoS systems; the large majority of flexible DCs in these
languages is balanced.
7.2.2.4 Summary
In this section it has been shown that, when considering DCs as an
undifferentiated group, there is no global match between flexibility versus
rigidity in the domains of PoS and DCs. However, when taking into account
the parameter of structural DC type, the following dependency relation is
revealed: Deranked flexible DCs are almost completely absent in languages
without any flexibility in their PoS system, but these languages quite often
do have balanced flexible DCs.
On the other hand, languages with flexible PoS systems exhibit both flexible
and rigid deranked DCs. Some of these languages display a combination of
flexible and rigid DCs, while others have rigid DCs only. In fact, however,
almost all languages with at a flexible PoS system have at least one rigid DC
construction. These rigid DCs can be deranked or balanced.

7.3 Specific functional matches: Types and amounts of
flexibility / rigidity in the PoS and DC domains
7.3.1 Introduction
In this section I investigate whether there are correlations between the
availability in a language of specific types of flexible and rigid PoS classes,
which can express a particular (set of ) propositional function(s), and the
availability of DC constructions with the same functional possibilities. It
is expected that each secondary DC construction of a specific flexible / rigid
type will have a primary lexical counterpart of the same flexible / rigid type.
This hypothesis was operationalized in Chapter 4 in the form of Prediction
C, repeated in (16) below (cf. (6) in Chapter 4):
Prediction C:
(16) If a language has a DC construction of a specific flexible or rigid type
X, then it should also have a PoS class of type X.
In what follows, Prediction C is tested first for specific flexible constructions,

and then for specific rigid constructions. In every case, I will first consider
DCs as a single group of constructions, and then differentiate them according
to internal structural type.
7.3.2 Specific matches for flexible constructions
Starting with flexible PoS and DC constructions, Prediction C in (16) above
can be broken down into three sub-predictions, one for each functional type,
as in (17a-c):
(17) a. If a language has contentive and / or multi-functional clauses,

then it should also have lexical contentives and / or non-verbs.
b. If a language has nominal clauses, then it should also have
lexical nominals.
c. If a language has modifier clauses, then it should also have
lexical modifiers.
Note that, unlike in Chapter 4 (see (7) in that Chapter), the prediction
in (17a) treats two flexible construction types as one: (i) constructions

that can be used in all four functions (lexical contentives and contentive
clauses), and (ii) constructions that can be used in all functions except the
head of a predicate phrase (lexical non-verbs and multifunctional clauses).
This is because, as shown in Chapter 6, contentive clauses are very rare.
No predictions are formulated either that make reference to flexible DC
types involving the functional slot for head of a predicate phrase (predicative
clauses, head clauses, and Flex Clauses A, B, and D, see Chapter 3, section
3.2.3), since these constructions are not attested in the sample languages (see
Chapter 6, section 6.2.3.2). Finally, while flexible DCs of type C (that can
be used as the head of a referential phrase and the modifier in a predicate
phrase) are indeed attested, their lexical equivalent (Flex PoS C) is not (see
Chapter 5, section 5.3.4) 106. Therefore, no predictions are formulated about
constructions with this particular functional pattern.
Each of the predictions in (17) can be combined with the parameter of
structural DC type. For example, the prediction in (17a) can be broken down
into four sub-predictions, as in (18a-d) (cf. (9a-d) of Chapter 4). As before, it is
expected that the prediction in (18a) is less likely to hold than the one in (18b),
and that the prediction in (18c) is less likely to hold than the one in (18d).
(18) a. If a language has balanced contentive and / or multifunctional

clauses of type 1, then it should also have lexical contentives
and / or non-verbs.
b. If a language has deranked contentive and / or multifunctional
clauses of type 2 / 3, then it should also have lexical contentives
and / or non-verbs.
c. If a language has deranked contentive and / or multifunctional
and / or non-verbs.
d. If a language has deranked contentive and / or multifunctional
and / or non-verbs.
In the remainder of this section I present the results for the three predictions
in (17a-c) one by one. Each of these three predictions will first be tested
106
It is noteworthy that there are thus DC types without a functional lexical counterpart.
However, flexible DCs of type C are infrequently attested (N=3), and involve either balanced
clauses (in Basque and West Greenlandic) or an infinitive construction that is reduplicated in
the function of modifier in a predicate phrase (Imbabura Quechua).

considering DCs as a single group, and then differentiating for structural
DC types, along the lines of (18a-d).
7.3.2.2 Flexible match 1:Lexical contentives / non-verbs and

contentive / multi-functional clauses
Consider first the results for the prediction in (17a). This prediction
involves the most pervasively flexible construction types, namely lexical
contentives / non-verbs and contentive / multi-functional clauses. The
relevant frequencies are given in Table 7.9. As expected, they reveal a
significant correlation: p = 0.037 (2-sided). The CC value is 0.351, indicating
a moderately strong effect. The relevant language data are summarized in
Tables 7.35 and 7.36 of section 7.5.
Table 7.9: Frequencies for languages with / without lexical contentives / non-verbs and
contentive clauses / multi-functional clauses
DCs: contentive / multi-functional
clauses
+ – total
PoS + observed freq 3 4 7
contentives / expected freq 1 6
non-verbs
expected freq 5 34
total 6 40 46
Applying Maslova’s method to the data in Table 7.9 reveals a two-sided

asymmetrical dependency, as Tables 7.10a and 7.10b make clear:

Pos contentive / non-verb + Pos contentive / non-verb –
PoS = DC 3 36
PoS DC 4 3

Table 7.10b: Maslova test no. 2: change PoS parameter to PoS=DC
DC contentive / multi- DC contentive / multi-
functional + functional –
PoS = DC 3 36
PoS DC 3 4
The significant correlations in Tables 7.10a and 7.10b show that the negative
values of both the PoS and the DC parameters constrain the event of the
PoS and DC parameters having the same value107. These results are
interpreted as evidence for the universal in (19), which is in fact a pair of two
implicational universals:
(19) [- PoS contentive/non-verb] ™2 [- DC contentive/multi-functional]

Languages without lexical contentives or non-verbs also lack
contentive or multi-functional clauses, and vice versa (languages
without contentive or multi-functional clauses also lack lexical
contentives and non-verbs).
Notably however, the original distribution in Table 7.9 makes clear that the
PoS and the DC parameters under investigation, involving the availability of
maximal flexibility, are both strongly skewed towards the negative value. In
other words, lexical contentives / non-verbs and contentive / multi-functional
clauses are rare phenomena, independently of each other. This is why there
are three nearly empty cells in Table 7.9, rather than exactly one nearly empty
cell as would be expected in the case of a ‘classical’ implicational universal.
This means that the implicational universals in (19), which have negative
values in the consequent parts, are not particularly informative.
A much more interesting generalization that can be made on the basis
of Table 7.9 is that the occurrence of a positive value on the PoS parameter
strongly increases the likelihood that the DC parameter will also have a
positive value, even though this likelihood is still only about 50%. In other
words, while maximally flexible PoS classes are rare, when they are attested
in a language, this dramatically increases the chances that that the same type
of flexibility will be attested in the DC system, resulting in a (nearly) even
107
Note that this pattern differs again from the hypothetical one for a two-sided
asymmetrical dependency relation as illustrated in Chapter 4, Tables 4.11a-b. In the latter
case, the positive value of the PoS parameter (Table 4.11a) and the negative value of the DC
parameter (Table 4.11b) constrain the PoS = DC parameter.

distribution. This finding supports the general hypothesis that there is a
strong tendency for the PoS and DC parameters to have the same value 108.
I will now investigate the effect of the parameter of structural DC type on
the correlation between maximally flexible PoS and DCs, as operationalized
in the set of predictions listed in (18a-d) above. First, consider the results
for the prediction in (18a), concerning balanced clauses. The relevant
frequencies are given in Table 7.11. Like in al the previous cases involving
balanced clauses, these counts do not reveal a significant correlation: p =
0.160 (2-sided).
Table 7.11: Frequencies for languages with / without contentives / non-verbs and contentive
clauses / multi-functional clauses of type 1
clauses type 1 (balanced)
+ – total
contentives / expected freq 1 6
non-verbs
expected freq 4 35
total 5 41 46
Turning to deranked contentive and multi-functional clauses (see (18b-d)

above), we find that all such constructions attested in the sample are of
type 3 rather than type 2 (cf. Table 7.35 in section 7.5). Therefore, it is not
possible to differentiate between clauses of type 3 and clauses of type 2.
The frequencies for deranked contentive and multi-functional clauses of
type 3 are presented in Table 7.12. As expected, these frequencies reveal a
significant correlation: p = 0.000 (2-sided). Moreover, the CC value is 0.589,
which indicates a strong effect.
108
I am indebted to Elena Maslova for her helpful comments on the interpretation of
these data.

Table 7.12: Frequencies for languages with / without lexical contentives / non-verbs and
contentive clauses / multi-functional clauses of type 3
clauses type 3 (D-ALT)
+ – total
contentives expected freq 1 6
/ non-verbs
expected freq 3 36
total 4 42 46
Applying Maslova’s method makes clear that this correlation involves a one-
sided asymmetrical dependency. The relevant data are presented in Tables
7.13a and 7.13b:

PoS contentive / non-verb + PoS contentive / non-verb
–
PoS = DC 4 39
PoS DC 3 0
Table 7.13b: Maslova test no. 2: change PoS parameter to PoS=DC

DC contentive / multi- DC contentive / multi-
functional + functional type 3 –
PoS = DC 4 39
PoS DC 0 3
Table 7.13a shows that the negative value of the PoS parameter constrains
the event of PoS and DCs having the same value. Table 7.13b, in contrast,
shows that the DC parameter and the PoS=DC parameter do not interact.
On the basis of these results, the universal in (20) can be formulated:

(20) [- PoS contentive / non-verb] 2 [- DC contentive / multi-functional
type 3]
Languages without lexical contentives or non-verbs also lack
deranked contentive / multi-functional clauses of type 3.
Notably, the distribution in Table 7.12 is similar to the one in Table 7.9, to the
extent that it displays three (nearly) empty cells, as a result of the fact that both
the PoS and the DC parameter are, independently of each other, strongly
skewed towards the negative value. In view of this, the implicational universal
in (20), which has a negative value in the consequent part, does not have much
explanatory power. Again, it is much more telling that the likelihood of the
DC parameter having the rare positive value increases to over 50% under the
influence of the PoS parameter having this rare positive value.
7.3.2.3 Flexible match 2: Nominals and nominal clauses

Continuing with less pervasive flexibility, I will now investigate the
relationship between lexical nominals and nominal clauses. First consider
the data in Table 7.14, in which nominal clauses are considered as one
undifferentiated group. The relevant language data can be found in Tables
7.37 and 7.38 of section 7.5. The observed frequencies in Table 7.14 do not
deviate at all from the expected ones, so that p = 1 (2-sided). This means that
there is no dependency relation between the presence of lexical nominals
and nominal clauses.
Table 7.14: Frequencies for languages with / without nominals and nominal clauses
DCs: nominal clauses
+ – total
PoS: + observed freq 2 3 5
nominals expected freq 2 3
expected freq 14 27
total 16 30 46
Taking into account the parameter of structural DC types, we can formulate

the more specific predictions in (21a-d), in parallel with the ones listed in
(18a-d) for contentives / non-verbs and contentive / multi-functional clauses.

(21) a. If a language has balanced nominal clauses of type 1, then it
should also have lexical nominals.
b. If a language has deranked nominal clauses of type 2 / 3, then it
c. If a language has deranked nominal clauses of type 2, then it
d. If a language has deranked nominal clauses of type 3, then it
First consider the prediction in (21a). In line with the other results found so
far, there is no correlation between lexical nominals and balanced nominal
clauses. The relevant frequencies appear in Table 7.15; Fisher’s Exact yields
p = 0.301 (2-sided). In fact, as this table shows, none of the 13 languages
with nominal clauses of type 1 has lexical nominals. (For the language data
pertaining to Table 7.15, see again Tables 7.37 and 7.38, in section 7.5.)
Table 7.15: Frequencies for languages with/without nominals and nominal clauses of type 1
DCs: nominal clauses type 1
(balanced)
+ – total
expected freq 12 29
total 13 33 46
Regarding the prediction in (21b), the frequencies for deranked nominal

clauses of type 2 and 3 appear in Table 7.16. Contrary to the expectations,
the observed frequencies do not yield a significant correlation (p = 0.120).

Table 7.16: Frequencies for languages with/without nominals and nominal clauses of type 2/3
DCs: nominal clauses type 2 / 3
(deranked)
+ – total
expected freq 5 36
total 6 40 46
Notably, the total number of languages with deranked nominal clause

constructions attested in the whole sample is quite low (6). Therefore,
splitting up the group of deranked nominal clauses into constructions of
type 2 versus type 3 does not influence the results in any interesting way. For
the sake of completeness, the relevant frequencies are presented in Table 7.17
(for nominal clauses of type 2) and Table 7.18 (for nominal clauses of type 3).
Not surprisingly, no significant correlations are found: The relevant p-values
are p = 0.208 (2-sided) for Table 7.17, and p = 0.379 (2-sided) for Table 7.18.
+ – total
expected freq 2 41
total 2 44 46
+ – total
expected freq 4 39
total 4 42 46

7.3.2.4 Flexible match 3: Modifiers and modifier clauses
Finally, consider the results for the third flexible construction type: lexical
modifiers and modifier clauses, presented in Table 7.19 below. It must be
noted that the total number of languages relevant for this prediction is only
38. This is because in 10 languages neither of the two modifier slots can be
expressed by means of a large, open PoS class 109. Some of these 10 languages
do have a small, rigid class of adjectives and / or adverbs, while others have
no lexical strategy at all. Both groups are interpreted as lacking a value for
the PoS parameter. These cases are excluded from the analysis, since it is
impossible to compare the distributional pattern of the DC construction
with the pattern of any large, open PoS class. The language data pertaining
to lexical modifiers and modifier clauses can be found in Tables 7.39 and 7.40
in section 7.5.
Table 7.19: Frequencies for languages with / without modifiers and modifier clauses
DCs: modifier clauses
+ – total
modifiers expected freq 0 4
expected freq 1 33
total 1 37 38
The data in Table 7.19 make clear that there is no correlation at all between the
availability of lexical and clausal constructions that can express both modifier
functions: p = 1. In fact, there is only one language with modifier clauses,
and this language, Babungo, does not have lexical modifiers, but rather rigid
derived adjectives (and a small set of simple ones) and a small class of rigid
manner adverbs 110,111. Moreover, there are only four languages with lexical
modifiers attested in the sample, and none of these have modifier clauses.
109
These languages are Mandarin Chinese, Tamil, Nung, Krongo, Hixkaryana, Slave, Nivkh,
West Greenlandic, Nunggubuyu, and Tuscarora.
110
Note that the modifier clause construction in Babungo is balanced. This means that it
does not show any formal similarity to a lexical expression, which links up with the lack of
functional similarity.
111
There are two more languages in the sample that display a flexible modifier clause
construction: Mandarin Chinese and Krongo (see Table 7.40 in section 7.5). However, neither
of these languages is taken into account in Table 7.19, since both lack (large, open) lexical
classes to express the functions of modification, and thus do not allow for a functional
comparison between PoS and DCs.

7.3.2.5 Summary
To sum up the results for specific matches between flexible constructions,
it was shown that the most pervasively flexible (deranked) type of DC is
hardly ever attested in languages without a PoS class with the same type
of flexibility. However, this generalization is rather uninformative to the
extent that maximal flexibility in both the PoS and the DCs domains are
rare phenomena in the first place. The interesting observation is that, even
though the presence of maximally flexible PoS classes in a language does not
imply the presence of maximally flexible deranked DCs, the availability of
the former strongly increases the chances of also having the latter.
In contrast, no dependency relations were found between the (non-)
attestation of less flexible lexical and clausal constructions, i.e. between
nominals and nominal clauses, and between modifiers and modifier clauses
(neither balanced nor deranked). The number of cases with a positive value
on either the PoS or the DC parameter was again found to be low, but unlike
in the case of contentives / non-verbs and contentive / multi-functional clauses,
there was no tendency for the two parameters to have the same value.
7.3.3 Specific matches for rigid constructions
I now turn back to Prediction C in (16) above and apply this prediction to
the various types of rigid PoS classes and DC constructions. This allows for
the formulation of the three sub-predictions listed in (22):
(22) a. If a language has complement clauses, then it should also have
lexical nouns.
b. If a language has relative clauses, then it should also have lexical
adjectives.
c. If a language has adverbial manner clauses, then it should also
have lexical manner adverbs.
Note that, unlike in Chapter 4 (see (8) in that chapter), constructions

specialized for the function of head of a predicate phrase (verbs and
predicate clauses) are not taken into account, since, as shown in Chapter 6,
rigid predicate clauses are not attested in any of the sample languages.
Each of the three predictions in (22) can be further differentiated when
taking into account the parameter of structural DC type. In particular, the

prediction in (22a) can be split up into the sub-predictions listed in (23a-d)
(cf. (10) of Chapter 4). As in the previous cases, the prediction in (23a) is less
likely to be confirmed than the one in (23b), and the one in (23c) is less likely
to hold than the one in (23d).
(23) a. If a language has balanced complement clauses of type 1, then it

b. If a language has deranked complement clauses of type 2 / 3,
then it should also have lexical nouns.
c. If a language has deranked complement clauses of type 2, then
it should also have lexical nouns.
d. If a language has deranked complement clauses of type 3, it
Parallel lists of sub-hypotheses can be set up for the predictions in (22b) and
(22c) above, concerning lexical and clausal constructions that are specialized
for the function of modifier in a referential phrase, and for modifier in a
predicate phrase, respectively. In what follows, I present the results for every
rigid construction type, first considering DCs as a single group, and then
differentiating according to structural DC type.
7.3.3.2 Rigid match 1: Nouns and complement clauses

I start out with the results for the prediction in (22a), involving rigid nouns
and rigid complement clauses, without differentiating for structural DC
type. The observed frequencies are presented in Table 7.20 below. The
relevant language data can be found in Tables 41 and 42, section 7.5 112. They
do not deviate from chance frequency, so that p = 1 (2-sided).
112
Note that, as mentioned earlier, there is one language, Warao, which has two lexical
strategies available to express the function of head of a referential phrase: a class of simple
non-verbs and a class of derived nouns. To avoid double values, prominence is given to the
class of simple lexemes. This means that Warao is counted as a language without a class of
rigid nouns.

Table 7.20: frequencies for languages with / without nouns and complement clauses
DCs: complement clauses
+ – total
PoS: nouns + observed freq 30 5 35
expected freq 30 5
expected freq 9 2
total 39 7 46
Turning to the sub-predictions regarding the different structural types of

complement clauses, consider the results for the prediction in (23a), which
appear in Table 7.21. In accordance with the findings for all balanced clause
types considered so far, there is no correlation between rigid lexical nouns
and rigid complement clauses of type 1. The counts in Table 7.21 hardly
deviate from chance frequency: p = 0.730 (2-sided).
Table 7.21: Frequencies for language with / without nouns and complement clauses of type 1
DCs: complement clauses type 1
(balanced)
+ – total
expected freq 19 16
expected freq 6 5
total 25 21 46
Unexpectedly, considering deranked DCs separately does not reveal a

correlation either. The frequencies for complement clauses of type 2 / 3,
and for types 2 and 3 separately, are presented in Tables 7.22, 7.23 and 7.24,
respectively. None of these yields a significant result. For Table 7.22, p =
0.497 (2-sided); for Table 7.23, p = 0.702 (2-sided); and for Table 7.24, p =
1 (2-sided).

Table 7.22: Frequencies for language with/without nouns and complement clauses of type 2/3
DCs: complement clauses type 2 / 3
(deranked)
+ – total
expected freq 18 17
expected freq 6 5
total 24 22 46
+ – total
expected freq 10 25
expected freq 3 8
total 13 33 46
+ – total
expected freq 11 24
expected freq 4 7
total 25 31 46
In sum, no significant correlation exists between rigid nouns and rigid

complement clauses, neither for balanced nor for (different types of )
deranked clauses.
7.3.3.3 Rigid match 2: Adjectives and relative clauses

Consider now lexical and clausal constructions that are specialized for the
function of modifier in a referential phrase: adjectives and relative clauses.
The distribution pertaining to the prediction in (22b) (without differentiation

for structural DC type) is presented in Table 7.25. This table shows that the
observed frequencies are again equal to the expected frequencies, so that the
p-value is 1 (2-sided). For the relevant language data, see Tables 7.43-7.47 in
section 7.5.
Note that the total number of analyzed cases (N) in Table 7.25 is 34.
This is because there are 14 languages that either lack a class of adjectives
altogether (8 languages) or have only a small set of them (6 languages).
These 14 cases are not taken into account because they are interpreted as
lacking a construction in the PoS domain that can serve as the functional
comparative standard 113. This means that the 15 cases with a negative value
on the PoS parameter in Table 7.25 involve languages with a flexible lexeme
class that can express the function of modifier in a referential phrase 114.
Table 7.25: Frequencies for languages with / without adjectives and relative clauses
DCs: relative clauses
+ – total
adjectives expected freq 15 4
expected freq 12 3
total 27 7 34
Taking into account the parameter of structural DC type, we can formulate

the set of predictions listed in (24), in parallel to those in (23) above. The
same expectations as above hold regarding the likelihood of each of the
predictions to be confirmed: (24a) is less likely than (24b), and (24c) less
likely then (24d):
(24) a. If a language has balanced relative clauses of type 1, then it

should also have lexical adjectives.
113
One of the languages without adjectives, namely Hixkaryana, also lacks a relative clause
construction (Derbyshire 1979: 26).
114
Note that there is one language, namely Ket, which has a large, open class of flexible
modifiers, as well as a class of derived rigid adjectives, and small classes of simple adjectives
and manner adverbs. In order to avoid double values, prominence is given to the large, open
lexeme class, i.e. the modifiers, and the derived and small classes are not taken into account.
This means that Ket is counted as a language without adjectives. In the case of a choice
between a derived and a small lexeme class, the former is considered primary and the latter
is disregarded. This occurs in one language, namely Koasati, which has a class of derived
modifiers and a small class of adjectives, but is counted as a language without rigid adjectives.

b. If a language has deranked relative clauses of type 2 / 3, then it
c. If a language has deranked relative clauses of type 2, then it
d. If a language has deranked relative clauses of type 3, then it
The results for the prediction in (24a) are presented in Table 7.26. The
observed and expected frequencies deviate only very slightly from each
other, so that there is no significant correlation: p =0.724 (2-sided).
Table 7.26: frequencies for languages with / without adjectives and relative clauses of type 1
DCs: relative clauses of type 1
(balanced)
+ – total
expected freq 10 5
total 22 12 34
The results for the prediction in (24b), concerning deranked relative clauses
(type 2 and 3) are given in Table 7.27. As can be seen, there is again only a
minimal deviation between the observed and the expected frequencies: p =
0.718 (two-sided).
Table 7.27: frequencies for languages with/without adjectives and relative clauses of type 2/3
DCs: relative clauses of type 2 / 3
(deranked)
+ – total
expected freq 4 11
total 10 24q 34

In Tables 7.28 and 7.29 below the data are split up for deranked relative
clauses of type 2 and of type 3, respectively. Neither the results in Table
7.28 nor those in Table 7.29 are in accordance with the predictions: There
is no correlation between the availability of lexical adjectives and deranked
relative clauses of type 2 and 3, respectively. The p-values are 0.718 (2-sided)
for Table 7.28, and p = 1 (2-sided) for Table 7.29.
It should be mentioned that there are some languages (in particular:
Kharia, Hungarian, Kayardild, and Georgian) that have both relative clauses
of type 2 and of type 3, which is why the totals for relative clauses of Tables
7.28 and Table 7.29 add up to 14, i.e. more than the total of 10 languages
with deranked relative clause constructions in Table 7.27. Note further that
there are no languages with a relative clause construction of type 3 that do
not also have a relative clause construction of type 2. This explains why the
frequencies in Tables 7.27 and 7.28 are identical.
Table 7.28: frequencies for languages with / without rigid adjectives and relative clauses of
type 2
+ – total
expected freq 4 11
total 10 24 34
Table 7.29: frequencies for languages with / without rigid adjectives and relative clauses of
type 3
+ – total
expected freq 2 13
total 4 30 34
In sum, the same conclusion is reached for adjectives and relative clauses
as for nouns and complement clauses: There is no correlation between the

availability of lexical and (different structural types of ) clausal constructions
specialized for the function of modifier in a referential phrase.
7.3.3.4 Rigid match 3: Manner adverbs and adverbial manner clauses

Finally, consider the data for lexical and clausal constructions that are
specialized for the function of modifier in a predicate phrase: manner
adverbs and adverbial manner clauses. The distribution pertaining to the
prediction in (22c) appears in Table 7.30. The relevant language data can be
found in Tables 7.48-7.52 in section 7.5.
Note that, as in the case of adjectives, only those languages are taken into
consideration that have a large, open class of simple or derived lexemes to express
the function of modifier in a predicate phrase. Languages with a small, closed
lexical class (8 cases), or no lexical option at all (11 cases) are excluded from the
analysis. In addition, there are 3 languages with a large simple or derived class
of manner adverbs, but without a DC strategy for the function of modifier in a
predicate phrase. There are another 3 languages that do have a class of flexible
lexemes that can be used in this function, but which are excluded from the
analysis as well, since they lack a DC strategy to express this function. This
leaves us with a rather small number of relevant languages, namely N = 23.
The observed frequencies in Table 7.30 deviate from the expected ones
in the predicted way, but not significantly: p = 0.083 (2-sided).
Table 7.30: frequencies for languages with / without a (derived) manner adverbs and
adverbial manner clauses
DCs: adverbial manner clauses
+ – total
PoS: manner + observed freq 16 0 16
adverbs expected freq 15 1
expected freq 6 1
total 21 2 23
We may now formulate the specific sub-predictions for manner adverbs and
adverbial manner clauses, taking into account the different structural DC
types. These sub-predictions are listed in (25), in parallel with (24) and (23)
above, and the by now familiar expectations regarding the likelihood of their
confirmation apply.

(25) a. If a language has balanced adverbial manner clauses of type 1,
then it should also have lexical manner adverbs.
b. If a language has deranked adverbial manner clauses of type
2 / 3, then it should also have lexical manner adverbs.
c. If a language has deranked adverbial manner clauses of type 2,
d. If a language has deranked adverbial manner clauses of type 3,
The results for the prediction in (25a) appear in Table 7.31. The observed
frequencies equal the expected frequencies, so that p = 1 (2-sided).
Table 7.31: Frequencies for languages with / without (derived) manner adverbs and
adverbial manner clauses of type 1
DCs: adverbial manner clauses of
type 1 (balanced)
+ – total
(incl.
derived)
expected freq 3 4
total 11 12 23
The results for the prediction in (25b), concerning deranked adverbial

manner clauses, appear in Table 7.32. The observed frequencies deviate from
the expected frequencies in the predicted direction, but not enough to yield
a significant correlation: p = 0.193 (2-sided).

Table 7.32: Frequencies for languages with / without (derived) manner adverbs and
adverbial manner clauses of type 2 / 3
DCs: adverbial manner clauses of
type 2 / 3 (deranked)
+ – total
(incl.
derived)
expected freq 4 3
total 12 11 23
Since there is only one language in the sample with an adverbial manner
clause construction of type 3, the data are insufficient to further differentiate
within the group of deranked adverbial manner clauses.
In sum, the findings for rigid constructions specialized for the function
of modifier in a predicate phrase parallel those for the other two rigid
construction types: there is no correlation between the presence of lexical
manner adverbs and the presence of (different structural types of ) adverbial
manner clauses.
7.3.3.5 Summary
No dependency relations are identified between the availability of rigid
lexical and clausal constructions that are specialized for the expression of
the same propositional function. Differentiating between balanced and
(different types of ) deranked DCs does not influence this result.
7.4 Summary, conclusion

To conclude, the results presented in this chapter make clear that significant
correlations between the functional possibilities of PoS and DCs are found
along the following lines:
(i) For flexible rather than for rigid constructions;

(ii) For pervasively flexible constructions rather than for less flexible
constructions;
(iii) For deranked DCs rather than for balanced ones.
I have presented the following evidence for these findings: First, Tables 7.4
and 7.5a-b reveal a one-sided asymmetrical dependency between flexible

PoS and flexible deranked DCs. In particular, it was shown that flexible
deranked DCs occur almost exclusively in languages with a flexible PoS
system. Second, Tables 7.7 and 7.8a-b show the same type of dependency
relation, but specify it for deranked flexible DCs of type 3 (D-ALT).
Third, Tables 7.9 and 7.10a-b contain the data that allow for a further fine-
tuning of these findings in terms of specific flexible construction types.
In particular, these data shows that maximal flexibility at the level of DC
constructions (contentive / multi-functional clauses) is possible only when a
maximally flexible PoS class (contentives / non-verbs) is also available. More
interestingly, in view of the fact that both maximally flexible PoS classes and
maximally flexible DC constructions are rare phenomena independently of
each other, the data show that the presence of a maximally flexible PoS class
in a language strongly increases the chances of also finding a maximally
flexible DC construction in that language. On the basis of the data in Tables
7.12 and 7.13 a-b this generalization could be further specified as pertaining
to deranked contentive and multi-functional clauses of type 3 (D-ALT).
These results suggest, in sum, that the presence of (pervasive) flexibility
in the domain of deranked DC constructions is dependent on the presence
of (pervasive) flexibility in the domain of PoS classes. On the other hand,
flexibility in the PoS system of a language does not imply flexibility in the
domain of deranked dependent clauses. In Chapter 8 I will further explore
the nature of the relationship between distributional patterns of PoS and
DCs, as well as the functional factors that may motivate this relationship.
7.5 Language data
Table 7.33: Availability and type of flexible DCs in languages with one or more flexible
PoS classes
Language Flex DCs Structural Functional Type Structural
type coding
1 Tagalog + 3, 1 contentive / pag-,
multi-functional Ø
2 Kharia + 3 contentive / RDP / Ø,
multi-functional -na
3 Samoan + 1 nominal Ø
4 Guaraní – (1)
5 Warao – (1, 2)

Language Flex DCs Structural Functional Type Structural
type coding
6 Turkish + 3, 1 nominal -DIK / -(y)AcAK,
ki (borrowed)
7 Kayardild + 3, 1 multi- -n-
functional, ntha
nominal
8 Paiwan + 1 nominal a (Ø)
9 Quechua + 2 nominal -j / -shka / -na
10 Ma’di + 2/3 nominal -lɛ̄, -dƷɔ
11 Gooniyandi + 1 multi-functional Ø
12 Hungarian – (1, 2, 3)
13 Japanese – (1, 2, 1 / 3)
14 Hmong – (1)
Njua
15 Lango – (1, 3)
16 Ket + 3, 1 nominal Ø , Ø (bare INF)
17 Koasati – (2)
Table 7.34: Availability and type of flexible DCs in languages with rigid PoS classes only
Language Flex Structural Functional type Structural
DCs type coding
1 Itelmen – (1, 2)
2 Thai + 1 nominal Ø / thiî
3 Basque + 1 nominal -en, bait
4 Abun – (1)
5 Bambara – (1, 2)
6 Georgian + 1 nominal rom
7 Bukiyip – (1)
8 Abkhaz – (1, 3)
9 Polish – (1, 2, 3)
10 Burushaski + 2,1 nominal, multi- -(á)as,
functional ke / ki
11 Lavukaleve – (1, 2)
12 Alamblak – (2, 3)
14 Pipil + 1 nominal ka(h)
14 Wambon – (1, 2)
15 Dhaasanac + 1 nominal DET(+DEM)

Language Flex Structural Functional type Structural
DCs type coding
16 Berbice Dutch – (1, 2)
17 Babungo + 1 modifier fan / yúu
18 Nama – (1, 2)
19 Hdi + 3 nominal tá
20 Mandarin Ch. + 1 multi- Ø, de
functional,
modifier
21 Tamil – (2)
22 Kisi – (1, 2)
23 Nung + 1 multi-functional Ø
24 Garo – (1, 2, 3)
25 Krongo + 2/3 modifier m- / n-
26 Hixkaryana – 3
27 Slave + 1 nominal Ø
28 Nivkh – (1, 2)
29 Greenlandic + 1 nominal, PTCmood /
comp / adv CONT mood
30 Nunggubuyu none
31 Tuscarora none
Table 7.35: Availability and type of contentive / multi-functional clauses in languages

with lexical contentives / non-verbs
Language Contentive / multifunctional Type Structural
clause coding
1 Tagalog + 3,1 pag-,Ø
2 Kharia + 3 RDP, -na
3 Samoan –
4 Guarani –
5 Warao –
6 Turkish –
7 Kayardild + 3 -n-,

Table 7.36: Availability and type of contentive / multi-functional clauses in languages
without lexical contentives / non-verbs
Language Contentive / multi-functional Type Structural
clause coding
1 Paiwan –
2 Quechua –
3 Ma’di –
4 Gooniyandi + 1 Ø
5 Hungarian –
6 Japanese –
7 Hmong Njua –
8 Lango –
9 Ket –
10 Koasati –
11 Itelmen –
12 Thai –
13 Basque –
14 Abun –
15 Bambara –
16 Georgian –
17 Bukiyip –
18 Abkhaz –
19 Polish –
20 Burushaski –
21 Lavukaleve –
22 Alamblak –
23 Pipil –
24 Wambon –
25 Dhaasanac –
26 Berbice –
Dutch
27 Babungo –
28 Nama –
29 Hdi –
30 Mandarin Ch. + 1 Ø
31 Tamil –
32 Kisi –
33 Nung + 1 Ø

Language Contentive / multi-functional Type Structural
clause coding
34 Garo –
35 Krongo –
36 Hixkaryana –
37 Slave –
38 Nivkh –
39 Greenlandic –
40 Nunggubuyu none
41 Tuscarora none
Table 7.37: Availability and type of nominal clauses in languages with lexical nominals
Language Nominal clauses Type Structural
coding
1 Quechua + 2 -j / -shka / -na
2 Ma’di + 2/3 -lɛ̄, -dƷɔ
3 Gooniyandi –
4 Hungarian –
5 Japanese –
Table 7.38: Availability and type of and nominal clauses in languages without lexical
nominals
Language Nominal clause Type Structural coding
1 Tagalog –
2 Kharia –
3 Samoan + 1 Ø
4 Guarani –
5 Warao –
6 Turkish + 3, 1 -DIK / -(y)AcAK, ki
(borrowed)
7 Kayardild + 1 -ntha
8 Paiwan + 1 a
9 Hmong Njua –
10 Lango –
11 Ket + 3, 1 Ø, Ø (bare INF)
12 Koasati –

Language Nominal clause Type Structural coding
13 Itelmen –
14 Thai + 1 Ø, thiî
15 Basque + 1 -en
16 Abun –
17 Bambara –
18 Georgian + 1 rom
19 Bukiyip –
20 Abkhaz –
21 Polish –
22 Burushaski + 2,1 -(á)as, ke / ki
23 Lavukaleve –
24 Alamblak –
25 Pipil + 1 ka(h)
26 Wambon –
27 Dhaasanac + 1 DET(+DEM)
28 Berbice Dutch –
29 Babungo –
30 Nama – (1)
31 Hdi + 3 tá
32 Mandarin Ch. –
33 Tamil –
34 Kisi –
35 Nung –
36 Garo –
37 Krongo –
38 Hixkaryana –
39 Slave + 1 nį
40 Nivkh –
41 Greenlandic + 1 PTCmood
42 Nunggubuyu none
43 Tuscarora none
Table 7.39: Availability of modifier clauses in languages with (derived) lexical modifiers
Language Modifier clause
1 Hmong Njua –
2 Lango –

Language Modifier clause
3 Ket –
4 Koasati –
Table 7.40: Languages without (derived) lexical modifiers, but with modifier clauses
Language Modifier clause Type Structural coding
1 Babungo + 1 fan / yúu
2 Mandarin Ch. + 1 de
3 Krongo + 2/3 m- / n-
Table 7.41: Availability and type of complement clauses in languages without nouns
Language Rigid complement clauses Type Structural coding
1 Tagalog –
2 Kharia + 1 no, gam-kon
3 Samoan + 3 -ga, unmarked
NMLZ
4 Guaraní + 1 há(gwe)
5 Warao + 2 -kitane
6 Turkish + 1, 2, 3 -ma, -maK, diye, Ø
7 Kayardild –
8 Quechua + 2 -y, ngapay
9 Ma’di + 1, 2 / 3 ka, Ø
10 Gooniyandi + 2 -woo
11 Hungarian + 3, 1 -ni, -ás / -és, hogy
Table 7.42: Availability and type of complement clauses in languages with nouns
1 Paiwan + 1 tu(a) / tjai
2 Ket –
3 Japanese + 1/3 no / mono
4 Hmong Njua + 1 qhov / kuam /
(has)tas
5 Lango + 1, 3 nî,-(kk)ɔ̀
6 Koasati + 2 NMLZ (various
forms)

7 Itelmen + 1, 2 Ø, INF (various
forms)
8 Thai + 1 wâa
9 Basque + 2 -t(z)e
10 Abun + 1 do / Ø
11 Bambara + 1 ka
12 Georgian + 3 -a
13 Bukiyip + 1 Ø
14 Abkhaz + 1, 3 NFIN / -ra
15 Polish + 1, 2, 3 ze, INF, -nie
16 Burushaski –
17 Lavukaleve + 2 -e / -i
18 Alamblak + 3 -nef / -(kfë)t
19 Pipil –
20 Wambon + 1 -e
21 Dhaasanac + 3 -ɲ / -an
22 Berbice + 1, 2 bifi / dati, fu / fi
Dutch
23 Babungo + 1 l āa
24 Nama + 1 !xáis-à
25 Hdi + 1 ká
26 Mandarin Ch –
27 Tamil + 2 -atu, -(kk)a
28 Kisi + 1, 2 (m)àà, Ø (INF)
29 Nung –
30 Garo + 1, 3, 2 in-e, -a, -a-ni, -na,
-kan-a
31 Krongo + 1,2 / 3 àni tiŋ, (t)-
32 Hixkaryana + 3 nɨ / -thɨ- / -hɨto + rɨ
33 Slave + 1 gú, ghá
34 Nivkh + 1 -vut / -vur, Ø
35 Greenlandic + 3 -niq
36 Nunggubuyu none
37 Tuscarora none

Table 7.43: Availability and type of relative clauses in languages without adjectives, but
with a flexible lexical strategy for the function of modifier in a referential phrase
Language Rigid Type Structural Lexical strategy
relative coding
clauses
1 Tagalog – contentives
2 Kharia + 2, 3, 1 na-wala, -al, contentives
je, a / i
3 Samoan + 1 -e contentives
4 Guaraní + 1 va contentives
5 Warao + 1 kotai non-verbs
6 Turkish + 2 -An non-verbs
7 Quechua – nominals
8 Ma’di + 2 -rɛ̄ / -ɓá nominals
9 Gooniyandi – nominals
10 Hungarian + 2, 3, 1 -ó, óttt, andó, nominals
RelPron
11 Japanese + 1 Ø (RSP) nominals
12 Hmong Njua + 1 kws modifiers
13 Lango + 1 à-mɛ̀ modifiers
14 Ket + 1 RelPron, -s / -bes modifiers, S / D
adjectives
15 Koasati + 2 various forms D modifiers, S
adjectives
Table 7.44: Availability and type of relative clauses in languages with an open class of
simple adjectives

relative coding
clauses
1 Kayardild + 1, 2, 3 Ø, -Thirri-n, adjectives
-n-garrba
2 Thai – adjectives (S.
modifiers)
3 Basque – adjectives
(S / modifiers)
4 Abun + 1 gato adjectives (S.
modifiers)

relative coding
clauses
5 Georgian + 1, 2, 3 Several adjectives
participles, REL
PRON
6 Bukiyip + 1 (ú)li adjectives
7 Abkhaz + 1 REL.PRON + adjectives
NFIN
8 Polish + 1, 2 Several adjectives
participles, REL
PRON
9 Burushaski + 2 -im / -um / -am adjectives
10 Lavukaleve + 1 Person suffix adjectives
11 Alamblak + 2 ɨnd adjectives
12 Pipil + 1 ne, ke adjectives
13 Wambon + 1 -a + o adjectives
14 Dhaasanac – adjectives
15 Berbice + 1 Wh adjectives
Dutch
Table 7.45: Availability and type of relative clauses in languages without an open class
of simple adjectives, but with derived adjectives (and a small class of simple
adjectives)
relative coding
clauses
1 Itelmen + 1 min D adjectives,
S modifiers
2 Nama + 1 Ø no INDIC S / D adjectives
(RSP)
3 Babungo – S / D adjectives
4 Kisi + 1 CL S / D adjectives

Table 7.46: Availability and type of relative clauses in languages without an open class of
simple / derived class adjectives, but with a small class of adjectives
relative coding
clauses
1 Paiwan + 2 in- an + a S adjectives
2 Bambara + 2, 1 -le / -ne, S adjectives
-min(u) /
-mun(u)
3 Hdi + 1, 2 tà, tà + NMLZ, S adjectives
-a
4 Mandarin Ch – S adjectives
5 Tamil + 2 -a S adjectives
6 Nung + 1 (tị-vạ) (+DEM) S adjectives
(+FOC)
Table 7.47: Availability and type of relative clauses in languages without a lexical
strategy for the function of modifier in a referential phrase
relative coding
clauses
1 Garo + 3 -gip-a X
2 Krongo – X
3 Hixkaryana none X
4 Slave + 1 i, sį́i, líi X
5 Nivkh + 1 NONFIN X
6 West + 3 -ta / -sa X
Greenlandic
7 Nunggubuyu none X
8 Tuscarora none X

Table 7.48: Availability and type of adverbial manner clauses in languages without
an open class of simple / derived manner adverbs, but with a flexible lexical
strategy for the function of modifier in a predicate phrase (and in some cases
also a small class of rigid manner adverbs)
adverbial coding
manner
clauses
1 Tagalog – Flex
(contentives)
2 Kharia + 2, 3 -ker, -ga, -ta Flex
+RDP (contentives)
3 Samoan none Flex
(contentives) +
S MAdv
4 Guaraní + 1 vo Flex
(contentives)
5 Warao none Flex (non-verbs)
6 Turkish + 2 -(y)ArAK, -(y) Flex (non-verbs,
A..-(y)A D modifiers)
7 Kayardild – Flex
(non-verbs /
predicatives)
+ S MAdv
8 Paiwan + 1 a parhu Flex
(predicatives)
+ S MAdv?
9 Hmong Njua none Flex (modifiers)
10 Ket + 1 ɛta qɔr’a, ásqà, Flex (modifiers)
-PROSEC + S MAdv

Table 7.49: Availability and type of adverbial manner clauses in languages with an open
class of simple manner adverbs
adverbial coding
manner
clauses
1 Gooniyandi + 2 -wadda, MAdv
-mawoo, -bari,
-ya / -gowaaya
2 Hungarian + 1, 2 úgy (a-)hogy, MAdv
-vá / -vé, -vén
3 Japanese + 2 -te / -de / -ite, MAdv
-i / -Ø
4 Lango none MAdv
5 Abun + 1 sa gato MAdv
6 Bambara + 2 -tò MAdv
7 Georgian + 1 ra, rogorc MAdv
8 Bukiyip + 1 bwidou -(u)mu MAdv
Table 7.50: Availability and type of adverbial manner clauses in languages without an
open class of simple manner adverbs, but with derived manner adverbs (and
in some cases a small class of simple manner adverbs)
adverbial coding
manner
clauses
1 Ma’di + 2 -zɨ̄ + sɨ̀ S / D MAdv
2 Koasati + 2 -n, -k, -t S / D MAdv
3 Itelmen + 1 QUATZ D MAdv
4 Thai none D MAdv
5 Basque + 1 en bezala D MAdv
6 Abkhaz + 1 -ş- + -NFIN S / D MAdv
7 Polish + 1, 2 PRS.PL-c-, jak S / D MAdv
(gdy)by
8 Burushaski + 2 n-STEM-(a)n S / D MAdv
9 Nama + 2 se / !’aa / tsii (no D MAdv
INDIC)
10 Hdi none S / D MAdv
11 Garo + 2 -e / -e-min / -e-r S / D MAdv

Table 7.51: A
vailability and type of relative clauses in languages without an open class of
simple / derived class manner adverbs, but with a small class of manner adverbs
adverbial coding
manner
clauses
1 Quechua + 2 -shpa, -y RDP S Madv
2 Lavukaleve – S Madv
3 Alamblak – S Madv
4 Babungo + 2 kɨ̀(i) / Ø S Madv
5 Mandarin Ch. + 2 zhe S Madv
6 Tamil – S Madv
7 Krongo – S Madv
8 Hixkaryana – S Madv
Table 7.52: A
vailability and type of adverbial manner clauses in languages without a
lexical strategy for the function of modifier in a predicate phrase
adverbial coding
manner
clauses
1 Pipil + 1 ke:n-aken, ADV X
2 Wambon + 1, 2 -no / -o, ka X
3 Dhaasanac – X
4 Berbice – X
Dutch
5 Kisi – X
6 Nung + 1 b ạt X
7 Slave + 1 gháré X
8 Nivkh + 2 r / -t-ř / / -n X
9 West – X
Greenlandic
10 Nunggubuyu none X
11 Tuscarora none X

Discussion: Flexibility and
Functional Transparency
8.1 Introduction
The results presented in the previous chapter suggest that there is a
dependency relation between the amount of flexibility displayed by the
parts of speech classes of a particular language, and the amount of flexibility
displayed by the dependent clause constructions of that language. More
specifically, the data show that, as expected, the possibility of having
(pervasively) flexible deranked DC constructions in a language is dependent
of the presence of (pervasively) flexible PoS classes in that language.
However, there were also a number of unexpected findings. First, not all
languages with flexible PoS classes also have flexible deranked DCs. Morever,
those that do have flexible deranked DCs combine them with rigid (deranked
or balanced) DCs. Second, there are no dependency relations between
lexical and (deranked) clausal constructions with less pervasive flexibility, i.e.
between nominals and nominal clauses, and between modifiers and modifier
clauses. Third, the presence of specific rigid DCs is not dependent upon the
presence of parallel types of rigid PoS classes. Finally, there are in general no
correlations between the functional possibilities of PoS and balanced DCs.
The aim of the present chapter is to interpret the results of Chapter
7 from a specific functionalist perspective. In particular, recent functional-
typological research has advanced the idea that flexibility and rigidity are
relative rather than absolute notions, which may be applicable in various
degrees to specific construction types at different levels of grammar. These
313
studies suggest that the categorial specificity of linguistic constructions
increases – or their flexibility decreases – when they become structurally
more complex (Haig 2006, Lehmann 2008). This generalization can
be regarded as a specific instance of a more general functional principle
regarding complexity in language systems, namely that flexibility or multi-
functionality in one area of the grammar must be counterbalanced or ‘traded
off ’ by rigidity or categorial specificity in another area, in order to guarantee
the functional identifiability of linguistic units within an actual utterance.
The structure of this chapter is as follows. I start out, in section 8.2,
with a more extensive discussion of the relevant literature. Subsequently,
in section 8.3, I apply the hypothesis that categoriality increases with
grammatical complexity to the case at hand: (simple) PoS classes and
(complex) dependent clauses. It will be shown that this approach accounts
for the functional behaviour of virtually all deranked DCs attested in the
sample. In section 8.4 I turn to balanced DCs, and show that, even though
they do not behave in accordance with the specific hypothesis of increased
complexity / categoriality, they do fit into a larger explanatory picture based
on the principle of functional transparency. I conclude, in section 8.5, that
the functionalist perspective developed in the present chapter provides a
complete account of the data.
8.2 Theoretical background:

Trade-off effects between flexible and rigid constructions
Recently, the debate in the functional-typological literature about the
existence or non-existence of ‘truly’ flexible languages (see Chapter 2,
section 2.5) has taken a new turn. In particular, it has been suggested that
flexibility (or pre-categoriality) and rigidity (or categorial specificity) are not
absolute notions that do or do not apply to complete language systems, or
even sub-systems. Rather, they are viewed as gradable notions that may
apply in certain degrees to individual constructions, pertaining to specific
levels of grammar.
A first proposal to this effect is Haig’s (2006). He distinguishes four
grammatical levels at which any language may display a certain amount of
categorial specificity: First, there is the basic level of so-called radical elements.
At this level, languages may differ in their distinctions of major lexical
categories. Second, at the level of input of lexical derivation, languages may
display variation in the extent to which their derivational processes select
certain base categories but not others. At the third level we find output of

lexical derivation, which concerns the degree to which derivational processes
result in categorially specified forms. At the final, highest level of complexity,
languages may vary in terms of the categorial distinctions drawn in inflectional
and other syntactic categories (Haig 2006: 44). In short, languages differ in
the amount and type of categorial specificity (or lack of it) displayed by their
constructions at different levels of grammatical complexity.
Haig (2006) puts forward the hypothesis that this variation is not
random, but rather constrained by what he terms the Principle of Successively
Increasing Categorisation (henceforth PSIC), and formulates as follows:
“The extent to which different levels of linguistic structure impose

categorial distinctions increases monotonically as we move to the right
of the following hierarchy:
Radical elements > Derivational input > Derivational output >
Inflection / Syntax”
(Haig 2006: 45)
Haig discusses a number of languages that conform to the PSIC, and also
points out some possible counterexamples to it 115. In general, however, he
suggests that cross-linguistic differences in ‘flexibility’ can be captured as
points on a cline defined by the hierarchical complexity level at which maximal
categorial specificity is reached in a language. At one extreme of this cline,
we find early-categorizing languages, in which categorial distinctions are
built into the inventory of radical elements 116, and are observed throughout
the morphology and projected into the syntax. At the other extreme we find
late-categorizing languages, in which the categorization of linguistic elements
happens only through their syntactic or phrase-structural configurations.
In between these two extremes we may find languages in which categorial
115
Interestingly, the counterexamples that Haig (2006) mentions are all flexible languages:
Tagalog, Mundari, and Riau Indonesian (Gil 2000). As Haig (2006: 46) puts it, in these
languages “the syntax unpacks categorial distinctions drawn in morphology, rather than refining
them”. However, as I have argued in Chapter 2, the morphological categories of flexible
languages are not defined in terms of phrase structure. Therefore, categorization at the
morphological level is not only of a different degree than categorization at the syntactic level,
but also of an entirely different nature (see also Himmelmann (2007) on Tagalog).
116
Here the question becomes relevant whether or not one assumes that, at some level of
abstraction, all languages have a stock of un-categorized roots (cf. Arad 2003). For the
present purposes, it is not necessary to take a specific stance in this respect; the general point
is that the degree of categorial distinctiveness is supposed to be lower at the level of roots
than at the level of stems, words, and phrases. For more discussion see also Haig (2006), who
provides an overview of relevant literature since Aronoff (1976).
Chapter 8 – Discussion: Flexibility and Functional Transparency | 315

distinctions are introduced at intermediate (morphological) levels, and / or
in which lower level categorial distinctions become more fine-grained when
moving upwards along the complexity hierarchy.
Lehmann (2008) arrives at a very similar generalization on the basis of
data from a small sample of six languages (German, English, Spanish, Latin,
Yucatec Maya, and Mandarin Chinese). In particular, Lehmann finds that,
for each language in this sample, the degree of categoriality at the level of
stems is higher than at the level of roots (cf. Haig’s ‘radical elements’). In
addition, Lehmann argues that every linguistic unit has a unique category at
the sentence level, so that “we can safely generalize that categoriality increases
with the grammatical levels.” (Lehmann 2008: 561) The term that Lehmann
uses for this generalization is staggering level-dependent categoriality, which I
will abbreviate as SLC.
Interestingly, Lehmann (2008: note 17, p. 565) points out a possible
contradiction between his SLC principle and Dik’s (1985, 1997) Principle of
Formal Adjustment (PFA), the latter of which has been discussed in Chapter
3 (section 3.3.3.3). Lehmann argues that if, on the one hand, categoriality
increases with grammatical levels, “this would invite the inference that higher
levels dictate the necessary categories, and categorizations made at lower levels
follow that model”. Dik’s PFA, on the other hand, claims that derived,
secondary constructions are coined on the basis of less complex, primary
constructions. This, as Lehmann puts it, “seems to entail that categories of lower
levels serve as models for categorization at higher levels”. Lehmann does not
propose a solution to this problem, but I believe that, from the perspective
of the present chapter, there are two points worth making.
First, consider the possibility that higher level categories determine
lower level ones. Lehmann argues that the ultimate level of categorization
of linguistic structures is determined by the propositional functions that
must be performed within an actual utterance, i.e. reference, predication,
and modification. According to him, this level of categorization constitutes
the syntactic level. However, it seems that Lehmann confuses propositional
functions with syntactically complex (rather than morphologically complex
or lexically simplex) linguistic constructions that may express these
functions. In other words, reference, predication, and modification are
not constructions themselves; they are functions that may be expressed by
constructions of different degrees of structural complexity.
Note that it is not necessary to measure the level of categorial specificity
of linguistic constructions relative to some universal set of functions, which

would determine the maximal degree of categorical distinctiveness in any
language. It is quite possible to assume that the ultimate set of relevant
functional distinctions is a language-specific one. This still allows one to
compare the set of function(s) that can be expressed by one construction
with the functional possibilities of another construction that has a higher or
lower degree of structural complexity.
A second issue concerns the status of derived, secondary constructions
as they figure in Dik’s PFA. The PFA predicts that derived, secondary
constructions will be formally modelled on their non-derived, primary
counterparts. However, it is not always obvious to what extent a more
complex construction of a particular type can be regarded as derived from
or secondary in relation to a simpler construction. The soundest basis on
which to decide this would probably be a functional one: two constructions
can be regarded a primary and secondary instantiation of the same category
if they perform the same function(s). Obviously, however, this undermines
the possibility to compare the functional possibilities of the two construction
types, and as such leaves the Principle of Increasing Categoriality toothless.
Therefore, a formal criterion is required in order to establish a derivational
(primary-secondary) relation between two constructions, before the nature
of their functional connection can be the subject of any testable prediction.
Haig (2006) and Lehmann (2008) have tested the hypothesis of
Increasing Categoriality on data that involve derivations from simple roots
to (derived) stems, and from stems to (inflected) words. In these cases,
the formal connection between the simpler and the more complex form is
quite clear. In the case of DCs, in contrast, such a derivational connection
is less straightforwardly established. In particular, while deranked DCs are
formally modelled on lexical or phrasal expressions, balanced DCs are not;
the latter retain the formal properties of independent clauses. In view of
this difference, I will assume that deranked but not balanced DCs can be
regarded as secondary constructions in the sense of Dik’s PFA 117.
The assumption behind the PFA is that the formal similarity between
a primary and a secondary construction reflects a functional similarity.
However, taking into account the generalizations pointed out by Lehmann
and Haig, this hypothesis should be further specified: In particular, it is
117
This difference between deranked and balanced DCs was already built into the hypotheses
formulated in Chapter 4 and tested in Chapter 7, to the extent that a weaker functional match
was expected between PoS and balanced DCs than between PoS and deranked DCs. In fact
however, no significant correlations at all were found between PoS and balanced DCs.

expected that the functional possibilities of secondary, derived constructions
(deranked DCs) are not necessarily identical to those of their primary
bases (PoS classes), but may also be – and usually are – more restricted. In
other words, Lehmann’s observation that the PFA suggests an influence
of the degree of categoriality of lower-level (or primary) constructions on
the categoriality of higher-level (secondary) constructions seems justified,
but this influence is not expected to take the form of direct modelling or
copying. Rather, the simple, primary construction determines the minimal
level of categoriality (or maximal level of flexibility) of the secondary, more
complex construction.
Thus, in general, the degree of categorial distinctiveness displayed by
a particular language at lower levels of complexity determines the amount
of categorization that needs to be implemented at higher levels, in order
to achieve maximal categorial distinctiveness at the level of the utterance.
Lehmann describes this as follows:
“In a language with low stem categoriality, the speaker’s task of assigning
the words to categories is fully achieved only at the level of syntax. This
is done by inserting words into certain syntactic templates, which force
a syntactic category on them. In a language with high stem categoriality
the syntax contributes nothing to the categorization of words, which
means that such templates play a minor role in constructions.” (Lehmann
2008: 557) 118
In fact, this idea is advanced already in an earlier study, namely Hengeveld

et al. (2004). In particular, they predict that:
“Languages with flexible lexemes will have certain morpho-syntactic

strategies at their disposal that provide the hearer with clues as to the
correct interpretation of the flexible lexeme. Morpho-syntactically, there
are basically two possibilities to guide the hearer as regards the intended
function of a flexible lexeme in the actual linguistic expression:
118
Lehmann describes the implementation of categorical specificity as a task for the speaker.
In fact, however, in order for this functional motivation to work, one should also take into
account the perspective of the hearer: He or she should be able to identify the right function
of each form. This point is acknowledged in the work of Hengeveld et al. (2004) (see below).

(i) morphological markers in the immediate linguistic context; and (ii)
syntactic patterning.”
(Hengeveld, Rijkhoff & Siewierska 2004: 547-548)
This prediction is strongly supported by data from a balanced sample of

50 languages. In particular, it is shown that languages with flexible PoS
resolve the functional ambiguity of their lexemes either by means of word
order restrictions or by means of special markers indicating the functions in
which lexemes appears. Note that this is an implicational universal in the
weak sense, since languages with rigid lexeme classes may also make use of
fixed word order or function-indicating markers. Thus, Hengeveld et al.
(2004) interpret their results as evidence for a trade-off effect between the
lexical, morphological, and syntactic coding means that a language has at
its disposal. In particular, these coding means must together ensure that the
function of any linguistic units is identifiable within an utterance. While
this means that functional ambiguity must be resolved, it does not mean
that redundant marking is necessarily excluded (see also Van Lier 2006, and
cf. Chapter 9).
Functional trade-off effects have also been studied for more specific
functional sub-domains. For instance, Sinnemäki (2008) finds such an effect
between three types of strategies for core argument encoding: fixed word
order, head marking, and dependent marking. In addition, and in line with
the findings of Hengeveld et al. (2004), Sinnemäki suggests that languages
adhere more strongly to distinctiveness than to economy, which is another
way of saying that they allow for redundancy more than for ambiguity. Finally,
Sinnemäki emphasizes that complexity trade-off should not be regarded as an
all-encompassing principle, but rather as one of several factors that motivate
certain aspects of the encoding of functions in a particular domain.
Similarly, Frajzyngier & Shay (2003) present their Principle of Functional
Transparency as an explanatory mechanism in the division of labour between
coding devices pertaining to different grammatical levels, in the expression
of a certain functional domain. In particular, their principle states that:
“Every constituent must have a transparent function within the utterance.

[…] This is achieved either through the inherent properties of lexical
items in the clause or through the system of grammatical means, which
may include affixes, adpositions, linear order and free morphemes with
grammatical functions.” (Frajzyngier & Shay 2003: 4-5)

Frajzyngier & Shay (2003: 7) explicitly do not assume a universal level of
maximal categorial specificity. Rather, in their view, functional transparency
is determined relative to the set of functional distinctions that are relevant in
a particular language. As mentioned above, such a stance does not preclude
the possibility to find cross-linguistic generalizations with respect to the
way in which the workload of categorization is divided over the available
grammatical resources.
To sum up, recent functional-typological literature suggests that
there is a relationship of interdependence or trade-off between different
construction types, pertaining to different levels of grammatical complexity,
as regards their contribution to the ultimate goal of establishing functional
transparency at the utterance level. In the next two sections (8.3 and 8.4) I
will discuss the results from Chapter 7 against this theoretical background.
8.3 Applying the complexity hypothesis to PoS and

deranked DCs
8.3.1 Hypothesis
As explained in the previous section, deranked DCs can be regarded
as secondary constructions, derived from primary, underived lexical
constructions, in terms of the Principle of Formal Adjustment. Secondary,
derived constructions are assumed to be more complex than primary,
underived ones. Therefore, the Principle of Increasing Categorization or
Staggering Level-dependent Categoriality, which states that categoriality
increases with complexity, predicts that:
(1) In any language, the functional flexibility of a PoS class is

either greater than or equal to the flexibility of a deranked DC
construction that can be used in at least one of the same function(s).
This hypothesis may be formally represented as a constraint, in the following

way:
(2) [flex PoS ≥ flex deranked DC]
There are three possibilities concerning the functional possibilities of a

particular deranked DC that are in accordance with the constraint in (2).
They are listed in (3):

(3) a. A deranked DC has the same amount of flexibility as the
relevant PoS class;
b. A deranked DC has less flexibility than the relevant PoS class;
c. A deranked DC has no flexibility (i.e. it is a rigid deranked
DC).
The constraint in (2) is violated when a deranked DC has more flexibility

than the relevant PoS class, i.e. when the DC can express more propositional
functions than the PoS class.
Note that the situations described in (3b-c) would constitute
counterexamples to some of the hypotheses tested in Chapter 7, but do not
count as such in view of the prediction in (1) / (2). Consider first the situation
in (3b): It would not constitute a counterexample to the hypotheses about
global matches between flexibility in PoS systems and deranked DCs, since
for a global match it is not necessary that flexible PoS and DCs can express
exactly the same type and amount of functions. However, it would indeed
constitute a counterexample to the hypotheses in Chapter 7 about one-to-
one functional matches between specific types of flexible PoS classes and
flexible deranked DC constructions (see the predictions in (18b-d), and
(21b-d) of Chapter 7). Furthermore, situations like those described in (3c)
would constitute counterexamples to all hypotheses in Chapter 7, whether
they make reference to global matches between flexibility in PoS and DC
systems, or to specific matches between particular flexible PoS classes and
flexible deranked DC constructions.
Note further that in cases where the relevant flexible PoS class can express
two functions (i.e. when it is a class of nominals or modifiers) and the DC
only one, the difference between situation (3b) and (3c) becomes irrelevant:
When the DC can express one function less than the PoS class, then there
is only one function left, so that the DC construction is automatically rigid
rather than flexible. I will return to this issue in more detail below.
Furthermore, recall that virtually all languages with a flexible PoS system
have one or more rigid PoS classes. It is predicted that, whenever a language
has a deranked DC construction that is used in a function for which a rigid
PoS class is available, this deranked DC will be rigid too. In other words,
if there is already maximal categoriality at the level of the simple, primary
construction (PoS), then this situation should be retained at the level of the
derived, secondary construction (deranked DC).

The same prediction, of course, applies to languages with a rigid PoS
system, i.e. a system consisting of rigid classes only: Their deranked DCs,
if they have any, are expected to all be rigid too. Note that, whenever a
language lacks a lexical category for the expression of one or more particular
function(s), there is no basis on which to predict the functional pattern of
DC(s) appearing in any of those functions.
In the next sub-section I present the results for the hypothesis formulated
in (1) / (2). I first discuss languages with flexible PoS systems (section 8.3.2.1),
followed by languages with rigid PoS systems (section 8.3.2.2).
8.3.2 Results
8.3.2.1 Distribution of deranked DCs in languages with flexible PoS

systems
Introduction
Table 8.1 below lists all languages in the sample with flexible PoS classes, and
compares the functional possibilities of these PoS classes with the functional
possibilities of deranked DCs (if the language has any) occurring in one
or more of the same function(s). The rightmost column indicates either
that a particular DC has the same functional distribution as the relevant
flexible PoS (=), or that it can express a smaller amount of functions than
the flexible PoS (>).The data in Table 8.1 thus show that the hypothesis
in (1) / (2) above is confirmed in all cases (with the possible exception of
one case that will be discussed further below), since there are no deranked
DCs with a higher degree of flexibility than the relevant PoS class (<). The
figures between the brackets in the rightmost column of Table 8.1 indicates
the difference in flexibility between the DC and the PoS, in terms of the
amount of expressible functions: (1) means that the DC can express one
function less than the PoS, (2) means two functions less, and (3) means three
functions less.
In short, there are some cases in which a flexible PoS class and a
flexible DC construction can express exactly the same set of functions. This
corresponds to the situation described under (3a) above. In other cases, the
DC can express a subset of the functions that are expressed by the PoS class.
These latter cases can be subdivided into two groups. In accordance with (3b)
above, there is one group of cases in which the DC can express less functions

than the PoS class, but still remains flexible. And, in accordance with (3c)
above, there is another group of cases in which the subset of functions that
the DC can express is a singleton set, so that the DC is in fact a rigid rather
than a flexible construction. In what follows I discuss and illustrate these
three types of cases in turn.
Table 8.1: Languages with a flexible PoS class, and the distribution of deranked DCs that
appear in at least one of the same function(s)
Language Relevant Deranked DCs in the same function(s)
flexible PoS Form Function Type Flex PoS ≥
Flex DC?
1 Tagalog contentives pag- contentive/ 3 =
2 Kharia contentives RDP/Ø contentive 3 =
-na multi-functional 3 > (1)
-na-wala relative 2 > (3)
-al relative 3 > (3)
-ker relative 3 > (3)
-ga adverbial 2 > (3)
manner
-ta + RDP adverbial 2 > (3)
manner
3 Samoan contentives -ga/Ø complement 3 > (3)
ona/ina complement 2 > (3)
4 Warao non-verbs -kitane complement 2 > (2)
5 Turkish non-verbs -DIK/ nominal 3 > (1)
-AcAk,
-mA complement 3 > (2)
-mAk complement 2 > (2)
-An relative 2 > (2)
-(y)ArAk adverbial 2 > (2)
manner
-(y)A…. adverbial 2 > (2)
-(y)A manner
6 Kayardild non-verbs -n- multi-functional 3 =
-Thirri-n relative 3 > (2)
-ngarrba relative 2 > (2)

flexible PoS Form Function Type Flex PoS ≥
Flex DC?
7 Quechua nominals -shka nominal 2 =
-ngapaj complement 2 > (1)
-chun complement 2 > (1)
-y complement 2 > (1)
8 Ma’di nominals -lɛ̄, -dƷɔ nominal 2/3 =
-ka complement 3 > (1)
-rɛ̄/-ɓá relative 2 > (1)
9 Gooniyandi nominals -woo complement 2 > (1)
10 Hungarian nominals -ni complement 2 > (1)
-ás/-és complement 3 > (1)
-ó relative 2 > (1)
-ótt relative 3 > (1)
-andó/ relative 2 > (1)
-endó
11 Japanese nominals no/mono complement 3 > (1)
12 Ket modifiers Ø (bare nominal 2 > (1)
INF)
13 Koasati modifiers -n adverbial 2 > (1)
manner
-k adverbial 2 > (1)
manner
-t adverbial 2 > (1)
manner
Deranked DCs with the same distribution as flexible PoS
Table 8.1 shows that there are five cases in which a flexible PoS class and a
flexible deranked DC construction can express exactly the same type and
amount of functions (as signalled by the = sign). An example of such a case
is Kayardild, which has lexical non-verbs and deranked multi-functional
clauses. The latter are demonstrated in (4a-c, cf. (33a-c) in Chapter 6), where
the same DC construction, marked by -n-, appears in the functions of head
of a referential phrase (a), modifier in a referential phrase (b), and modifier
in a predicate phrase (c):

Kayardild (Evans 1995: 472, 474)
(4) a. Ngada kurri-ja [ki-l-wan-ji
1sg.nom see-act 2-pl-poss-mloc
dalwani-n-ki thawal-urrk]
dig.up-nmlz-mloc yam-mloc:a.obl
‘I saw you digging up yams.’
b. Nga-ku-l-da [wirr-n-ku] dangka-wu

1-inc-pl-nom dance-ptc-m.prop man-m.prop
kurri-ju
see-pot
‘We will watch the dancing man.’
c. [Bilaangka-nurru kari-i-n-da] ngada warra-j

blanket-assoc cover-m-conv-nom 1sg.nom go-act
‘I went along, covering myself in a blanket.’
Deranked DCs with less flexibility than flexible PoS
In all other cases listed in Table 8.1, the DC can express a subset of the
functions of the PoS class (as indicated by the > signs). Two of these cases
correspond to the situation described in (3b) above: They involve DCs that
can express less functions than the relevant PoS classes, but are nevertheless
still flexible (i.e. can be used in at least two functions). These cases are
multifunctional -na clauses in Kharia and nominal -DIK / -AcAK clauses in
Turkish. The former DC construction can express three functions, which
is one function less than lexical contentives in Kharia. This distribution of
the -na construction is illustrated in (5a-c) below, where it appears in the
functions of head of a referential phrase (a), modifier in a referential phrase
(b), and modifier in a predicate phrase (c). The second DC construction,
Turkish nominal clauses, can express two functions, which is one function
less than the lexical non-verbs in this language. This DC has already been
illustrated several times (see Chapter 3 examples (23a-b); Chapter 6 example
(37a-b); and Chapter 7 examples (8a-b)).

Kharia (Peterson 2006: 259, 243, 249)
(5) a. [iɲ u ikuɁɖ sundar kontheɁɖ=ki=te
1sg this very beautiful bird=pl=obl
‘I don’t want to torment these beautiful birds.’
b. [ɖoli=te ɖoɁɖ=na] bhere

palanquin=obl take=inf time
‘the time [they were] to take away the palanquin’
c. …lekin [lam=na lamna] souɁb=te ikuɖ jughay

but search=inf rdp all=obl very much
ɖaɁ-piyas laɁ=ki.
water-thirst emot=m.pst
‘But searching and searching, [they] all became very thirsty.’
Full and partial retention of flexibility in DCs: Chapter 7 revisited
Above I discussed the cases in Table 8.1 that represent the situations described
in (3a) and (3b) above: Deranked DCs that have the same amount of flexibility
as the relevant flexible PoS classes, and deranked DCs that are less flexible
than the relevant PoS classes, but still flexible. I want to emphasize here that
these are the cases behind the general finding of Chapter 7 that flexibility in
the domain of deranked DCs is dependent upon flexibility in the domain of
PoS classes. This finding can now be reinterpreted as follows: The maximal
amount of flexibility (or the minimal amount of categorial specificity) of
deranked DCs in a language is constrained by the amount of flexibility (or
categorial specificity) displayed by the PoS classes in that language.
Moreover, and as already pointed out in section 8.3.1, those cases in
Table 8.1 which involve a PoS class that expresses two functions and a DC
that expresses only one function can be interpreted as instances of the
situation in (3b) as well, since the DCs express one function less than the
PoS classes. This holds for all DCs in Table 8.1 that are marked with > (1)
from Quechua down to Koasati 119. If these cases are accordingly counted
as representing partial loss of flexibility as compared to the relevant PoS,
then the following picture emerges: Among the total 13 languages in Table
8.1 with flexible PoS classes and at least one deranked DC (as opposed to

balanced clauses only), there are 11 in which this deranked DC can either
express all the functions that can be expressed by the relevant PoS class, or
one function less 120. The situation in these 11 languages is summarized in
Table 8.2 below. In terms of PoS classes (see the two leftmost columns), a
distinction is made between languages with very flexible PoS (contentives
or non-verbs) and languages with less flexible PoS (nominals or modifiers).
In terms of deranked DCs (see the two top rows), a distinction is made
between cases of equal flexibility (=), and cases in which the DC expresses
one function less than the PoS class (>(1)).
Deranked DCs
= > (1) total
PoS contentives / non-verbs Tagalog Turkish 4
Kharia
Kayardild
nominals / modifiers Quechua Gooinyandi 7
Ma’di Hungarian
Japanese
Koasati
Ket*
total 5 6 11
Table 8.2: The Principle of Increasing Categorization: Flexible PoS and deranked DCs
Note that there is one somewhat quirky case in Table 8.2, marked by an
asterix: Ket has lexical modifiers and deranked nominal clauses. This means
that, while both the lexical and the clausal construction can be used in two
functions, there is overlap in only one of these functions, namely modifier in
a referential phrase. Therefore, the DC is counted as expressing one function
less than the PoS class 121.
In short, in the large majority of languages with flexible PoS systems,
deranked DCs either display full retention or partial loss of the functional
possibilities available at the lexical level. This is in accordance with the
hypothesis in (1) / (2).
119
These cases stand in contrast to rigid deranked DCs that can express two or three
functions less than the relevant PoS. This is indicated in Table 8.1 as > (3) or > (2). These cases
will be discussed shortly.
120
As can be seen in Table 8.1 (and see also example (5)), Kharia has both a deranked DC that
displays the same flexibility as lexical contentives and a DC that can express one function less
than lexical contentives. In order to avoid double values, in Table 8.2 Kharia is mentioned as
possessing only the former, most flexible DC.

Flexible deranked DCs and Functional Transparency
Before turning to the remaining cases in Table 8.1 – those that involve
complete loss of flexibility in the DC (cf. (3c) above) – I will consider flexible
deranked DCs from the perspective of functional transparency, as discussed
in the previous section. Put simply, any flexible construction, including a
flexible deranked DC, presents a potential problem of functional ambiguity.
As in the case of flexible lexeme classes (cf. Hengeveld et al. 2004), it is
expected that languages will have certain morpho-syntactic means to
resolve this ambiguity. The data from my sample confirm this prediction. In
particular, the relevant languages employ the same phrase-structural devices
that are used for lexical or phrasal constituents to unambiguously mark the
functions of flexible deranked DCs. Consider for instance the examples in
(6a-b) from Tagalog. Example (6a) illustrates the use of the flexible gerund
construction in a functional slot marked by the locative marker sa. Example
(6b) shows the use of a simple lexical construction in the same functional
slot, marked by the same element:
Tagalog (Himmelmann 2005: 373; Himmelmann 2007: 252)

(6) a. pag-bawal-an mo ang bataɁ-ng iyó
sf-forbidden-lv 2sg.poss spec child-lk dist
sa [pag-la-laróɁ sa lansangan]
loc ger-rdp-play loc street
‘Forbid that child to play in the street.’
b. um-sakay silá sa bangká

av-passenger 3pl loc boat
‘They got on the boat.’
Other strategies used to resolve the functional ambiguity of flexible deranked

DCs include case marking and rigid word order. For instance in Quechua,
-shka clauses are case-marked when they function as the head of a referential
121
It is possible, however, that the bare infinitival form in what I have classified as a nominal
clause construction is in fact a lexical derivation when it appears in the function of modifier
in a referential phrase. This is hard to say, since in this function the patient argument of
the dependent predicate is gapped, and the agent is demoted, so that there are no overtly
expressed arguments (cf. example (16) in Chapter 6).

phrase, but not when they modify a referential head. Nominal clauses in
Ma’di follow the main predicate when functioning as object complements,
while they follow their nominal head when functioning as relative clauses.
I will return to the issue of functional transparency in section 8.4, where I
discuss how languages may disambiguate the multiple functions of flexible
balanced DCs.
Complete loss of flexibility in the DC
I now turn to the cases in Table 8.1 involving DCs that (i) can express a
subset of the functions of the relevant PoS class, and (ii) where this subset is
a singleton set, so that the DC is a rigid rather than a flexible construction.
This corresponds to the situation described in (3c) of section 8.3.1: complete
loss of flexibility in the deranked DC. As explained above, cases that involve
a PoS class that can express two functions and a DC that can express only
one are regarded as instances of partial loss of flexibility. Thus, the cases that
remain to be discussed here involve rigid deranked DCs that can express
two or three functions less than the relevant PoS. These cases are indicated
in Table 8.1 above as > (3) or > (2).
Two languages in Table 8.1 only have deranked DCs that display
complete loss of flexibility, i.e. loss of two or three functions compared to
the relevant flexible PoS classes in these languages. The first case is Samoan,
which has lexical contentives and rigid deranked clausal nominalizations;
the second is Warao, with lexical non-verbs and rigid deranked infinitival
clauses. The other languages with deranked DCs that show complete loss
of flexibility (Kharia, Turkish, and Kayardild) also have deranked DCs that
show no loss or only partial loss of flexibility.
Obviously, the categorial distinctiveness of any rigid deranked DC
construction is by definition higher than that of any flexible PoS class.
As such, these DCs all conform to the [flex PoS ≥ flex deranked DC]
constraint, and do not require further explanation. Nonetheless, there seem
to be additional motivations for the attestation of many rigid deranked DCs
in languages with flexible PoS. A first reason may be borrowing. This occurs
in Kharia, which has borrowed the rigid participial forms -na-wala, -al, and
-ker from Indo-Aryan. A second reason – more relevant from the present
perspective – may be the presence of an additional (possibly derived) rigid
PoS class in an otherwise flexible system, serving as the primary category
on which the rigid deranked DC is modelled. Warao, for instance, has

simple non-verbs plus a class of derived rigid nouns (see also Chapter 7,
section 7.2.2.3).
The rigid deranked complement clause construction in Samoan also
presents an interesting case. These nominalized clauses may occur with a
special marker on the dependent predicate, namely the suffix -ga. In Chapter
2 (section 2.5.2.2), I have shown that this suffix is also used for lexical
derivation. I explained there that lexically derived -ga formations remain
flexible (see example (39) in Chapter 2). In contrast, when -ga formation
applies at a higher level of grammatical complexity, i.e. when -ga is a marker
of syntactic derivation, then its output is apparently no longer flexible. This
is in accordance with the [flex PoS ≥ flex deranked DC] constraint. An
example of a syntactic -ga nominalization is given in (7). Note that this
construction can be distinguished from a regular NP (with a -ga-derived
head) by the fact that the second argument of the dependent predicate (apu
ma moli ‘apples and citrus’) remains expressed as in an independent clause.
Note also that the term ‘NP’ is not meant here to refer to a phrase with
a nominal head, to the extent that Samoan has no lexical nouns (neither
simple nor derived).

(7) A le faalavelave le tupu i [le ai-ga
pst neg trouble art king ld art eat-nmlz
apu ma moli a le pipili ma le tauaso]
apple and citrus poss art lame and art blind
‘The king was not troubled that the lame and the blind ate the
apples and oranges.’
Interestingly, Samoan rigid nominalized clauses can also occur without any
structural coding on the dependent predicate. This is illustrated in (8):
Samoan (Mosel 1992: 267)

(8) E lelei [l=a=u tunu ia]
genr good art=poss=2sg roast fish
‘Your fish roasting is good.’
Arguably, these unmarked rigid nominalizations are indeed modelled on a

rigid primary construction, but on a phrasal rather than a lexical one: an NP.
Samoan NPs are characterized by the possibility to take determiners / case-

markers, and to contain possessive modifiers. Since in Samoan any lexeme
can function as the head of an NP, the dependent predicate does not
need to be re-categorized when it comes to function as the nucleus of a
clausal nominalization. This explains why no special marker appears on the
dependent predicate.
Similar types of unmarked nominalized clauses are attested in other
Central-Eastern Malayo-Polynesian languages with a high degree of lexical
flexibility, such as Tuvaluan (see example (9)) and Kambera (see example
(10)). In the latter language nominalized clauses are cross-referenced with a
dative marker on the matrix predicate, just like lexical and phrasal objects.
Tuvaluan (Besnier 2000: 512, 513)

(9) Koe na e poto i [te fakkaa o
You there npast clever in art caus-burn
o te kasa], nee?
poss art pressure-lamp tag
‘You know how to light a pressure-lamp, don’t you?’

(10) Nda ku-pí-a-nya [na ngàndi-mu
neg 1sg.nom-know-mod-3sg.dat art take-2sg.gen
rú kuta]
leaf pepper plant
‘I dind’t know that you would bring kuta.’
(lit.: ‘I didn’t know (of ) your bringing kuta.’)
In short, there may be independent motivations, apart from the [flex PoS ≥
flex deranked DC] constraint, for the attestation of rigid deranked DCs in
functions that can be expressed by means of a flexible PoS class, including
borrowing and modelling on an additional rigid PoS class or a rigid
phrasal construction.
Rigid PoS and deranked DCs in languages with flexible PoS systems
Finally, consider rigid PoS classes in languages with flexible PoS systems,
and their clausal counterparts. It was hypothesized that any deranked DC
appearing in a function for which a rigid PoS class is available, should be a
rigid construction. In other words, this DC should not be able to express

any more function(s), since this would obviously violate the [flex PoS ≥
flex deranked DC] constraint. The data in Table 8.3 make clear that this
prediction is born out: For every rigid PoS class in a language with a flexible
PoS system it is true that, if there is a deranked DC available to express the
same function, then this DC is also a rigid construction.
Summary
In sum, in this section I have tried to reinterpret three types of results from
Chapter 7 in terms of the Principle of Increasing Categorization. One was
an expected result, in terms of the hypotheses in Chapter 7, namely that
flexibility in the domain of deranked DCs is dependent on flexibility in the
domain of PoS classes. I have shown that this dependency relation can be
interpreted as a constraint, imposed by the amount of flexibility attested in
the PoS system of a particular language, on the maximal amount of flexibility
(or minimal amount of categorial specificity) that can be displayed by the
deranked DC(s) of that language.
Furthermore, I accounted for two results that were unexpected in terms of
the predictions in Chapter 7. First, I explained the lack of correlations between
lexical nominals and modifiers on the one hand, and deranked DCs with the
same functional patterns on the other hand (see Tables 7.16-7.18 and Table 7.19
in Chapter 7). This finding can be understood as involving loss of flexibility in
the DCs as compared to the PoS classes, and thus as supporting the Principle
of Increasing Categorization. The same holds for the finding that virtually all
languages with very flexible PoS classes have at least one rigid deranked DC
(whether or not in combination with flexible deranked DC(s)).
Table 8.3: Rigid PoS classes in languages with flexible PoS systems, and the distribution
of deranked DCs that appear in the same function as these rigid PoS classes
Rigid PoS Form Function type Type Flex PoS ≥
Flex DC?
1 Warao D Noun -kitane complement 2 =
2 Kayardild Adj -Thirri-n relative 3 =
Adj -n-garrba relative 2 =

Rigid PoS Form Function type Type Flex PoS ≥
Flex DC?
3 Gooniyandi Madv -wadda adverbial manner 2 =
Madv -mawoo adverbial manner 2 =
Madv -bari adverbial manner 2 =
Madv -ya/ adverbial manner 2 =
-gowaaya
4 Hungarian (D) MAdv -vá/-vé adverbial manner 2 =
(D) MAdv -vén adverbial manner 2 =
5 Japanese Noun no/mono complement 3 =
etc
MAdv -te/-de/-ite adverbial manner 2 =
MAdv -i/-Ø adverbial manner 2 =
6 Lango Noun -(kk)ɔ̀ complement 3 =
7 Koasati Noun NMLZ complement 2 =
(various
forms)
Noun -:sáya complement 2 =
Noun -:yólli complement 2 =
Noun -:ka complement 2 =
Noun -:kítta complement 2 =
Noun -laho:li:sáya complement 2 =
S/D -n adverbial manner 2 =
MAdv
S/D -k adverbial manner 2 =
MAdv
S/D -t adverbial manner 2 =
MAdv
8.3.2.2 Distribution of deranked DCs in languages with rigid PoS systems
This section is concerned with deranked DCs in languages with rigid PoS
only. It was predicted that in such languages any deranked DC that appears in
a function for which a lexical class is available should be a rigid construction
too. The relevant data are presented in Table 8.4. These results make clear
that the prediction is confirmed, except for the two counterexamples that
were already discussed in Chapter 7: In Burushaski and Hdi we find flexible
deranked nominal clauses, while these languages both have rigid lexical

nouns, and Burushaski also has rigid lexical adjectives. These cases violate
the [flex PoS ≥ flex deranked DC] constraint, since the degree of flexibility
of the deranked DCs is higher (or their categorical distinctiveness lower)
than that of the corresponding PoS classes.
Table 8.4: Rigid PoS classes in languages with rigid PoS only, and the distribution of
deranked DCs that appear in the same function
rigid PoS Form Function Type Flex PoS ≥
Flex DC?
1 Itelmen Noun INF (various complement 2 =
forms)
2 Basque Noun -t(z)e, complement 2 =
Adj -tu/-du/-i/- Ø relative 2 =
3 Bambara Adj -le/-ne relative 2 =
Madv -tò adverbial 2 =
manner
4 Georgian Noun -a complement 3 =
Adj m- (-a-)(-el/-al)) relative 3 =
Adj -ul/-il/m- relative 3 =
-ar/-al
Adj sa- (-el/-al/r) relative 2 =
5 Abkhaz Noun -ra complement 3 =
6 Polish Noun -nie complement 2 =
Noun INF (various complement 2 =
forms)
Adj -c relative 2 =
Adj -any etc. relative 2 =
Adj (PST.PASS) PTC relative 2 =
(various forms)
S/D PRS.PL-c- adverbial 2 =
Madv manner
7 Burushaski Noun -(á)as nominal 2 >
Adj -im/-um/-am relative 2 =
S/D n-STEM-(a)n adverbial 2 =
Madv manner
8 Lavukaleve Noun -e/-i complement 2 =

rigid PoS Form Function Type Flex PoS ≥
Flex DC?
9 Alamblak Noun -nef complement 3 =
Noun -(kfë)t complement 3 =
Adj ɨnd relative 2 =
10 Dhaasanac Noun -ɲ/-an complement 3 =
11 Berbice Noun fu/fi/Ø complement 2 =
Dutch
12 Nama D MAdv -se/!’aa/tsii adverbial 2 =
(no INDIC) manner
13 Hdi Noun tá nominal 3 >
14 Tamil Noun -atu complement 3 =
Noun -(kk)a complement 2 =
15 Kisi Noun Ø complement 2 =
16 Garo Noun -a complement 3 =
Noun -a-ni complement 2 =
Noun -na complement 2 =
Noun -kan/-kan-a complement 2 =
S/D -e/-e-min/-e-r adverbial 2 =
Madv manner
17 Krongo Noun -(t) complement 3 =
18 Hixkaryana Noun nɨ/-thɨ-/-hɨto complement 3 =
+ rɨ
19 W. Noun -niq complement 3 =
Greenlandic
Recall that in Chapter 7 no correlations were found between specific rigid

PoS and rigid deranked DCs, i.e. between nouns and deranked complement
clauses (see Tables 7.22-7.24 in Chapter 7), between adjectives and deranked
relative clauses (see Tables 7.27-7.29), and between manner adverbs and
deranked adverbial manner clauses (see Table 7.32). This is because, instead
of deranked DC's, languages may employ balanced DC constructions, the
formal and functional properties of which are unrelated to PoS (cf. section
8.4 below). Nonetheless, as the data in Table 8.4 show, if a language does
employ a deranked DC, then this construction displays the same functional
specialization as the relevant rigid PoS class in that language.

8.3.3 Summary
The findings of this section can be summarized as follows:
(i) In languages with flexible PoS systems, there are no cases of
deranked DCs that are more flexible than the relevant flexible
PoS classes;
(ii) In languages with flexible PoS systems, deranked DCs occurring in
a function that can be expressed by a rigid PoS class are rigid too;
(iii) In languages with rigid PoS systems, there are only two cases of
flexible deranked DCs occurring in functions that are expressed by
rigid PoS classes.
Thus, except for the two counterexamples mentioned under (iii), the
functional patterns of all deranked DCs in the sample language are in
accordance with the [Flex PoS ≥ Flex deranked DC] constraint. In the next
section, I will consider the functional possibilities of balanced DCs.
8.4 Balanced constructions: functional transparency and

other factors
8.4.1 Introduction
The results presented in Chapter 7 make clear that the functional patterns
of balanced DCs, unlike those of deranked DCs, are not dependent on the
degree of flexibility displayed by the PoS classes of the same language. This
holds for global matches as well as for specific matches; and for languages
with flexible PoS systems as well as for languages with rigid PoS systems. In
Chapter 7 and in section 8.2 I have argued that this absence of a functional
connection ties in with the absence of any formal similarity between lexical
expressions and balanced DCs. In the present section, I attempt to further
interpret the results for balanced DCs, taking into account the Principle of
Functional Transparency as well as other functional factors.
8.4.2 Functional Transparency

From the point of view of functional transparency, rigid balanced DCs, both
in languages with and languages without lexical flexibility, do not present any
problem: they are functionally transparent by definition. Flexible balanced
clauses, in contrast, are functionally ambiguous by definition. In the remainder
of this sub-section I try to explain the distribution of flexible balanced DC

from the perspective of functional transparency; first in languages with
flexible PoS systems, and then in languages with rigid PoS systems.
8.4.2.1 Flexible balanced DCs in languages with flexible PoS systems

The data presented in Chapter 7 show that a number of languages with
flexible PoS systems display flexible balanced DCs (even though no
dependency relation was established between these two construction types;
cf. Tables 7.3, 7.11, 7.15, and 7.33 in Chapter 7). The relevant cases are listed in
Table 8.5:
Language Form Function Type Flexible PoS class

in same functions
1 Tagalog Ø multi-functional 1 contentives
2 Samoan Ø nominal 1 contentives
3 Turkish ki nominal 1 non-verbs
4 Kayardild -ntha- nominal 1 non-verbs
5 Paiwan a (Ø) nominal 1 predicatives
6 Gooniyandi Ø multi-functional 1 nominals
7 Ket Ø nominal 1 modifiers
Table 8.5: Flexible balanced DCs in languages with flexible PoS systems
Interestingly, in many of these cases, the flexible balanced DC is not marked

by means of a dedicated subordinating conjunction. Rather, it makes use
of the same function-indicating marking, at the level of phrase-structure,
as constituents with a different internal structure, such as simple lexemes,
phrases, or deranked DCs (see for instance the Tagalog examples in (6) of
section 8.3.2.1 above). This fits in with the idea that languages with (more or
less pervasive) lexical flexibility use a kind of ‘grid’ of morpho-syntactically
marked slots to compensate for the functional ambiguity of their lexical
material (cf. Hengeveld et al. 2004, Lehmann 2008). Apparently, these
functional slots may not only accommodate lexical and phrasal constituents,
but also complete clausal structures, which do not structurally deviate from
independent clauses.
We may again take Tagalog as an example: In this language a balanced
DC is marked with the same function word that would be used for any
other linguistic unit expressing the same function. In (11), for instance, the
‘specifier’ ang is used to mark a balanced object complement clause, while in
(12) ang marks a simple lexical object.

Tagalog (Palmer 2003: 268)
(11) Moo-ng una hindi ko ma-alam-an
rem-lg first neg 1sg:gen irr:nc-knowledge-loc
ang [akin-g ga-gaw-in]
spec 1sg-lg rdp-do-irr
‘At first I didn’t know what to do.’
(12) I-in-abót ng manggagamot sa sundalo ang itlóg

cv-real-reach gen doctor loc soldier spec egg
‘The physician handed the egg to the soldier.’
Similarly, example (13) shows a balanced clause in relative function that is

marked by the linking element na. The same marker is used for a simple
lexical attribute in (14):

(13) Sa mga lalaki na [maN-ibig nung
loc pl man lk av-love dist.gen:lk
kanyá-ng anák]
3sg.dat-lk child
‘(So he held a contest) between the men who courted his child.’
(14) ulól na unggó

foolish lk monkey
‘foolish monkey’
Samoan also has the possibility to insert balanced DCs into functional
slots without a dedicated subordinating conjunction. Example (15) shows a
complement clause that is zero-marked for absolutive (the subject Tigilau
is marked for ergative). This can be compared with (16), which shows that
simple lexical arguments in the same functions are marked in the same way.

(15) Na iloa e Tigilau [úa sau Sina]
pst know erg Tigilau pfv comes Sina
‘Tigilau knew that Sina had come.’

(16) Sā fasi le maile e le teine
pst hit art dog erg art girl
‘The girl hit the dog.’
Example (17) shows a balanced relative clause, which is marked by means of

rigid word order: it obligatorily appears to the right of the head. As is shown
in (18), the same order is used for lexical modifiers.

(17) Ua tu le alii lea [na ua e Popi]
pfv stand.up art man that pst bite erg Popi
‘That man who was bitten by Popi stood up.’
(18) le ta’avale mumu

art car red
‘the red car’
As a final example, consider balanced nominal clauses in Paiwan. They are

marked with the particle a, as is shown in (19a-b), where the construction
functions as a subject complement clause and as a relative clause, respectively.
In (20) it can be seen that the same element is used with lexical constituents
as a marker of the argument in focus (the agent in (20a)), and as an attributive
marker in NPs (see (20b)):
Paiwan (Egli 1990: 230, 178)

(19) a. na˙nguaq a [ma-ngetjez sun]
good foc pass-come you
‘It is good that you have come.’
b. qala a [na tem-ker tua vaua]

stranger lk pfv drink-ag obl wine
‘he stranger who has drunk wine’
Paiwan (Egli 1990: 180, 181)

(20) a. na qem-aung a alak
pfv cry-ag foc child
‘The child has cried.’

b. vavayan a burai
woman lk beautiful
‘beautiful woman’
There are two flexible balanced clauses in Table 8.5 that do not conform to
the general picture of using the same phrase-structural coding for lexical
and clausal constructions, but these both represent rather special cases. First,
the balanced nominal clause construction in Kayardild displays so-called
complete concord (see Dench 2006): All elements of the DC are separately
marked with the oblique complementizer case -ntha. Example (21) shows
this construction in the function of object complement:
Kayardild (Evans 1995: 516)

(21) Ngada kamburri-ja niwan-ji [walbu-ntha dathin-inja
1sg.nom say-act 3sg-mloc raft-c.obl that-c.obl
barji-nyarra-nth]
capsize-appr-c.obl
‘I told him that the raft would capsize.’
Second, the balanced nominal clause construction in Turkish, marked by

the general subordinating conjunction ki, is special to the extent that it is
borrowed from Persian.
In sum, the majority of flexible balanced DCs in languages with flexible
PoS systems conform to the Principle of Functional Transparency by making
use of the same function-indicating morpho-syntactic markers that are also
employed for the disambiguation of non-clausal constituents.
8.4.2.2 Flexible balanced DCs in languages with rigid PoS systems

The data presented in Chapter 7 make clear that flexible balanced clauses
also occur quite often in languages without any lexical flexibility. The
relevant cases are listed in Table 8.6 below (cf. Table 7.33 in Chapter 7) 122.
In fact, Thai and Basque both have small classes of flexible modifiers; this marginal
122
amount of lexical flexibility is ignored here.

Language Form Function Type Rigid PoS class(es) in
same functions
1 Thai thiî nominal 1 nouns, adj
2 Basque -en nominal 1 nouns, adj
3 Georgian rom nominal 1 nouns, adj
4 Burushaski ke/ki multi-functional 1 nouns, adj, S/D Madv
6 Pipil ka(h) nominal 1 nouns, adj
7 Dhaasanac DET (+ DEM) nominal 1 nouns, adj
8 Mandarin Ø/de multi-functional/ 1 nouns, S. Adj, S. Madv
modifier
9 Nung Ø multi-functional 1 nouns, S. Adj, X
8 Greenlandic PTC mood nominal 1 nouns, X
Table 8.6: Flexible balanced DCs in languages with rigid PoS systems
These DCs are again inherently functionally ambiguous. Therefore, on the

basis of the Principle of Functional Transparency, it is expected that they
combine with additional morpho-syntactic means that distinguish between
their multiple functions. This expectation is confirmed. Some of the flexible
balanced DCs in Table 8.6 employ disambiguating strategies that pertain
exclusively to the domain of subordination. Other DCs make use of more
general strategies that are also used with other types of constructions
(cf. Hengeveld & Van Lier 2008).
As a first example of the former, subordination-specific type of strategy,
consider Thai. This language has a balanced nominal clause construction
marked with thîi, as illustrated in (22a-b) below. When functioning as a
relative clause, as in (22b), this construction can be combined with a
resumptive pronoun. Even though this grammatical device does not count
as structural coding in the sense of a relativizer, it does mark the function
of the relativized item in the DC, and as such helps to show that we are
dealing with the modifying use of the thîi-constuction, rather than with its
referential use.
Thai (Iwasaki & Ingkaphirom 2005: 245, 255)

(22) a. tɛ̀-wâa dii ná [thîi mây mii khay pen alay]
but good prt comp neg have who cop what
‘But it was good that no one was hurt.’

b. Khon [thîi kháw pay yùu kan taam roŋrian]
people rel rsp3 go stay recip school
‘people who want to stay at school’
A second example is the Georgian DC construction marked with rom. This

is also a nominal clause construction: it can be used as a complement clause
and as a relative clause, as is illustrated in (23a-b) below. Example (23a)
shows that the rom-construction, when functioning as a complement clause,
can be combined with a pronominal element in the main clause. This is
even obligatory when the complement clause is dependent on a postposition
or functions obliquely (i.e. follows a verb that takes instrumental case).
Moreover, the examples in (23) show a contrast in terms of the position of
the conjunction: Very often (although not obligatorily) complement clauses
start with rom, while in relative clauses rom avoids the first position.
Georgian (Hewitt 1995: 613, 607)

(23) a. (is) uk’ve
(that(nom)) already
še-v-a-mčn-i-e rom es
prev-1sg-nv-notice-ths-aor.ind comp this
xalx-i sando ar ar-i-s
people-nom trustworthy not be-prs-it
‘I have already noticed that this people is not trustworthy.’
b. [gušin rom m-a-čuk-e,]

yesterday rel me-lv-present-aor.ind
is beč’ed-I sad ar-i-s?
that(nom) ring-nom where be-prs-it
‘Where is that ring which you presented to me yesterday?’
There are also languages that use more generally applicable strategies to
indicate the function of flexible balanced DCs. An obvious example of
such a strategy would be fixed word order 123. Consider for instance Pipil:
the complementizer ka(h) in this language is sometimes also used to
123
See Hengeveld et al. (2004), who show that disambiguating morpho-syntactic measures,
such as rigid word order, are not confined to languages with flexible PoS systems. While such
measures are always present in the case of lexical flexibility, they may be present in languages
with rigid PoS systems. cf. section 8.2 above.

introduce relative clauses (Campbell 1985: 128). The two functions of the
ka(h)-construction are illustrated in (24a-b). These examples show that
complement clauses are post-verbal, while relative clauses must follow their
nominal head. Since Pipil has rigid classes of verbs and nouns, these ordering
restrictions resolve the functional ambiguity of this DC construction.
Pipil (Campbell 1985: 126, 129)

(24) a. Tesu ki-mati [ka ne i-siwa:-w
no it-know comp the his-wife-poss
se: bru:hah]
a witch
‘He didn’t know that his wife is a witch.’
b. Ni-k-miktih ne mistun [ka ki-kwah ne tu:tut]

I-it-killed the cat rel it-ate the bird
‘I killed the cat that ate the bird.’
The -en construction in Basque can also be used as a complement clause and
as a relative clause. This is illustrated in (25a-b). The first example shows that
this DC combines with a determiner when used as a complement clause. In
contrast, when the construction is used as a relative clause, the determiner
appears on the modified head:

(25) a. Entzun dut [Amaiaren neba hil d-en]-a
hear aux Amaia.gen brother die aux-comp-art
b. [Pellok ekarri du-en] diru-a galdu dut

Peter.erg bring aux-rel money-det lose aux
‘I lost the money Peter brought.’
Finally, consider the somewhat exceptional case of the West Greenlandic

participial mood construction. This is again a construction that can be used
as a complement clause and as a relative clause, but, as already discussed
in Chapter 6 (see example (57a-b)), it behaves quite differently in each of
these two functions: Intransitive participial forms are fully balanced when
functioning as complements, whereas in relative clause function they are not

marked for person other than third. Also, it is not possible to create relative
clauses with corresponding transitive participial mood forms.
In sum, flexible balanced DCs in languages with rigid PoS systems
conform to the Principle of Functional Transparency; their multiple
functions can be disambiguated by means of several strategies. Some of
these strategies, such as the use of resumptive pronouns, are specific to
subordinate clauses, while others, such as word order constraints, are used
more generally in the language system as function-indicating devices.
8.4.3 Other factors: diachrony and cognition

Tables 8.5 and 8.6 above show that many flexible balanced DCs display a
particular pattern of flexibility: They are nominal clauses constructions, i.e.
they can be used as complement clauses and as relative clauses. There appear
to be two functional factors that motivate this pattern.
The first factor is a diachronic one. According to Cristofaro (1998),
relative clause markers may develop complementizer use along the following
pathway 124: The original construction involves a predicate meaning ‘see’,
‘think’, ‘know’ etcetera (i.e. a potentially complement-taking predicate)
with a (pro-)nominal object followed by a non-restrictive relative clause.
Cristofaro illustrates this type of construction with an example from Biblical
Hebrew (taken from Givón 1991):
Biblical Hebrew (Givón 1991: 289)

(26) Ɂal tirɁu-ni [she-Ɂani shaxoret]
neg see:ipfv:3sg:masc-me rel-I be.dark:sg:f
‘Don’t see me that I am dark-skinned.’
As Cristofaro (1998: 65) explains, in such constructions “the relative clause

[…] has high communicative value: it conveys a specification about the main clause
object, and this specification represents the communicative focus of the sentence”.
As a result of this pragmatic saliency, the relative clause is reanalyzed
as the object of the matrix clause, so that the relative pronoun becomes
a complementizer, and the (pro-)nominal head is dropped. Example (27)
shows that the original relative marker she in Biblical Hebrew is used as
a complementizer:
124
The diachronic scenario to be described is probably most relevant for balanced DCs
that are marked by means of subordinating conjunctions, as opposed to zero-marked
constructions.

Biblical Hebrew (Givón 1991: 297)
(27) yada’’ti [she-gam zeh hu ra’’yon ruah]
know:pfv:1sg comp-also this it notion of:wind
‘I knew that this too was total nonsense.’
Second, Croft (2001: 346-349) suggests that there is also a cognitive

connection between complementation and relativization, which seems to
be in line with Cristofaro’s point about pragmatic saliency. In particular,
Croft makes use of Langacker’s (1987) distinction between dependent and
autonomous concepts: A dependent concept elaborates on (i.e. makes more
specific) a certain aspect of an autonomous concept. According to Croft,
complement clauses and relative clauses are alike in that they are both
dependent concepts that elaborate on an aspect of the main clause event.
However, complement clauses and relative clauses don’t do this in exactly
the same way. Whereas complement clauses elaborate on the main clause
event by filling an argument position, relative clauses rather represent an
elaboration of a participant of the main clause event. Nonetheless, Croft
argues that the sharing of structural properties between complement clauses
and relative clauses in the same language can be explained in terms of their
shared cognitive function of main clause elaboration.
In short, the fact that many flexible balanced DCs involve nominal
clause constructions may be motivated in terms of diachronic and cognitive
factors. These motivations make reference to a similarity between the
functions of complementation and relativization constructions: both specify
a discursively important aspect of the matrix clause.
8.4.4 Summary
In this section I have tried to show that the distributional behaviour of
balanced DCs, even though it it is not related to the flexible or rigid functional
properties of PoS classes in particular languages, does fit into a larger
explanatory picture based on the Principle of Functional Transparency.
8.5 Summary
In this chapter I reconsidered the results obtained in Chapter 7 in light of two
closely related functional principles proposed in recent typologically-based
literature: the Principle of Increasing Categoriality (Haig 2006, Lehmann
2008), and the Principle of Functional Transparency (Frajzyngier & Shay
2003, Hengeveld et al. 2004, Sinnemäki 2008). In addition, a categorical

distinction was implemented between deranked and balanced dependent
clauses, based on Dik’s (1997) Principle of Formal Adjustment. In particular,
I hypothesized that deranked DCs, which are formally modelled on PoS, are
also functionally constrained by the latter. Balanced DCs, in contrast, have
no formal properties of PoS, and are not expected to display any functional
similarities with lexical constructions either.
The hypothesized functional connection between PoS and deranked
DCs was operationalized in accordance with the Principle of Increasing
Categoriality, which states that the degree of categorial specificity of
linguistic constructions increases with their structural complexity. In terms
of the present study, this means that (secondary) deranked DCs are predicted
to either display the same amount of flexibility as (primary) PoS classes, or
less flexibility, but not more. This prediction was confirmed for virtually all
deranked DC constructions attested in the sample.
Moreover, it was shown that flexible deranked DCs conform to the more
general Principle of Functional Transparency, which claims that languages
avoid ambiguity in the formal expression of a particular functional domain.
Interestingly, most flexible deranked DCs in languages with flexible PoS
systems do this by means of the same morpho-syntactic devices that
are used in these languages to disambiguate the functions of lexical and
phrasal constructions.
Regarding balanced DCs, it was shown that, even though their functional
possibilities cannot be correlated with those of PoS classes in the same
language, their distributional properties can be explained in terms of the
Principle of Functional Transparency. Whereas rigid balanced DCs are by
definition functionally transparent, flexible balanced DCs, like other flexible
constructions, require additional morpho-syntactic means to compensate
for their inherent functional ambiguity. I have illustrated several types of
strategies that serve this function for balanced DC constructions. Some such
strategies are confined to subordination constructions, while others are more
general applicable. It was shown that balanced flexible clauses in languages
with flexible PoS systems, like deranked ones, employ the same function-
indicating devices that are used to compensate for functional ambiguity at
the lexical / phrasal level.

In sum, the functional principles of Increasing Categoriality, Functional
Transparency, and Formal Adjustment together provide a complete account
of the results presented in Chapter 7. Moreover, the discussion in the present
chapter confirms the general picture emerging from Chapter 7, namely that
functional flexibility at the lexical level has pervasive repercussions at higher
grammatical levels.

Summary and Conclusion
The aim of this study was to investigate the functional relationship between
a language’s parts of speech classes and its dependent clauses, in terms of
the set of propositional functions that both construction types can express.
First, in Chapters 2 and 3, I discuss various functionalist approaches to the
theory and typology of parts of speech and dependent clauses, respectively.
It is shown that both parts of speech and dependent clauses can be defined
as formal mappings onto a space consisting of four propositional functions.
A basic distinction is adopted between rigid and flexible constructions:
The former type can express only a single propositional function, while
the latter can be used in two or more functions, without any difference in
structural coding.
Whereas both lexical and clausal constructions can thus be defined in
terms of their functional possibilities, only the latter are configurational, and
can as such also be classified according to their internal morpho-syntactic
properties. On the basis of earlier functional-typological studies, a distinction
is made between balanced and deranked dependent clauses. Balanced clauses,
which represent one extreme of a scale, are characterised by the fact that they
express the same set of categories as independent clauses. Towards the other
extreme we find various types of deranked clauses, the internal structure
of which is increasingly deviant from the structure of independent clauses.
This deviation may show in two ways: First, it may take the form of de-
categorization, i.e. (partial) loss of ‘verbal’ features, such as TAM distinctions
349
and person marking, and / or non-expression of arguments. Second, deranked
dependent clauses may show formal reflections of re-categorization, i.e. they
may acquire (some) ‘nominal’ features, such as the expression of case markers
and/or determiners and possessive coding of argument(s).
Against this theoretical backdrop, Chapter 4 formulates a set of predictions
concerning the expected similarity between the functional possibilities as
displayed by the parts of speech classes of a particular language and by its
dependent clause constructions. Some predictions are formulated in general
terms, i.e. they make reference to languages with some flexibility versus no
flexibility in their parts of speech system, and the expected reflection of this
difference on the functional properties of dependent clause constructions
in these languages. Other predictions are more specific; they aim at the
identification of one-to-one matches between the functional possibilities of
particular types of part of speech classes and dependent clause constructions.
In addition, all predictions are formulated first without differentiation for
dependent clauses in terms of their internal morpho-syntactic properties,
and then in sets of sub-predictions that make specific reference to balanced
clauses versus (various types of ) deranked clauses.
Chapters 5 and 6 present the classifications of parts of speech classes
and dependent clauses in a balanced sample of 50 languages in terms of
the typological frameworks developed in Chapters 2 and 3. Subsequently, in
Chapter 7, the two data sets of Chapter 5 and 6 are linked in order to identify
dependency relations between the functional patterns displayed by the parts
of speech classes of particular languages, and their (different structural types
of ) dependent clause constructions. The analyses reveal that the presence
of flexible deranked dependent clauses in a language is dependent upon the
presence of flexible parts of speech classes in that language. This does not
mean, however, that all languages with flexible parts of speech systems also
display flexible deranked dependent clauses.
Furthermore, it is shown that pervasively flexible parts of speech classes
and deranked dependent clauses are both, i.e. independently of each other,
cross-linguistically rare phenomena. Therefore, the generalization that
maximally flexible deranked dependent clauses do not occur in languages
without the same degree of flexibility in the lexical domain does not have
much explanatory power. More interestingly, it is shown that whenever
maximal lexical flexibility does occur in a language, this strongly increases
the chances of also finding the other rare phenomenon: maximally flexible
deranked dependent clauses.

In contrast to the result obtained for pervasively flexible constructions, no
dependency relations can be established between the presence of less flexible
or rigid parts of speech classes and deranked dependent clauses with the
same functional behaviour. Moreover, the functional patterns of balanced
dependent clauses are not related to those of parts of speech classes either.
In Chapter 8 an attempt is made to provide a functionalist explanatory
framework to account for these results. In particular, I suggest that the
findings concerning parts of speech and deranked dependent clauses can be
interpreted as conforming to the Principle of Increasing Categoriality (or
Staggering Level-dependent Categoriality), which states that the categorial
specificity of linguistic units increases – or their flexibility decreases – with
increased structural complexity. Deranked dependent clauses, as opposed
to balanced ones, are regarded as secondary constructions, derived from
primary lexical or phrasal constructions, in terms of the Principle of Formal
Adjustment. Deranked dependent clauses are more complex than their
lexical counterparts, and therefore the Principle of Increasing Categoriality
predicts that the former should be at least as categorially specific as the latter.
In other words, the degree of categorial specificity attested at the lexical level
is expected to determine the minimal degree of categorial specificity (or
the maximal degree of flexibility) that should be maintained at the level of
deranked dependent clauses. This principle explains the finding that flexible
deranked dependent clauses are almost exclusively found in languages with
(pervasively) flexible parts of speech systems.
Moreover, the Principle of Increasing Categoriality accounts for a
number of other results of Chapter 7: First, it captures the finding that the
functional matches between flexible parts of speech and flexible deranked
dependent clauses need not be one-to-one, but may also involve a partial
decrease in the flexibility of the latter construction type, as compared with
the former. Second, the principle accounts for the finding that languages
with pervasively flexible parts of speech systems almost always have rigid
deranked dependent clauses (possible alongside their flexible deranked
clauses). Third, the principle explains at least partly why a lower degree of
flexibility at the lexical level does not correlate with flexibility at the level of
dependent clauses: If a flexible part of speech class covers only two functions,
then even a loss in flexibility of one function on the part of the dependent
clause automatically results in a rigid construction.
Finally, the Principle of Increasing Categoriality and the Principle of
Formal Adjustment shed light on the fact that no correlations are found
Chapter 9 – Summary and Conclusion | 351

between particular types of rigid parts of speech classes and rigid deranked
dependent clauses with the same distribution. The reason for this is that,
instead of rigid deranked dependent clauses, languages may employ balanced
constructions. The latter, since they are presumably not modelled on lexical
categories, may be rigid as well as flexible. However, whenever a language
has a deranked dependent clause construction to express a function for
which a rigid parts of speech class is available, then this dependent clause
construction is rigid as well. This holds both for languages with flexible parts
of speech systems that include one or more rigid parts of speech classes, as
well as for languages with rigid parts of speech classes only. These findings
confirm the prediction that a deranked dependent clause cannot exhibit a
greater degree of flexibility than a parts of speech class appearing in the
same function.
Regarding balanced dependent clauses, I argue that even though their
functional possibilities cannot be linked to those of parts of speech, their
distributional behaviour is in accordance with the more general Principle
of Functional Transparency. This principle predicts that every linguistic
unit should ultimately, i.e. at the utterance-level, be endowed with a
categorial value, in order to make its function identifiable for the hearer.
As expected, the multiple possible functions of flexible balanced clauses,
both in languages with rigid and with flexible parts of speech systems, can
be disambiguated through additional morpho-syntactic strategies. Some of
these strategies, such as the use of resumptive pronouns, are confined to the
functional domain of subordination, while others, such as fixed constituent
order, are used much more generally to establish functional transparency in
grammatical (sub-)systems.
With regard to this last point, a particularly interesting finding is that
the flexible dependent clauses – both balanced and deranked ones – in
languages with very flexible parts of speech systems often make use of the
same morpho-syntactic strategies that are used to indicate the functions of
lexical and phrasal constituents. This supports a general pattern emerging
from the present study concerning languages with pervasive lexical flexibility:
They appear to have a particular type of phrase structure, consisting of a
‘grid’ of morpho-syntactically marked slots, which may be filled by formal
units of any type of internal structure: not just single lexemes, but also
complex phrasal and clausal constituents. These complex units can (but need
not) retain maximal flexibility until the final level of the structure-building
process, when they are inserted into a syntactic frame corresponding to a

particular propositional function. This pattern is in accordance with findings
from other studies, which suggest that, in the encoding of a particular
functional domain, flexibility in one grammatical area must be compensated
with rigidity in another area.
Both the Principle of Increasing Categoriality and the Principle of
Functional Transparency are ultimately motivated in terms of the functional
principle of Economy. In particular, languages either retain the functional
specialization present in simple units in the process of building more
complex ones, or gradually increase it, until the ultimate level of functional
specialization required at the utterance level is reached. If, in contrast,
the functional specialization of less complex units would be decreased or
altogether lost along the way towards more complexity, this would require
additional and as such uneconomical processing on the part of the speaker.
Moreover, if the loss of categorial specificity would not be repaired at some later
stage in the derivational process, this would imply an additional processing
load for the hearer, who would ultimately have to solve the problem of
functional ambiguity that he or she would be presented with. Therefore, the
Economy principle predicts that complexity-increasing linguistic processes
will produce output structures that are at least as categorially specific as their
input structures, in order to maximize processing ease for both participants
of a communicative situation.
On a final note, I would like to point out that redundancy seems to be a
common phenomenon in languages, while ambiguity is not. In other words,
languages often encode a particular functional distinction within a certain
domain by means of more than one formal device, while they apparently
tend to prevent situations in which there is too little grammatical coding to
unambiguously identify the function of a linguistic unit within an utterance.
Viewed from the perspective human communication, this suggests that
languages strongly adhere to processing ease for the hearer, although not
necessarily for the speaker.
Chapter 9 – Summary and Conclusion | 353

References 125
Aarts, Bas 2006. Conceptions of categorization in the history of

linguistics. Language Sciences 28, 361-385.
* Alpatov, Vladimir M. & Vera I. Podlesskaya 1995. Converbs in
Japanese. In: Martin Haspelmath & Ekkehard König (Eds.): Converbs in
cross-linguistic perspective. Berlin: Mouton de Gruyter. pp. 465-485.
* Anderson, Gregory D.S. 2002. Case-marked clausal subordination in
the Burushaski complex sentence. Studies in Language 26:3, 543-571.
Anward, Jan, Edith Moravcsik and Leon Stassen 1997. Parts of
speech: A challenge for typology. Linguistic Typology 1, 167-183.
Arad, Maya 2003. Locality constraints on the interpretation of roots:
the case of Hebrew denominal verbs. Natural Language and Linguistic
Theory 21, 737-778.
Aronoff, Mark 1979. Word formation in generative morphology.
Cambridge, Mass.: MIT Press.
* Asher, Ron E. 1982. Tamil [Croom Helm Descriptive Grammars].
London / New York: Croom Helm.
Baker, Mark 2003. Lexical categories: verbs, nouns, and adjectives.
Cambridge: Cambridge University Press.
125
References with an asterix are grammars or other data sources used for the sample
languages. Therefore they appear also in the list of ‘References for the sample languages’
below.
355
Bakker, Dik, forthcoming. Language sampling. In: Jae Jung Song
(Ed.): Handbook of Linguistic Typology. Oxford: Oxford University Press.
Bauer, Winifred 1993. Maori. London: Routledge.
Beck, David 2002. The typology of parts of speech systems:
The markedness of adjectives. New York: Routledge.
* Berger, Hermann 1998a. Die Burushaski Sprache von Hunza und
Nager. Teil I: Grammatik. Wiesbaden: Harrowitz Verlag.
* Berger, Hermann 1998b. Die Burushaski Sprache von Hunza und
Nager. Teil III: Wörterbuch. Wiesbaden: Harrowitz Verlag.
* Berry, Keith & Christine Berry 1999. A description of Abun, a West
Papuan language of Irian Jaya [Pacific Linguistics, Series B,
No. 115]. Canberra: Pacific Linguistics, Research School for Pacific and
Asian Studies, Australian National University.
Bhat, N.S. 1994. The adjectival category. Amsterdam: Benjamins.
* Bielec, Dana 1998. Polish: An essential Grammar. London / New York:
Routledge.
Bisang, Walter, forthcoming. Word Classes. In: Jae Jung Song (Ed.):
The Oxford handbook of Language Typology. Oxford: Oxford University Press.
* Blackings, Mairi & Nigel Fabb 2003. A Grammar of Ma’di
[Mouton Grammar Library 32]. Berlin: Mouton de Gruyter.
Bloomfield, Leonard 1962 (1933). Language. London: Allen & Unwin.
Broschart, Jürgen 1997. Why Tongan does it differently: Categorial
distinctions in a language without nouns and verbs. Linguistic Typology 1,
123-165.
* Bruce, Les 1984. The Alamblak language of Papua New Guinea
(East Sepik) [Pacific Linguistics, Series C, No. 81]. Canberra: Australian
National University.
* Burling, Robbins 2004. The language of the Modhupur Mandi (Garo).
Vol 1: Grammar. New Delhi / Morganville, New Jersey: Bibliophile South
Asia / Promilla & Co. Publishers.
Bybee, Joan 1985. Morphology. A study of the relation between meaning
and form. Amsterdam: Benjamins.
* Campbell, Lyle 1985. The Pipil Language of El Salvador [Mouton
Grammar Library 1]. Berlin / New York / Amsterdam: Mouton Publishers.
* Cerrón-Palomino, Rodolfo 1987. Lingüística Quechua. Cuzco:
Centro de Estudios Rurales Andinos ‘Bartolomé de las Casas’.
* Childs, Tucker G. 1995. A grammar of Kisi: A Southern Atlantic
language [Mouton Grammar Library 16]. Berlin: Mouton de Gruyter.

* Chirikba, Viacheslav, A. 2003. Abkhaz [Languages of the
World / Materials 119]. München: Lincom Europa.
Clark, E.V. & H. H. Clark 1979. When nouns surface as verbs.
Language 55, 767-811.
Cohen, Jacob 1995. A power primer. Psychological Bulletin 112:1, 155-159.
* Cole, Peter 1982. Imbabura Quechua [Lingua Descriptive Studies 5].
Amsterdam: North Holland Publishing Company.
Comrie, Bernard 1976. The syntax of action nominals:
a cross-linguistic study. Lingua 40, 177-201.
Comrie, Bernard & Tania Kuteva 2005. Relativization strategies.
In: Martin Haspelmath et al. (Eds.): The world atlas of language structures.
Oxford: Oxford University Press. pp. 494-497.
* Conrad, Robert J. & Kepas Wogiga 1991. An outline of Bukiyip
Grammar [Pacific Linguistics, Series C, No. 113]. Canberra:
Australian National University.
Cristofaro, Sonia 1998. Grammaticalization and clause linkage
strategies: A typological approach with particular reference to Ancient
Greek. In: Paul Hopper and Anna Giacalone Ramat (Eds.): The limits of
grammaticalization. Amsterdam / Philadelphia: Benjamins. pp. 55-88.
Cristofaro, Sonia 2003. Subordination. Oxford: Oxford University Press.
Croft, William 1991. Syntactic categories and grammatical relations.
Chicago: University of Chicago Press.
Croft, William 2000. Parts of speech as language universals and as
language-particular categories. In: Petra Vogel & Bernard Comrie (Eds.):
Approaches to the typology of word classes. Berlin: Mouton de Gruyter.
pp. 65-102.
Croft, William 2001. Radical Construction Grammar: Syntactic theory in
typological perspective. Oxford: Oxford University Press.
Croft, William 2003. Typology and Universals. Second Edition.
Croft, William 2005. Word classes, parts of speech, and syntactic
argumentation. Linguistic Typology 9:3, 431-441.
Cysouw, Michael 2003. Against implicational universals.
Linguistic Typology 7:1, 89-101.
Cysouw, Michael 2005. Quantitative methods in typology.
In: Gabriel Altmann, Reinhard Köhler & R. Piotrowski (Eds.):
Quantitative Linguistics: An International Handbook. Berlin: Mouton de
Gruyter. pp. 554-578.
References | 357
Cysouw, Michael 2007. Building semantic maps: the case of person
marking. In: Matti Miestamo & Berhard Wälchli (Eds.):
New challenges in typology. Berlin: Mouton de Gruyter. pp. 225-248.
Dench, Alan 2006. Case marking strategies in subordinate strategies
in Pilbara Languages – Some diachronic speculations.
Australian Journal of Linguistics 26:1, 81-105.
* Derbyshire, Desmond C. 1979. Hixkaryana
[Lingua Descriptive Studies 1]. Amsterdam: North Holland Publishing.
Dik, Simon C. 1989. The theory of Functional Grammar. Part I:
The structure of the clause. Dordrecht: Foris.
Dik, Simon C. 1997. The theory of Functional Grammar. Part 2:
Complex and derived constructions. Ed. Kees Hengeveld. Berlin / New York:
Mouton de Gruyter.
Dixon, R.M.W. & Alexandra Y. Aikhenvald 2006. Complementation.
A cross-linguistic typology. Oxford: Oxford University Press.
Don, Jan & Eva van Lier, forthcoming. Categorization and derivation
in flexible and differentiated languages. In: Jan Rijkhoff & Eva van Lier
(Eds.): Flexible word classes: a typological study of underspecified
parts-of-speech. Oxford: Oxford University Press.
Dryer, Matthew 2003. Significant and non-significant implicational
universals. Linguistic Typology 7:1, 108-127.
* Egli, Hans 1990. Paiwangrammatik. Wiesbaden: Otto Harrassowitz.
* Evans, Nicholas 1995. A grammar of Kayardild
Evans, Nicholas 2000: Word classes in the world’s languages.
In: Geert Booij, Christian Lehmann & Joachim Mugdan (Eds.):
Morphology. A handbook on inflection and word formation. Volume 1. Berlin:
Walter de Gruyter. pp. 708-732.
Evans, Nicholas & Toshiki Osada 2005. Mundari: The myth of a
language without word classes. Linguistic Typology 9:3, 351-390.
Everitt, B.S. 1977. The analysis of contingency tables.
London / New York: Chapman & Hall.
Farrell, Patrick 2001. Functional shift as category underspecification.
English Language and Linguistics 5:1, 109-130.
Fischer, Rafael & Eva van Lier, forthcoming. Cofán subordinate
clauses in a typology of subordination. In: Katharina Haude,
Rik van Gijn, & Pieter Muysken (Eds.): Subordination in South American
Indian Languages. Amsterdam: Benjamins.

* Foley, William 1998. ‘Symmetrical Voice Systems and
Precategoriality in Philippine Languages.’ Paper presented at the
workshop Voice and Grammatical Functions in Austronesian of the
LFG98 conference, Brisbane. Downloadable at: http:/ / www.sultry.arts.
usyd.edu.au / LFG98 / austro / foley / fintro.htm
* Fortescue, Michael 1984. West Greenlandic
[Croom Helm Descriptive Grammars]. London: Croom Helm
* Frajzyngier, Zygmunt with Erin Shay 2002. A grammar of Hdi
Frajzyngier, Zygmunt & Erin Shay 2003. Explaining language
structure through system interaction [Typological studies in language 55].
Amsterdam / Philadelphia: Benjamins.
* Georg, Stefan & Alexander P. Volodin 1999. Die Itelmenische Sprache:
Grammatik und Texte [Tunguso Sibirica 5]. Wiesbaden: Harrassowitz.
Gil, David 2000. Syntactic categories, cross-linguistic variation and
universal grammar. In: Petra Vogel & Bernard Comrie (Eds.): Approaches to
the typology of word classes. Berlin: Mouton de Gruyter. pp. 173-216.
Givón, Talmy 1980. The binding hierarchy and the typology of
complements. Studies in Language 4, 453-473.
Givón, Talmy 1991. The evolution of dependent clause
morpho-syntax in Biblical Hebrew. In: Elisabeth Closs Traugott & Bernd
Heine (Eds.): Approaches to grammaticalization. Amsterdam / Philadelphia:
Benjamins. pp. 257-310.
Givón, Talmy 2001. Syntax. Volume 1. Amsterdam / Philadelphia:
Benjamins.
* Göksel, Aslɩ & Celia Kerslake 2005. Turkish.
A comprehensive grammar. London / New York: Routledge.
Gordon, Raymond G. Jr. (Ed.) 2005. Ethnologue:
Languages of the world. 15th Edition. Dalles, Texas: SIL International.
Greenberg, Joseph H. 1966. Some universals of language, with particular
reference to the order of meaningful elements. In: Joseph H. Greenberg
(Ed.): Universals of Language. Cambridge: MIT Press. pp. 73-113.
* Gregores, Emma & Jorge Suárez 1967.
A description of colloquial Guaraní. The Hague: Mouton.
* Hagman, Roy S. 1979. Nama Hottentot Grammar. Ann Arbor,
Michigan: University Microfilms.
References | 359
Haig, Geoffrey 2006. ‘Word-class distinctions and morphological type:
agglutinating and fusional languages reconsidered.’ Unpublished manuscript,
University of Kiel.
Haiman, John 1983. Iconic and economic motivation.
Language 59, 781-819.
* Harriehausen, Bettina 1990. Hmong Njua: syntaktische Analyse
einer gesprochenen Sprache mithilfe datenverarbeitungstechnischer Mittel
und sprachvergleichende Beschreibung des südostasiatischen Sprachraumes
[Linguistische Arbeiten 245]. Tübingen: Niemeyer.
Haspelmath, Martin 1994. Passive participles across languages.
In: Barbara Fox & Paul J. Hopper (Eds.): Voice: Form and Function
[Typological Studies in Language 27]. Amsterdam: Benjamins. pp. 151-177.
Haspelmath, Martin 1995. The converb as a cross-linguistically valid
category. In: Martin Haspelmath & Ekkehard König (Eds.): Converbs in
cross-linguistic perspective. Berlin / New York: Mouton de Gruyter. pp. 1-55.
Haspelmath, Martin 2001. Word classes and parts of speech.
In: Paul B. Baltes & Neil J. Smelser (Eds.): International Encyclopedia of the
Social and Behavioral Sciences. Amsterdam: Pergamon. pp. 6538-16545.
Haspelmath, Martin 2002. Understanding Morphology. New York:
Oxford University Press.
Haspelmath, Martin 2007. Pre-established categories don't exist:
consequences for language description and typology.
Linguistic Typology 11:1, 119-132
Haspelmath, Martin 2008. Frequency vs. iconicity in explaining
grammatical asymmetries. Cognitive Linguistics 19:1, 1-33.
Haspelmath, Martin, Matthew S. Dryer, David Gil & Bernard
Comrie (Eds.) 2005. The world atlas of language structures. Oxford:
Hawkins, John A. 1983. Word order universals. New York:
Academic Press.
* Heath, Jeffrey 1984. Functional grammar of Nunggubuyu.
Canberra: Australian Institute of Aboriginal Studies.
Hengeveld, Kees 1992. Non-verbal predication. Theory, typology,
diachrony. Berlin: Mouton de Gruyter.
Hengeveld, Kees 1998. Adverbial clauses in the languages of Europe.
In: Johan van der Auwera (Ed.): Adverbial constructions in the languages of
Europe. Berlin / New York: Mouton de Gruyter. pp. 335-419.

Hengeveld, Kees, Jan Rijkhoff and Anna Siewierska 2004.
Parts-of-speech systems and word order. Journal of Linguistics 40, 527-570.
Hengeveld, Kees & Jan Rijkhoff 2005. Mundari as a flexible language.
Linguistic Typology 9:3, 406-431
Hengeveld, Kees & J. Lachlan Mackenzie 2008.
Functional Discourse Grammar. Oxford: Oxford University Press.
Hengeveld, Kees & Eva van Lier 2008. Parts of speech and dependent
clauses in Functional Discourse Grammar. In: Umberto Ansaldo, Jan Don
and Roland Pfau (Eds.): Parts of Speech: descriptive tools, theoretical
constructs. Special Issue of Studies in Language 32:3, 753-785.
Hengeveld Kees & Eva van Lier 2009. An implicational functional
map of parts of speech. In: Andrej Malchukov, Michael Cysouw &
Martin Haspelmath (Eds.): Semantic Maps: Methods and Applications.
Special Issue of Linguistic Discovery.
* Hewitt, Brian G. 1979. Abkhaz [Lingua Descriptive Studies 2].
* Hewitt, Brian G. 1987. The typology of subordination in Georgian and
Abkhaz. Berlin: Mouton de Gruyter.
* Hewitt, Brian G. 1995. Georgian: a structural reference grammar
[London Oriental and African Language Library 2].
Amsterdam: Benjamins.
* Himmelmann, Nikolaus P. 2005. Tagalog. In: Alexander Adelaar &
Nikolaus P. Himmelmann (Eds.): The Austronesian Languages of Asia and
Madagascar. [Routledge Language Family Series]. New York:
Routledge. pp. 350-376.
* Himmelmann, Nikolaus P. 2007. Lexical categories and voice in
Tagalog. In: Peter Austin & Simon Musgrave (Eds.): Voice and grammatical
relations in Austronesian Languages. Stanford: CSLI Publications.
pp. 246-293.
Himmelmann, Nikolaus P. & Eva Schulze-Berndt 2005. Issues in the
syntax and semantics of participant-oriented adjuncts: an introduction.
In: Nikolaus P. Himmelmann & Eva Schulze-Berndt (Eds.): Secondary
predication and adverbial modification. The typology of depictives. Oxford:
Oxford University Press. pp. 1-67.
* Hinds, John 1986. Japanese [Croom Helm Descriptive Grammars 16].
London: Croom Helm.
Hockett, Charles F. 1958. A course in modern linguistics. New York:
The Macmillan Company.
References | 361
Hockett, Charles F. 1966. The problem of universals in language.
In: Joseph H. Greenberg (Ed.): Universals of language. Cambridge,
Mass.: MIT Press. pp. 1-29.
Hopper, Paul J. & Sandra A. Thompson 1984. The discourse basis for
lexical categories in Universal Grammar. Language 60:4, 703-752.
Hopper, Paul J. & Sandra A. Thompson 1985. The iconicity of the
universal categories ‘noun’ and ‘verb’. In: John Haiman (Ed.):
Iconicity in syntax. Amsterdam: Benjamins. pp. 151-183.
* Hualde, José Ignacio & Jon Ortiz de Urbina 2003. A grammar of
Basque [Mouton Grammar Library 26]. Berlin: Mouton de Gruyter.
* Iwasaki, Shoichi & Preeya Ingkaphirom 2005. A reference grammar of
Thai. Cambridge: Cambridge University Press.
* Kenesei, István, Robert M. Vago & Anna Fenyvesi 1998. Hungarian.
London / New York: Routledge.
* Klamer, Marian 1998. A grammar of Kambera
König, Ekkehard 1995. The meaning of converb constructions.
In: Martin Haspelmath & Ekkehard König (Eds.): Converbs in cross-
linguistic perspective. Berlin / New York: Mouton de Gruyter. pp. 57-95.
Koptjevskaja-Tamm, Maria 1993. Nominalizations.
Koptjevskaja-Tamm, Maria 2003. Action Nominal Constructions in
the languages of Europe. In: Franz Plank (Ed.): Noun Phrase Structure in
the Languages of Europe. Berlin: Mouton de Gruyter. pp. 723-759.
Koptjevskaja-Tamm, Maria 2005. Action nominal constructions.
In: Martin Haspelmath et al. (Eds.): The world atlas of language structures.
Oxford: Oxford University Press. pp. 253-255.
* Kornfilt, Jaklin 1997. Turkish. London / New York: Routledge.
Kortmann, Bernd 1998. Adverbial subordinators in the languages
of Europe. In: Johan van der Auwera (Ed.): Adverbial constructions in the
languages of Europe. Berlin / New York: Mouton de Gruyter.
* Kouwenberg, Silvia 1994. A grammar of Berbice Dutch Creole
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar,
Volume I: Theoretical prerequisites. Stanford: Stanford University Press.
Langacker, Ronald W. 2002. Concept, image, and symbol. The cognitive
basis of grammar. Second edition. Berlin / New York: Mouton de Gruyter.

Lehmann, Christian 1988. Towards a typology of clause linkage.
In: John Haiman & Sandra A. Thompson (Eds.): Clause combining in
grammar and discourse. Amsterdam / Philadelphia: Benjamins. pp. 181-225.
Lehmann, Christian 1990. Towards lexical typology. In William Croft
et al. (Eds.): Studies in typology and diachrony. Papers presented to Joseph H.
Greenberg on his 75th birthday. Amsterdam / Philadelphia: Benjamins.
pp. 161-185.
* Lehmann, Christian 2005. Wortarten und Grammatikalisierung.
In: Clemens Knobloch & Burkhard Schaeder (Eds.): Wortarten und
Grammatikalisierung. Perspektiven in System und Erwerb. Berlin / New York:
Mouton de Gruyter. pp. 2-20.
Lehmann, Christian 2008. Roots, stems and word classes.
In: Umberto Ansaldo, Jan Don and Roland Pfau (Eds.): Parts of Speech:
descriptive tools, theoretical constructs.
Special Issue of Studies in Language 32:3, 546-567.
Lehmann, Christian & Edith Moravcsik 2000. Noun.
In: Geert Booij, Christian Lehmann & Joachim Mugdan (Eds.):
Morphology: An international handbook on inflection and word-formation.
Volume1. Berlin: Mouton de Gruyter. pp. 732-757.
* Li, Charles N. & Sandra A. Thompson 1981. Mandarin Chinese:
A functional reference grammar. Berkeley: University of California Press.
* Lombardi Vallauri, E. 1997. Clause linkage in Japanese.
Linguistics 35, 481-512
* Lorimer, D.L.R. 1935. The Burushaski Language. Volume I:
Introduction and grammar. Oslo: Instituttet for sammenlignende
kulturforskning.
Lyons, John 1977. Semantics. Cambridge: Cambridge University Press.
Mackenzie, J. Lachlan 1996. English nominalizations in the layered
model of the sentence. In: B. Devriendt et al. (Eds.): Complex structures: a
functionalist perspective. Berlin: Mouton de Gruyter. pp. 325-357.
Malchukov, Andrej 2004. Nominalization /Verbalization: constraining
a typology of transcategorial operations. München: Lincom Europa.
Malchukov, Andrej 2006. Constraining nominalization:
function / form competition. Linguistics 44:5, 973-1009.
Maslova, Elena 2003. A case for implicational universals.
References | 363
Maslova, Elena & Tatiana Nikitina, submitted. ‘Stochastic
universals and dynamics of cross-linguistic distributions: the case
of alignment types.’ Available on-line at: http:/ / www.stanford.
edu / ~emaslova / Publications / ProbabilityPubl.html
* Matissen, Johanna & Werner Drossard 1998. Lexical and
syntactic categories in Nivkh [Theorie des Lexicons: Arbeiten des
Sonderforschungsbereich 282: 85]. Düsseldorf: Heinrich Heine Universität.
* McGregor, William 1990. A functional grammar of Gooniyandi
[Studies in language companion series 22]. Amsterdam / Philadelphia:
Benjamins.
* Mithun-Williams, Marianne 1976. A grammar of Tuscarora.
New York / London: Garland Publishing.
* Mithun, Marianne 2000. Noun and verb in Iroquoian languages:
multicategorization from multiple criteria. In: Petra M. Vogel & Bernard
Comrie (Eds.): Approaches to the typology of word classes. Berlin:
* Moravcsik, Edith 2001. On the nouniness of Hungarian adjectives.
In: Chris Schaner-Wolles et al. (Eds). Naturally! Linguistic studies in honour
of Wolfgang Ulrich Dressler presented on the occasion of his 60th birthday.
Torino: Rosenberg and Sellier. pp. 337-346.
* Mosel, Ulrike 1992. On nominalisation in Samoan.
In: Tom Dutton, Malcolm Ross & Darrell Tryon (Eds.): The language
game. Papers in memory of Donald C. Laycock. Canberra: Research School of
Pacific Linguistics. pp. 263-281.
* Mosel, Ulrike 2004. Juxtapositional constructions in Samoan.
In: Isabelle Bril & Françoise Ozanne-Rivierre (Eds.): Complex predicates in
Oceanic Languages. Berlin: Mouton de Gruyter. pp. 261-294.
* Mosel, Ulrike & Even Hovdhaugen 1992. Samoan Reference
Grammar. Oslo: Scandinavian University Press.
* Neukom, Lukas 2001. Santali [Languages of the World / Materials
323]. München: Lincom Europa.
Newmeyer, Frederick 2007. Linguistic typology requires crosslinguistic
formal categories. Linguistic Typology 11:1, 133-157.
Nikitina, Tatiana 2007. Nominalization /Verbalization: constraining
a typology of transcategorial operations (Review of Malchukov 2004).

Noonan, Michael 1985 (2007). Complementation.
In: Timothy Shopen (Ed.): Language typology and syntactic description.
Volume 11: Complex constructions (Second edition). Cambridge:
Cambridge University Press. pp. 52-150.
* Noonan, Michael 1992. Lango [Mouton Grammar Library 7].
Berlin: Mouton de Gruyter.
* Nordhoff, Sebastian 2004. Nomen /Verb-Distiktion in Guaraní.
Arbeitspapier 48 (Neue Folge). Köln: Allgemeine Sprachwissenschat,
Institut für Linguistik, Universität zu Köln.
* Palmer, Gary 2003. Talking about thinking in Tagalog.
Cognitive Linguistics 14:2 / 3, 251-280.
* Paul, Waltraud 2005. Adjectival modification in Mandarin Chinese
and related issues. Linguistics 43:4, 757-793.
* Peterson, John 2005. There’s a grain of truth in every “myth”, or,
Why the discussion of lexical classes in Mundari isn’t quite over yet.
* Peterson, John 2006. Kharia. A South Munda language. Volume I:
Grammatical analysis. Habilitationsschrift, Universität Osnabrück.
* Peterson, John, forthcoming. Parts of speech in Kharia – A formal
account. In: Jan Rijkhoff & Eva van Lier (Eds.): Flexible word classes: a
typological study of underspecified parts-of-speech. Oxford: Oxford
University Press.
Pustet, Regina 2000. How arbitrary is lexical categorization?
Verbs vs. adjectives. Linguistic Typology 4:2, 175-212.
Pustet, Regina 2003. Copulas. Universals in the categorization of the
lexicon. Oxford: Oxford University Press.
* Rau, Felix, forthcoming. Proper names, predicates, and the part of
speech system of Santali. In: Jan Rijkhoff & Eva van Lier (Eds.):
Flexible word classes: a typological study of underspecified parts-of-speech.
Oxford: Oxford University Press.
* Reh, Mechtild 1985. Die Krongo Sprache (nìino módì). Beschreibung,
Texte, Wörterverzeichnis [Kölner Beiträge zur Afrikanistik 12].
Berlin: Reimer.
* Rice, Keren D. 1989. A grammar of Slave
Rijkhoff, Jan 2002. The Noun Phrase. Oxford: Oxford University Press.
Rijkhoff, Jan 2007. Word classes.
Language and Linguistics Compass, 1:6, 709-726
References | 365
Rijkhoff, Jan 2008a. Descriptive and discourse-referential modifiers in
a layered model of the noun phrase. Linguistics 46: 4, 789-829.
Rijkhoff, Jan 2008b. Layers, levels and contexts in Functional
Discourse Grammar. In: Daniel García Velasco & Jan Rijkhoff (Eds.):
The Noun Phrase in Functional Discourse Grammar. Berlin / New York:
Mouton de Gruyter. pp 63-115.
Rijkhoff, Jan, Dik Bakker, Kees Hengeveld & Peter Kahrel 1993.
A method of language sampling. Studies in Language 17, 169-203.
Rijkhoff, Jan & Dik Bakker 1998. Language sampling.
Linguistic Typology 2, 263-314.
* Romero-Figeroa, Andrés 1997. A reference grammar of Warao
[Lincom studies in native American linguistics 6]. München:
Lincom Europa.
* Rounds, Carol 2001. Hungarian. An essential grammar.
Ruhlen, Merritt 1991. A guide to the world’s languages, volume 1:
Classification (Second Edition). Stanford: Stanford University Press.
* Sadock, Jerrold M. 2003: A grammar of Kalaallisut (West
Greenlandic Inuttut) [Language of the world / Materials 162].
München: Lincom Europa.
Sasse, Hans-Jürgen 1993a. Syntactic categories and subcategories.
In: J. Jacobs et al. (Eds.): Syntax. Berlin: Mouton de Gruyter. pp. 646-686.
Sasse, Hans-Jürgen 1993b. Das Nomen – eine universale Kategorie?
Sprachtypology und Universaliënforschung (STUF) 46, 187-221.
Schachter, Paul 1985. Parts of speech systems. In: Timothy Shopen
(Ed.): Language typological and syntactic description. Volume 1: Clause
structure (First edition). Cambridge: Cambridge University Press.
* Schachter, Paul & Fe T. Otanes 1972. Tagalog reference grammar.
Berkeley: University of California Press.
Schachter, Paul & Timothy Shopen 2007. Parts-of-speech systems. In:
Timothy Shopen (Ed.): Language typology and syntactic description. Volume I:
Clause structure (Second Edition). Cambridge: Cambridge University Press.
* Schaub, Willi 1985. Babungo [Croom Helm Descriptive Grammars].
London: Croom Helm.
* Schroeder, Christoph 2004. Depiktive im Sprachvergleich Deutsch-
Türkish. Eine konstrastive-typologische Analyse. Habilitationsschrift,
Universität Osnabrück.

Searle, John R. 1969. Speech acts. An essay in the philosophy of language.
Smit, Niels 2001. Lexical categories in a diachronic perspective.
M.A. Thesis, University of Amsterdam.
Sinnemäki, Kaius 2008. Complexity trade-offs in core argument
marking. In: Matti Miestamo, Kaius Sinnemäki & Fred Karlssson (Eds.):
Language complexity. Typology, contact, change. Amsterdam / Philadelphia:
Benjamins. pp. 67-88.
Stassen, Leon 1985. Comparison and Universal Grammar.
Oxford: Basil Blackwell.
Stassen, Leon 1997. Intransitive predication. Oxford:
Thompson, Sandra A., Robert E. Longacre & Shin Ja J. Hwang
2007. Adverbial Clauses. In: Timothy Shopen (Ed.): Language typology
and syntactic description. Volume II: Complex constructions (Second edition).
Cambridge: Cambridge University Press. pp. 237-269.
* Tikkanen, Bertil 1995. Burushaski converbs in their South and
Central Asian context. In: Martin Haspelmath & Ekkehard König (Eds.):
Converbs in Cross-linguistic perspective. Berlin: Mouton de Gruyter.
pp. 487-528.
* Tosco, Mauro 2001. The Dhaasanac language. Cologne:
Rüdiger Köppe Verlag
* Uehara, Satoshi 1998. Syntactic categories in Japanese:
A cognitive and typological introduction. Tokyo: Kurosio Publishers.
* Vajda, Edward J. 2004. Ket [Languages of the World / Materials 204].
Van Lier, Eva 2006. Parts-of-speech and dependent clauses:
A typological study. Folia Linguistica 40:3 / 4, 239-304.
* Velázquez-Castillo, Maura 2002. Grammatical relations in active
systems. The case of Guaraní. Functions of Language 9:2, 133-167.
References | 367
References for the Sample
Languages
Abkhaz
Chirikba, Viacheslav, A. 2003. Abkhaz [Languages of the
World / Materials 119]. München: Lincom Europa.
Hewitt, Brian G. 1979. Abkhaz [Linga Descriptive Studies 2].
Hewitt, Brian G. 1987. The typology of subordination in Georgian and
Abkhaz. Berlin: Mouton de Gruyter.
Abun
Berry, Keith & Christine Berry 1999. A description of Abun: a West
Papuan language of Irian Jaya [Pacific Linguistics B 115]. Canberra: Pacific
Linguistics, Research School of Pacific and Asian Studies, The Australian
National University.
Alamblak
Bruce, Les 1984. The Alamblak language of Papua New Guinea
(East Sepik) [Pacific Linguistics C113]. Canberra: The Australian National
University.
Babungo
Schaub, Willi 1985. Babungo [Croom Helm Descriptive Grammars].
London: Croom Helm.
369
Bambara
Brauner, Siegmund 1974. Lehrbuch des Bambara. Leipzig:
Enzyklopädie.
Basque
Hualde, José Ignacio & Jon Ortiz de Urbina (Eds.) 2003:
A grammar of Basque [Mouton Grammar Library 26]. Berlin / New York:
Mouton de Gruyter.
Saltarelli, Mario 1988. Basque [Croom Helm Descriptive Grammars].
London: Croom Helm.
Berbice Dutch Creole

Kouwenberg, Silvia 1994. A grammar of Berbice Dutch Creole
[Mouton Grammar Library 12]. Berlin / New York: Mouton de Gruyter.
Bukiyip
Conrad, Robert J. with Kepas Wogiga 1991. An outline of Bukiyip
grammar. [Pacific Linguistics series C113] Canberra: Research School of
Pacific Linguistics.
Burushaski
Anderson, Gregory D.S. 2002. Case-marked clausal subordination in
the Burushaski complex sentence. Studies in Language 26:3, 543-571.
Berger, Hermann 1998a. Die Burushaski Sprache von Hunza und Nager.
Teil I: Grammatik. Wiesbaden: Harrowitz Verlag.
Berger, Hermann 1998b. Die Burushaski Sprache von Hunza und Nager.
Teil III: Wörterbuch. Wiesbaden: Harrowitz Verlag.
Lorimer, D.L.R. 1935. The Burushaski Language. Volume I: Introduction
and grammar. Oslo: Instituttet for sammenlignende kulturforskning.
Tikkanen, Bertil 1995. Burushaski converbs in their South and
Central Asian context. In: Martin Haspelmath & Ekkehard König (Eds.):
Converbs in Cross-linguistic perspective. Berlin: Mouton de Gruyter.
pp. 487-528.
Dhaasanac
Tosco, Mauro 2001. The Dhaasanac language. Cologne: Rüdiger Köppe
Verlag.

Garo
Burling, Robbins 1961. A Garo grammar [Deccan College Monograph
Series 25]. Poona: Deccan College.
Burling, Robins 2004. The language of the Modhupur Mandi (Garo) Vol. 1:
Grammar. New Delhi/Morganville: Bibliophile South Asia/Promilla & Co.
Georgian
Cherchi, Marcello 1999. Georgian [Languages of the world / Materials
147]. München: Lincom Europa.
Hewitt, Brian George 1987. The typology of subordination in Georgian
and Abkhaz. Berlin: Mouton de Gruyter.
Hewitt, Brian George 1995. Georgian: a structural reference grammar
[London Oriental and Africa Language Library 2]. Amsterdam:
Benjamins.
Vamling, Karina 1989. Complementation in Georgian. Ph.D. Thesis.
Lund: Lund University Press.
Gooniyandi
McGregor, William 1990. A functional grammar of Gooniyandi [Studies
in language companion series 22]. Amsterdam / Philadelphia: Benjamins.
Guaraní
Gregores, Emma & Jorge Suárez 1967. A description of colloquial
Guaraní. The Hague: Mouton.
Nordhoff, Sebastian 2004. Nomen /Verb-Distiktion in Guaraní.
Arbeitspapier 48 (Neue Folge). Köln: Allgemeine Sprachwissenschat,
Institut für Linguistik, Universität zu Köln.
Velázquez-Castillo, Maura 2002. Grammatical relations in active
systems. The case of Guaraní. Functions of Language 9:2, 133-167.
Hdi
Frajzyngier, Zygmunt with Erin Shay 2002. A grammar of Hdi.
Hixkaryana
Derbyshire, Desmond C. 1979. Hixkaryana [Lingua Descriptive
Studies 1]. Amsterdam: North Holland Publishing.
References for the Sample Languages | 371

Hmong Njua
Harriehausen, Bettina 1990. Hmong Njua: syntaktische Analyse
einer gesprochenen Sprache mithilfe datenverarbeitungstechnischer Mittel
und sprachvergleichende Beschreibung des südostasiatischen Sprachraumes
[Linguistische Arbeiten 245]. Tübingen: Niemeyer.
Hungarian
Kenesei, István, Robert M. Vago & Anna Fenyvesi 1998. Hungarian.
Moravcsik, Edith A. 2001. On the nouniness of Hungarian adjectives.
In: Chris Schaner-Wolles, John Rennison, and Friedrich Neubarth (Eds.):
Naturally! Linguistic studies in honor of Wolfgang Ulrich Dressler presented on
the occasion of his 60th birthday. Vienna: Rosenberg & Seiler. pp. 337-346.
Rounds, Carol 2001. Hungarian. An essential grammar. London /
New York: Routledge.
Imbabura Quechua
Cole, Peter 1982. Imbabura Quechua [Lingua Descriptive Studies 5].
Itelmen
Georg, Stefan & Alexander P. Volodin 1999. Die Itelmenische Sprache:
Grammatik und Texte [Tunguso Sibirica 5]. Wiesbaden: Harrassowitz.
Japanese
Alpatov, Vladimir M. & Vera I. Podlesskaya 1995. Converbs in
Japanese. In: Martin Haspelmath & Ekkehard König (Eds.): Converbs in
cross-linguistic perspective. Berlin: Mouton de Gruyter. pp. 465-485.
Hinds, John 1986. Japanese [Croom Helm Descriptive Grammars].
London: Croom Helm.
Lombardi Vallauri, E. 1997. Clause linkage in Japanese. Linguistics 35,
481-512.
Kambera
Klamer, Marian 1998. A grammar of Kambera [Mouton Grammar
Library 18]. Berlin: Mouton de Gruyter.

Kayardild
Evans, Nicholas D. 1988. Odd topic marking in Kayardild. In: Peter
Austin (Ed.): Complex constructions in Australian languages
[Typological studies in language 15]. Amsterdam: Benjamins. pp. 219-266.
Evans, Nicholas D. 1995. A grammar of Kayardild
[Mouton Grammar Library 15]. Berlin: Mouton de Gruyter
Ket
Vajda, Edward J. 2004. Ket [Languages of the World / Materials 204].
Werner, Heinrich 1995. Zur Typologie der Jenissej-Sprachen.
[Veröffentlichungen der societas uralo-altaica. Band 45]. Wiesbaden:
Harrowitz Verlag.
Werner, Heinrich 1997. Die Ketische Sprache. Wiesbaden:
Harrowitz Verlag.
Kharia
Peterson, John 2005. There’s a grain of truth in every “myth”, or, Why
the discussion of lexical classes in Mundari isn’t quite over yet. Linguistic
Typology 9:3, 391-441.
Peterson, John 2006. Kharia. A South Munda language. Volume I:
Grammatical analysis. Habilitationsschrift, Universität Osnabrück
Peterson, John, forthcoming. Parts of speech in Kharia – A formal
account. In: Jan Rijkhoff & Eva van Lier (Eds.): Flexible word classes: a
typological study of underspecified parts-of-speech. Oxford: Oxford
University Press.
Kisi
Childs, Tucker G. 1995. A grammar of Kisi
Koasati
Kimball, Geoffrey D. 1991. Koasati grammar. Lincoln: University of
Nebraska Press.

Krongo
Reh, Mechtild 1985. Die Krongo Sprache (nìino módì). Beschreibung,
Texte, Wörterverzeichnis [Kölner Beiträge zur Afrikanistik 12].
Berlin: Reimer.
Lango
Noonan, Michael 1992. Lango [Mouton Grammar Library 7].
Berlin: Mouton de Gruyter.
Lavukaleve
Terrill, Angela 2003. A grammar of Lavukaleve
Ma'di
Blackings, Mairi & Nigel Fabb 2003. A grammar of Ma’di
Mandarin Chinese
Li, Charles N. & Sandra A. Thompson 1981. Mandarin Chinese.
Lin, Hua 2001. A grammar of Mandarin Chinese
[Languages of the World / Materials 344]. München: Lincom Europa.
Paul, Waltraud 2005. Adjectival modification in Mandarin Chinese
and related issues. Linguistics 43:3, 757-793.
Nama
Hagman, Roy Stephen 1974. Nama hottentot grammar. Ann Arbor,
Michigan: University Microfilms.
Nunggubuyu
Heath, Jeffrey 1984. Functional grammar of Nunggubuyu. Canberra:
Australian Institute of Aboriginal Studies.

Nivkh
Gruzdeva, Ekaterina 1998. Nivkh [Languages of the World:
Material III]. München: Lincom Europa.
Matissen, Johanna & Werner Drossard 1998. Lexical and
syntactic categories in Nivkh [Theorie des Lexicons: Arbeiten des
Sonderforschungsbereich 282, 85]. Düsseldorf: Heinrich Heine Universität.
Nung
Saul, Janice E. & Nancy Freiberger Wilson 1980. Nung grammar.
Dallas: Summer Institute of Linguistics.
Paiwan
Egli, Hans 1990. Paiwangrammatik. Wiesbaden: Harrassowitz
Pipil
Campbell, Lyle 1985. The Pipil Language of El Salvador [Mouton
Grammar Library 10]. Berlin / New York / Amsterdam: Mouton Publishers.
Polish
Bielec, Dana 1998. Polish: An essential Grammar. London / New York:
Routledge.
Samoan
Mosel, Ulrike & Even Hovdhaugen 1992. Samoan reference grammar.
Oslo: Scandinavian University Press.
Mosel, Ulrike 1992. On nominalisation in Samoan. In: Tom Dutton,
Malcolm Ross & Darrell Tryon (Eds.): The language game. Papers in memory
of Donald C. Laycock. Canberra: Research School of Pacific Linguistics.
Mosel, Ulrike 2004. Juxtapositional constructions in Samoan. In:
Isabelle Bril & Françoise Ozanne-Rivierre (Eds.): Complex predicates in
Oceanic Languages. Berlin: Mouton de Gruyter. pp. 261-294.
Santali
Neukom, Lukas 2001. Santali. München: Lincom Europa.
Rau, Felix, forthcoming. Proper names, predicates, and the part of
speech system of Santali. In: Jan Rijkhoff & Eva van Lier (Eds.): Flexible
word classes: a typological study of underspecified parts-of-speech. Oxford:

Slave
Rice, Keren D. 1989. A grammar of Slave
Tagalog
Foley, William 1998. “Symmetrical Voice Systems and Precategoriality
in Philippine Languages.” Paper presented at the workshop Voice and
Grammatical Functions in Austronesian of the LFG98 conference,
Brisbane. Downloadable from: http:/ / www.sultry.arts.usyd.edu.
au / LFG98 / austro / foley / fintro.htm
Himmelmann, Nikolaus P. 2005. Tagalog. In: Alexander Adelaar
& Nikolaus P. Himmelmann (Eds.): The Austronesian Languages of
Asia and Madagascar [Routledge Language Family Series]. New York:
Routledge. pp. 350-376.
Himmelmann, Nikolaus P. 2007. Lexical categories and voice
in Tagalog. In: Peter Austin & Simon Musgrave (Eds.): Voice and
grammatical relations in Austronesian Languages. Stanford:
CSLI Publications. pp. 246-293.
Schachter, Paul & Fe T. Otanes 1972. Tagalog reference grammar.
Tamil
Asher, Ron E. 1982. Tamil [Lingua Descriptive Studies 7].
Amsterdam: North Holland.
Lehmann, Christian 2005: Wortarten und Grammatikalisierung.
In: Clemens Knobloch & Burkhard Schaeder (Eds.): Wortarten und
Grammatikalisierung. Perspektiven in System und Erwerb. Berlin:
Thai
Iwasaki, Shoichi & Preeya Ingkaphirom 2005.
A reference grammar of Thai. Cambridge: Cambridge University Press
Smyth, David 2002. Thai. An essential grammar. London / New York:
Routledge

Turkish
Göksel, Asli & Celia Kerlake 2005. Turkish: A comprehensive grammar.
Oxford: Routledge.
Kornfilt, Jaklin 1997: Turkish. London / New York: Routledge.
Tuscarora
Mithun-Williams, Marianne 1976. A grammar of Tuscarora. New York:
Garland Publishing.
Mithun, Marianne 2000. Noun and verb in Iroquoian languages:
Multicategorisation from multiple criteria. In: Petra M. Vogel & Bernard
Comrie (Eds.): Approaches to the typology of word classes. Berlin / New York:
Wambon
De Vries, Lourens 1986. The Wambon relator system.
Working Papers in Functional Grammar 17.
De Vries, Lourens & Robinia de Vries-Wiersma 1992. The morphology
of Wambon [Verhandelingen van het Koninklijk Instituut voor Taal-,
Land- en Volkenkunde 151]. Leiden: KITLV Press.
Warao
Romero-Figeroa, Andres 1997. A reference grammar of Warao
[Lincom studies in Native American Linguistics 6]. München: Lincom.
West Greenlandic
Fortescue, Michael. 1984. West Greenlandic [Croom Helm Descriptive
Grammars]. London: Croom Helm.
Sadock, Jerrold M. 2003. A grammar of Kalaallisut (West Greenlandic
Inuttut) [Language of the world / Materials 162]. München:
Lincom Europa.

Appendix i: Language Sample
Language Family (Ethnologue) Subfamily (Etnologue) Genus (WALS)

1 Hdi Afro-Asiatic Chadic Biu-Mandara
2 Dhaasanac Afro-Asiatic Cushitic Eastern-Cushitic
3 Turkish Altaic Altaic Turkic
4 Japanese Japanese Japanese Japanese
5 Pipil Uto-Aztecan Uto-Aztecan Aztecan
6 Hixkaryana Carib Carib Cariban
7 Tuscarora Iroquoian Iroquoian Northern Iroquoian
8 Koasati Moskogean Moskogean Muskogean
9 Guaraní Tupi Tupi Tupi-Guaraní
10 Warao Isolate Isolate Warao
11 Imbabura Quechuan Quechuan Quechuan
Quechua
12 Gooniyandi Australian Bunaban Bunaban
13 Kayardild Australian Tangkic Tangkic
14 Nunggubuyu Australian Gunwinyungan Nunggubuyu
15 Tagalog Austronesian Western Malayo- Meso-Philippine
Polynesian
16 Kambera Austronesian Central Malayo- Central Malayo-Polynesian
Polynesian
17 Samoan Austronesian Eastern Malayo- Oceanic
Polynesian
18 Paiwan Austronesian Paiwanic Paiwanic
19 Garo Sino-Tibetan Tibeto-Burman Baric
20 Thai Tai-Kadai Kam-Tai Kam-Tai
21 Santali Austro-Asiatic Munda Munda
22 Kharia Austro-Asiatic Munda Munda
23 Hmong-Njua Hmong-Mien Hmong-Mien Hmong-Mien
24 Abkhaz North-Causasian North-Caucasian North-west Caucasian
25 Georgian Kartvelian Kartvelian Kartvelian
26 Itelmen Chukotko-Kamchatkan Chukotko- Southern Chukotko-
Kamchatkan Kamchatkan
27 Tamil Dravidian Southern Dravidan Southern Dravidian
28 West Greenlandic Eskimo-Aleut Eskimo-Aleut Eskimo-Aleut
29 Polish Indo-European Slavic Slavic
30 Wambon Trans-New Guinea Awju-Dumut Awju-Dumut
31 Alamblak Sepik-Ramu Sepik Hill Sepik Hill
32 Lavukaleve East Papuan Central Solomons Solomon East Papuan
33 Abun West Papuan North Central Bird’s North Central Bird’s Head
Head
34 Bukiyip Torricelli Arapesh Kombio-Arapesh
35 Nama Khoisan Central Koisan Central Koisan
379
Language Family (Ethnologue) Subfamily (Etnologue) Genus (WALS)
36 Slave Na-Dene Athapaskan Athapaskan
37 Ket Yeniseian Yeniseian Yeniseian
38 Burushaski Isolate Isolate Burushaski
39 Basque Basque Basque Basque
40 Nivkh (Gilyak) Isolate Isolate Nivkh
41 Babungo Niger-Congo Benue Congo Bantoid
42 Kisi Niger-Congo Atlantic Southern-Atlantic
43 Bambara Niger-Congo Mande Western-Mande
44 Krongo Nilo-Saharan Kadugli Kadugli
45 Lango Nilo-Saharan Eastern-Sudanic Nilotic
46 Ma’di Nilo-Saharan Central Sudanic Moru-Ma’di
47 Berbice Dutch Creole Creole Creoles and Pidgins
Creole
48 Nung Tai-Kadai Kam-Tai Kam-Tai
49 Mandarin Chinese Sino-Tibetan Chinese Chinese
50 Hungarian Uralic Finno-Ugric Ugric
Geographical location of sample languages

Appendix ii: Excluded parts of speech systems
The following parts of speech systems are predicted not to occur in any language, because they violate one
or more of the constraints developed in Chapter 2 as part of the implicational map model of parts of speech.
These constraints are repeated below for convenience. Note that the constraints are numbered in the same way
as in Chapter 2, in order to facilitate comparison with the predicted systems discussed there. The PoS systems
are simply numbered (1)* through (34)*. The asterixes are added to indicate that these are ‘forbidden’ systems.

a. If a language has a rigid class of lexemes that can be used as the head of a referential phrase,
it must also have a rigid class of lexemes that can be used as the head of a predicate phrase.
b. If a language has a flexible class of lexemes that can be used as the head of a referential
phrase (but not as the head of a predicate phrase, since then the restriction becomes
irrelevant), it must also have a flexible or rigid class of lexemes that can be used as the head
of a predicate phrase.
(12) Head Modifier

a. If a language has a rigid class of lexemes that can be used as the modifier within a phrase, it
must also have a rigid class of lexemes that can be used as the head of that phrase.
b. If a language has a flexible class of lexemes that can be used as the modifier within a phrase
(but not as the head of that phrase, since then the restriction becomes irrelevant), it must
also have a flexible or rigid class of lexemes that can be used as the head of that phrase.
(13) ((Predication / Reference) (Head / Modifier))

If a language has distinct (rigid or flexible) classes of lexemes for heads and modifiers within any
phrase, then it must also have distinct (rigid or flexible) classes of lexemes for heads of predicate
versus referential phrases.
The system in (1)* has a flexible class of lexemes (non-verbs) that can be used as the head of a referential
phrase, but no class of lexemes that can be used as the head of a predicate phrase (verbs). Thus, it violates
constraint (11b) above. Furthermore, it has a flexible class of lexemes that can be used as a modifier in a
predicate phrase, but no class of lexemes that can be used as the head of a predicate phrase. Thus, it also
violates constraint (12b).
(1)*
Head Modifier
Predication –
Reference Non-verb
The system in (2)* is excluded on the basis of constraint (12b): It has a flexible class of lexemes that can be
used as the modifier (but not the head) in a referential phrase, and no corresponding class of heads.
(2)*
Head Modifier
Predication Flex
Reference –
The system in (3)* is excluded because it has a rigid head class in the referential, but not in the predication
domain. As such, it violates constraint (11a).
(3)*
Head Modifier
Predication Flex
Reference Noun
The system in (4)* involves a rigid modifier class in the referential domain without the corresponding rigid
head class and thus violates (12a). In addition, it has distinct classes for heads and modifier in the referential
domain, without distinct classes of heads of predicative versus referential phrases. Thus, it also violates (13).
381
(4)*
Head Modifier
Predication Flex
Reference Adjective
The system in (5)* is excluded because it violates constraint (12a) in the predication domain, where is has
a rigid class for modifiers, but no rigid class for heads. This system also violates constraint (13), since it
has distinct classes for heads and modifiers in the referential domain, but no distinct classes for heads of
predicate versus referential phrases.
(5)*
Head Modifier
Predication Flex Manner adverb
Reference
The system in (6)* is excluded because it has a class of flexible lexemes that can be used as the head of a
referential phrase, but no class of lexemes that can be used as the head of a predicate phrase. Thus, it violates
constraint (11b).
(6)*
Head Modifier
Predication – –
Reference Nominal
The system in (7)* violates two constraints: (11b) because it has a flexible class of lexemes that can be used as
the head of a referential phrase, but no class of lexemes that can be used as the head of a predicate phrase. And
(12a) because it has a rigid class of modifiers in the predication domain, but not the corresponding heads.
(7)*
Head Modifier
Predication – Manner adverb
Reference Nominal
The system in (8)* violates constraint (12b): It has a flexible class of modifiers, but no heads.
(8)*
Head Modifier
Predication – Modifier
Reference –
The system in (9)* is excluded because it violates constraint (13): There is a lexical distinction between heads
and modifiers in the predication domain, but there are no distinct classes for heads of predicative versus
referential phrases. In addition, it violates constraint (12b) in the referential phrase, where modifiers can be
expressed lexically, but not heads.
(9)*
Head Modifier
Reference –
The system in (10)* is also excluded because it violates constraint (13) and constraint (12b) (but this time in
the predication domain). In addition, constraint (11a) is violated: this system has a rigid class of heads for
referential but not for predicate phrases.

(10)*
Head Modifier
Predication – Modifier
Reference Noun
The system in (11)* violates constraint (13): There is lexical distinction between heads and modifiers, but there
are no distinct classes for heads of predicate versus referential phrases.
(11)*
Head Modifier
Predication Head Modifier
Reference
The system in (12)* is excluded because it has two rigid classes of modifiers, without the corresponding rigid
classes of heads. As such it violates constraint (12a) in both the predication and the reference domain.
(12)*
Head Modifier
Predication Head Manner adverb
Reference Adjective
The system in (13)* is excluded because it violates constraint (12a) in the predication domain: It has a class of rigid
lexemes that can be used as the modifier in a predicate phrase, but no rigid class for heads of predicate phrases.
(13)*
Head Modifier
Predication Manner adverb
Head
Reference –
The system in (14)* also violates constraint (12a), but in the referential domain, where it has a rigid class of
modifiers, but no rigid class for heads.
(14)*
Head Modifier
Predication –
Head
Reference Adjective
The system in (15)* violates constraint (11a): It has a rigid class of heads in the referential but not in the
predication domain.
(15)*
Head Modifier
For the same reason, the system in (16)* is excluded:
(16)*
Head Modifier
Reference Noun –
The system in (17)* violates constraint (12a): It has a class of rigid modifiers in the reference domain, but not
the corresponding rigid class of heads.
Appendix ii: Excluded parts of speech systems | 383

(17)*
Head Modifier
Reference – Adjective
The system in (18)* involves a flexible class of lexemes that can be used as the head of a referential phrase and
as a modifier in a predicate phrase (Flex C). It is excluded because it violates constraint (12a) in the reference
domain, where it has a rigid modifier class without a rigid head class.
(18)*
Head Modifier
Reference Flex C Adjective
The system in (19)* is excluded because it violates four constraints. It violates constraint (11b) because it has
a flexible lexeme class that can be used as the head of a referential phrase, but no lexeme class for heads of
predicate phrases. In addition, this system violates constraint (12a) in the reference domain, since it has a
specialized class of lexemes for referential modifiers, but no specialized class for referential heads. It also
violates constraint (12b), because it has a flexible class of lexemes that can be used as the modifier (but not the
head) of a predicate phrase, without a lexical class for heads of predicate phrases. Finally, this system violates
constraint (13): it has distinct classes for heads and modifiers in referential phrases, but no distinct class for
heads of predicate phrases as opposed to referential phrases.
(19)*
Head Modifier
Predication – Flex C
Reference Flex C Adj
The system in (20) is excluded because it violates (11b): it has a flexible lexeme class that can be used as the
head of a referential phrase, and as a modifier in a predicate phrase, but no lexical means to express the head
of a predicate phrase. In addition, this system violates constraint (12b) in the predication domain, where it
has a flexible lexeme class that can be used for modification, but no class for heads.
(20)*
Head Modifier
Predication – Flex C
The system in (21)* involves a flexible lexeme class that can be used as the head of a predicate phrase and as a
modifier in a referential phrase. It is excluded because it violates constraint (11a): It has a rigid class of heads
in the referential, but not in the predication domain. In addition, it violates constraint (12a) in the predication
domain, where it has a rigid modifier class but no rigid head class.
(21)*
Head Modifier
Predication Flex D Manner adverb
Reference Noun Flex D
The system is (22)* is excluded because it violates constraint (11a): It has a rigid class of heads in the referential,
but not in the predication domain.
(22)*
Head Modifier
Predication Flex D –
Reference Noun Flex D

The system in (23)* is excluded because it violates constraint (12a) in the predication domain, where it has
a rigid class of modifiers, but a flexible class for heads. In addition, this system violates constraint (12a) in
the reference domain: it has a flexible lexeme class that can be used as a modifier in a referential phrase, but
no lexical means to express the function of head of that phrase. Finally this system violates constraint (13):
it has distinct classes for heads and modifiers in the predication domain, but no distinct classes for heads of
referential phrases, as opposed to predicate phrases.
(23)*
Head Modifier
Predication Flex D Manner adverb
Reference – Flex D
The system in (24)* violates constraint (12b), since it has a flexible lexeme class that can be used as a modifier
in a referential phrase, but no lexeme class that can be used as the head of that phrase.
(24)*
Head Modifier
Predication Flex D –
Reference – Flex D
The system in (25)* is excluded since it has only one rigid class of heads, but in the reference rather than in
the predication domain. As such it violates constraint (11a).
(25)*
Head Modifier
Predication – –
Reference Noun –
The system in (26)* has a rigid class of modifiers in the referential domain, without the corresponding rigid
class of heads. As such it violates constraint (12a).
(26)*
Head Modifier
Predication – –
The system in (27)* also violates constraint (12a), since it has a rigid class of modifiers without the
corresponding rigid class of heads, but this time in the predication rather than the reference domain.
(27)*
Head Modifier
Reference – –
The system in (28)* has distinct classes of heads and modifiers in the predication domain, while it does not
have distinct classes for heads of predicative versus referential phrases. Thus, constraint (13) is violated.
(28)*
Head Modifier
Reference – –
The system in (29)* violates constraint (12a), since is has a rigid class of modifiers in the reference domain,
without the corresponding rigid class of heads. In addition, this system violates constraint (13) because it has
distinct classes of heads and modifiers in the predicate domain, but no distinct classes for heads of referential
phrases, as opposed to predicate phrases.
Appendix ii: Excluded parts of speech systems | 385

(29)*
Head Modifier
The system in (30)* is excluded because it violates (12a) in the reference domain, where it has rigid modifiers
without any class for heads.
(30)*
Head Modifier
The system in (31)* violates two constrains: (11a), since it has is a single rigid class of heads, but not in the
predication domain. And (13), since it has a head-modifier distinction in the referential domain, but no
distinct classes of lexemes for heads of predicative versus referential phrases.
(31)*
Head Modifier
Predication – –
The system in (32)* also violates two constraints: It has a single rigid class of heads, but not in the predication
domain, so that (11a) is violated. Furthermore, (12a) is violated in the predication domain, for which the
system has a rigid modifier class without the corresponding rigid head class.
(32)*
Head Modifier
Reference Noun –
The system in (33)* violates constraint (11a), since it has a rigid class for heads in the referential but not in the
predication domain. It violates (12a), since it has a rigid modifier class in the predication domain but no rigid
head class in that domain. And it violates (13), because it has a head-modifier distinction in the referential
domain, but no distinct classes for heads of predicative versus referential phrases.
(33)*
Head Modifier
The system in (34)* has two rigid classes of modifiers, without the corresponding rigid classes of heads, so
that (12a) is violated in both the predication and the reference domain.
(34)*
Head Modifier

Appendix iii: Dependent Clause Constructions Key Examples 1
Tagalog
pag-Pred (‘gerund’)
Functional distribution: F lex: Pred Head?, Ref Head, Ref Mod, Pred Mod = PoS minus Pred Head
(contentives) Himmelmann (2005: 372): “As with all Tagalog content words,
gerunds can be used in any syntactic function, provided their meaning fits.”
Structural type: 3 (D-ALT)
Verbal categories: No voice / mood marking, aspect can be expressed.
Nominal categories: Combines with the same phrase-marking function words as all other content
items. (Pred Head function is marked by sentence-initial position.)
Argument encoding: POSS - SENT 2
Examples:
Pred Head:
[pag-lu-lutoɁ ng pagkain] ang trabaho niyá
ger-rdp-cook gen food spec work 3sg.poss
‘His / her job is cooking food.’ (Himmelmann 2005: 372)
Ref Head:
pag-bawal-an mo ang bataɁ-ng iyó sa [pag-la-laróɁ sa lansangan]
sf-forbidden-lv 2sg.poss spec child-lk dist loc nmlz-rdp-play loc street
‘Forbid that child to play in the street.’ (Himmelmann 2005: 373)
Ref Mod: No example available
Pred Mod:
[pag-datíng naming doón] in-iwan namin don ang bangka
conv-arrival 1pl.excl.poss dist.loc real.uv-abandon 1pl.excl.poss dist.loc spec boat
‘When we arrived there we abandoned the boat, …’ (Himmelmann 2005: 373)
na / -ng( / kung) 3 + clause (=Ø)

Functional distribution: Flex: Ref Head, Ref Mod, Pred Mod = PoS minus Pred Head (contentives)
Structural type: 1 (Balanced)
Verbal categories: All retained
Nominal categories: Combines with the same phrase-marking function words as all other content
items. (Pred Head is marked by sentence-initial position.)
Argument encoding: SENT- SENT
Examples:
Ref Head:
Sinabi [kung maganda si Maria]
tell -.lk? beautiful nom Maria.
‘I said that Maria was beautiful.’ (Schachter & Otanes 1972: 173)
Gusto ni Pepito na [sagip-in ang dahon]

liking pn.poss Pepito lk salvage-pv spec leaf
‘Pepito wanted to catch the leaf.’ (Himmelmann 2005: 364)
Nakita ni Pedro[-ng / Manuel na [puno na ang bus]

see nom Pedro-lk / Manuel lk full lk rm bus
‘Pedro / Manuel saw that the bus was already full.’ (Schachter & Otanes 1972:177)
Ref Mod:
Sa mga lalaki na [maN-ibig nung kanyá-ng anák]
loc pl man lk av-love dist.gen:lk 3sgdat-lk child
‘(So he held a contest) between the men who courted his child.’ (Himmelmann 2005: 368)
1
The following symbols are used in this appendix: ‘=’ means ‘same functional possibilities as a PoS class in the same
language. This PoS class is added between brackets. ‘≠’ means ‘different functional possibilities than any PoS class in the
same language’. The relevant PoS classes, i.e. those that express the function(s) in which the DC is used, are added between
brackets. When there is no lexical class available for the relevant function(s) this is also indicated. For the meaning of other
abbreviations concerning the functions, the expression, and the classification of the DCs, see Chapters 3 and 6.
2
As Koptjevskaja-Tamm (1993: 119-120) explains, the SENT classification of coding of the second argument is not entirely
straightforward, because there is no difference between the marking of the second argument of an actor-voice predicate
and the possessor in Tagalog: both are marked by ng. However, there is a second type of possessive construction in which
the possessor is expressed as a sa-phrase. Since the first argument in a gerund construction can be both a sa and a ng-phrase
(just like possessors), while the second argument can only be a ng-phrase, Koptjevskaja-Tamm argues in favour of SENT
expression of the second argument.
3
The status of kung is unclear: it may be a combination of –ng with some other element.
387
Pred Mod:
[Biglá siyá]-ng nagbangon
sudden 3sg-lk real.av:rising
‘She got up quickly’ (Himmelmann 2005: 360)
Kharia
Pred-Ø / RDP (‘Freestanding form / masdar’)
Functional distribution: Flex: Pred Head (HAB), Ref Head, Ref Mod, (+ case) Pred Mod = PoS (contentives)
Verbal categories: No voice / tense marking and no Person agreement; retains valency-related marking
such as causative and passive / reflexive marking. In Pred Head function obligatorily
combined with the middle voice, indicating habituality.
Nominal categories: May take case and number
Argument encoding: POSS - SENT, occasionally also POSS - POSS
Examples:
Pred Head:
[iɲ ɖaɁ biɁɖ-biɁɖ]=ki=ɲ
1sg water pour.out-rdp=m.pst=1sg
‘I used to pour water out.’ (i.e. that was my job). (Peterson 2006: 74)
Ref Head:
[OɁ=yaɁ bay-bay] um=iɲ baɁj=ta.
house=gen build-rdp neg=1sg like=m.prs
‘I don’t like (the act of ) building houses.’ (Peterson 2006: 73)
Ref Mod:
[iɲ=aɁ dura=te ruɁ-ruɁ] kuɲji
1S=gen door-obl open-rdp key
‘The key I open the door with.’ (Peterson 2006: 73)
Pred Mod, without case-marking (with reduplication):

raʈa=te [ɖoko ɖoko] lemeɁɖ laɁ=ki
Rata=obl sit.down rdp sleep emot=m.pst
‘while he was seated, Rata became tired.’ (Peterson 2006: 249)
Pred Mod, with oblique case-marker:

[khõɖeɁj=ko ɖoko]=te=ga mukum
a.while=cntr sit=obl=foc doze.off=m.pst
‘Sitting for a while, he dozed off (…)’ (Peterson 2006: 295)
Pred=na (‘Infinitive’)
Functional distribution: F lex: Ref Head, Ref Mod, (+case / postposition) Pred Mod
= PoS minus Pred Head (contentives).
Verbal categories: No Voice /Tense / Person agreement
Nominal categories: Case
Argument encoding: POSS - SENT or Ø - SENT
Examples:
Ref Head:
iɲ [u ikuɁɖ sundar kontheɁɖ=ki=te bajhay=kon
1sg this very beautiful bird=pl=obl trap=seq.conv
‘I don’t want to trap and torment these beautiful birds.’ (Peterson 2006: 259)
Ref Mod:
Ho=kaɽ [khoɽi buli=na] modhe buŋ khaɽiya=ki=yaɁ jhaɽI
that=s.hum [village.section wander=ptc] means instr Kharia=pl=gen all
habhaw=te erikhudi koŋ may=sikh=oɁ
mannerisms=obl from.bottom.to.top know total=pfv=act.pst
‘She had learned by wandering through the village (= through the in the village wandering means) all of
the mannerisms of the Kharia inside out. (Peterson 2006: 306-307)
Pred Mod, without case marking (with reduplication):

…lekin [lam=na lamna] souɁb=te ikuɖ jughay ɖaɁ-piyas laɁ=ki.
but search=inf rdp all=obl very much water-thirst emot=m.pst
‘But searching and searching, [they] all became very thirsty.’ (Peterson 2006: 248)

Pred Mod, with oblique case-marking (and reduplication):
[Aw=na awna]=te khaɽiya=ki ho=te aɖoɁ oɁ
stay=inf rdp=obl Kharia=pl that=obl(=’there’) again house
dura bay=kon ikuɖ=ga memon jou aw=ki=may
door build=seq.conv very=foc year up.to stay=m.pst=3pl
‘Staying, the Kharia stayed there for several years, again building homes (houses and doors).’ (Peterson 2006: 249)
Pred=na-wala (‘participle’)
Remark: Borrowed from Hindi. Structural coding consists of the infinitive =na followed by =wala.
It denotes iterativity and habituality.
Functional distribution: Rig: Ref Mod ≠ PoS (contentives)
Structural type: 2 (D-SENT)
Nominal categories: None (no case agreement)
Argument encoding: The relativized argument is gapped, other argument(s) are SENT.
Example:
Ref Mod:
[Jharkanɖ=te aw=na=wala] lebu=ki ikuɖ jughay milansar aw=ta=ki
Jharkand=obl live=inf=ptc person=pl very much friendly cop=m.prs=pl
‘The people from Jharkhand are very friendly.’ (Peterson 2006: 307)
Pred-al (‘participle’):
Remark: The suffix attaches only to lexical predicates of Sadani origin which end in -a or -ay.
This suffix has thus been borrowed with the root.
Nominal categories: None (no case agreement)
Argument encoding: POSS - SENT
Example:
Muda moɲ Brahman [ho=koɽ=aɁ daru=te ʈaŋ-al] janew=te yo=yoɁ
But one brahman that=sg.hum=gen tree=obl hang=ptc holy.thread=obl see=act.pst
‘But a Brahman saw the holy thread which he had hung on a tree.’ (Peterson 2006: 307)
Pred-ker(r) / -kon / -kan (‘converb’)

Remark: “The first two of these markers are direct borrowings from Sadani. […] These markers denote,
among other things, that the two or more (sub-)predicates are portrayed by the speaker as being directly
related to one another in some way, combining to form a larger, more complex event. […] =kon […]
appears to be a calque from the Sadani form =ker. Like the cognate form -kar in Hindi, =ker in Sadani
appears to derive from the root kar ‘do’. The sequential converbal marker =kon in Kharia apparently derives
from a similar lexeme, i.e., ikon ‘make, do’. The core function of these forms is to denote the completion of
one action before another begins. [But they] are also often used to denote the manner in which an action is
carried out. In these cases, the action denoted by the converb is generally a more exact specification of that
of the morphologically finite predicate.” (Peterson 2006: 243-244)
Functional distribution: Rig: Pred Mod ≠ PoS (contentives)
Nominal categories: None.
Argument encoding: Ø - SENT
Examples:
Pred Mod:
…[lay koj=kon] goɁjuŋ bay=siɁ=may
dig scrape=conv path make=pfv=3pl
‘… they have built the path by digging an scraping [the dirt away].’ (Peterson 2006: 244)
raksin oɖoɁ jughay khisay=ta …ro [gone keb=kon] gam=te:..

witch even more angry=m.prs …and tooth grind=conv say=act.prs
‘The witch grows even angrier and … grinding her teeth, says:…’ (Peterson 2006: 244)
Pred=ga + RDP (‘(imperfective) converb’)

Remark: The =ga form is primarily a focus marker.
Nominal categories: None
Appendix iii: Dependent Clause Constructions Key Examples | 389

Example:
Pred Mod
ro ho kuda koloŋ=aɁ daru sumbhoɁ=te [iɲam=ga iɲamga] goj jom=ta
and that millet bread=gen tree base=obl cry=conv rdp die ap=m.prs
‘And crying and crying, she just died at the base of that millet bread tree.’ (Peterson in prep: 248)
Pred=ta (‘(imperfective) converb’)

=ta is homophonous with the general imperfective middle marker.
Example:
Pred Mod:
ele [am=pe=te goɁ=ta goɁta] han=tiɁj u=tɁj ɖoɽ=e=le
1pl.excl 2=2pl=obl carry.on.shoulders=conv rdp that=side this.side take=act.irr=1pl.excl
‘We will carry you around on our shoulders.’(=we will take you, carrying you on our shoulders’) (Peterson
2006: 248)
no / Ø + clause
Remark: Used for object complement clauses, especially with utterance predicates.
Functional distribution: Rig: Ref Head ≠ PoS (contentives)
Argument encoding: SENT - SENT
Example:
Ref Head:
ap=ɖom raʈa=te remakh-oɁ ro gam-oɁ [no babu musa iɲ kimir
father=3poss Rata=obl call=act.pst and say=act.pst comp child today 1sg forest
co=na um=iɲ pal=e]
go=inf neg=1sg be.able=act.irr
‘His father called Rata and said “child, to day I will be unable to go to the forest.’ (Peterson 2006: 298)
gam=kon + clause
Remark: The form gam=kon is the sequential converb of gam ‘to say’. It is occasionally found instead of no
as a kind of quotative form.
Functional distribution: Rig: Ref Head ≠ PoS (contentives)
Example:
Ref Head:
[Je janwar tar=e=ki ho janwar=yaɁ ghos ol=e=ki” gam=kon] gam-oɁ
cr animal kill=act.irr=pl that animal=gen meat take=act.irr quot say=act.pst
‘Whatever animal they kill, that animal’s meat they should bring, he said.’ (Peterson 2006: 299)
Correlative construction
Remark: There are two types of correlative constructions:
(i) With je-class markers: all correlative forms begin with j- and have been borrowed from
Indo-Aryan.
(ii) With a / i / other question particle-class markers: all correlative forms are homophonous with
interrogatives. This construction is not borrowed from Indo-Aryan, although it could be an older
calque of the Indo-Aryan correlative construction, using purely language-internal means.
In both construction types, the head is usually repeated in the main clause, preceded by a demonstrative.
Alternatively, the head is not repeated and only the demonstrative is there.
Argument encoding: SENT - SENT (optional gapping)

Examples:
Ref Mod:
(Je-class)
…aɖi [je bhere] eŋ=ki se bhere aɖi=ya poʈom=te soreŋ kui=ki
anaph cr time return=m.pst dem time anaph=gen bundle=obl stone find=m.pst
‘Which time he returned, (at) that time he found a stone in the bundle.’ (Peterson 2006: 302)
(a / i-class)
[a=boɁ=te pujapaʈh karay=na aw=ki,] ho boɁ=te ɖam=ke, ….
q=place=obl sacrifice do=inf cop=m.pst dem place=obl arrive=seq.conv
‘Having arrived at the place where the sacrifice was to be done … (Peterson 2006: 302)
Unmarked gapped relative clauses with a (partially) finite predicate

Remark: This construction is balanced, except that person marking may be lost on the dependent predicate
(see second example).
Argument encoding: SENT - Ø / Ø - SENT (gapping)
Examples:
Ref Mod:
[iɲ yo=yoɁj] lebu=ki iɲ=aɁ hoʈel=te aw=ta=ki
1sg see=act.pst.1sg person=pl 1sg=gen hotel=obl live=m.prs=pl
‘The people I saw live in my hotel.’ (Peterson 2006. 303)
[iɲ=te yo=yoɁ] lebu=ki ulaɁ likha=yoɁ=ki

1s=obl see=act.pst person=pl letter write=act.pst=pl
‘The people who saw me wrote a letter.’ (Peterson 2006. 303)
Kambera
Pa-deranked clause
Remark: In combination with the prepositional verb wàngu ‘use’ this construction can be used as an adverbial
clause with an interpretation of simultaneity or immediate sequence (see Pred Mod example below).
Functional distribution: F lex: Ref Head (same-subject), Ref Mod (object of DC), (+ prep / prepositional
verb: Pred Mod) = PoS minus Pred Head (contentives)
Structural type: 2 / 3 (D-SENT / D-ALT)
Verbal categories: No aspect, no mood marking. (Kambera has no tense marking.)
Nominal categories: DET + number agreement (in Ref Mod function)
Argument encoding: Ø - SENT/POSS - Ø In Ref Head function the subject remains unexpressed
under co-referentiality; the object is SENT and cross-referenced on the dependent
predicate, in the DAT form. In Ref Mod function the subject is POSS, and the object
is gapped, but remains cross-referenced on the dependent predicate in the DAT form.
Examples:
Ref Head:
Ta-pakiring [pa-tinu-nya na lau] haromu
1pl.nom-start comp-weave-3sg.dat art sarong tomorrow
We will start to weave the sarong tomorrow.’ (Klamer 1998: 338)
Ref Mod:
Ta-pakiri-nja da lau [pa-tinu-nda]
1pl.nom-start-3sg.dat art sarong rel-weave-1pl.dat
‘We start (with) (them) the sarongs woven by us.’ (Klamer 1998: 338)
na kalembi na [pa-kei wà-nggu-nya]

art shirt art rel-buy use-1sg.gen-3sg.dat
‘the shirt that I bought’ (Klamer 1998: 326)
Pred Mod:
Patiang ana mandài-ndài [wàngu pa-buta ana rumba]
wait dim rdp-belong use comp-pick dim grass
‘(We) wait a while weeding some grass in the meantime.’ (Klamer 1998: 240)
ma-deranked clause
Functional distribution: Rig: Ref Mod (subject / possessor clauses) ≠ PoS (contentives)
Verbal categories: No aspect, no mood marking. (Kambera has no tense marking.)

Nominal categories: DET + number agreement (in Ref Mod function)
Argument encoding: Ø - SENT (gapping)
Examples:
Ref Mod:
Na-meti-ka na tau na [ma-piti-ya na kabela-nggu]
3sg.nom-die-pfv art person art rel-take-3sg.acc art machete-1sg.gen
‘The person that took my machete died already.’ (Klamer 1998: 315)
Ita-nggu-nya na tau na [ma-meti kuru uma-na].

see-1sg.gen-3sg.dat art person art rel-die wife-3sg.gen
‘I saw the man whose wife died.’ (Klamer 1998: 320)
Unmarked nominalized clause

Remark: Nominal clauses can be dependent or independent. In combination with a conjunction, the construction
can apparently also be used in adverbial function as a simultaneity clause (see Pred Mod examples below).
Functional distribution: F lex?: Pred Head / main clause, Ref Head (+ CONJ also Pred Mod, simultaneity).
≠ PoS (contentives, adverbs), but apparently Pred Head / Ref Head flexibility.
Structural type: 3 (ALT-SENT)
Verbal categories: (Some) aspect and mood marking is retained (also lexically). (Kambera has no
tense-marking).
Nominal categories: DET, when functioning as such construction is cross-referenced on the main
predicate as an object, with a DAT form.
The subject is expressed through a genitive pronominal enclitic.
Examples:
Pred Head:
[Na apu-mu, katuda-na] la pino bolsak-ka una
art grandmother-2sg.gen sleep-3sg.gen loc top mattress-pfv emph.3sg
‘Your granny, she will sleep on a mattress.’ (Lit. ‘Your grandmother’s sleeping is on a mattress.’ (Klamer
1998: 97)
Ref Head:
Nda ku-pí-anggau [na ngàndi-mu rú kuta]
neg 1sg.nom-know-mod-2sg.dat art take-2sg.gen leaf pepper plant
‘I dind’t know that you would bring kuta.’ (lit.: I didn’t know (of ) your bringing kuta.’) (Klamer 1998: 97)
Nda ku-mbuti-nya [na taka-mu]

neg 1sg.nom-expect-3sg.dat art arrive-2sg.gen
‘I did not expect your coming.’ (Klamer 1998: 315)
Pred Mod:
[Ba meu-meu-na,] ba na-imbu-ya
conj rdp-roar-2sg.gen conj 3sg.nom-search-3sg.acc
‘And it roared (a tiger), while it went after him.’ (Klamer 1998: 100)
[ba jiapa-ma na hanggapit-na-nya dá]

conj continously-emph art clasp-3sg.gen-3sg.dat inside
‘while he kept clasping it inside…’ (Klamer 1998: 97)
wà + clause
Remark: Quotative construction.
Functional distribution: Rig: Ref Head ≠ PoS (contentives, adverbs)
Example:
Ref Head:
Ka [na-ngàndi-ya na mbuku] wà-nggu-nya làti
conj 3sg.nom-take-3sg.acc art book say-1sg.gen-3sg.dat in fact
‘In fact, I told him that he should take the book.’ (Klamer 1998: 347)
Samoan
Pred-ga
Functional distribution: Rig: Ref Head, ≠ PoS (contentives, adverbs)

Verbal categories: No TAM marking
Nominal categories: DET / CASE
Argument encoding: POSS - SENT (occasionally also POSS - POSS)
Examples:
Ref Head:
A le faalavelave le tupu i [le ai-ga
pst neg trouble art king ld art eat-nmlz
apu ma moli a le pipili ma le tauaso]
apple and citrus poss art lame and art blind
‘The king was not troubled that the lame and the blind ate the apples and oranges.’
Ae na oo lava in moumou malie atu le pisa

But pst reach emph conj disappear gentle dir art noise
o [e sapini-ga o Pale ma Maria e o la Tina]
poss art whip-nmlz poss Pale and Maria erg poss 3du mother
‘But finally the noise of the whipping of Pale and Maria by their mothers gently faded away.’
(Mosel 1992: 279)
Unmarked nominalized clause

Functional distribution: Rig: Ref Head, ≠ PoS (contentives, adverbs)
Nominal categories: DET / CASE
Example:
Ref Head:
E lelei [l-a-u tunu ia]
genr good art-poss-2sg roast fish
‘Your fish roasting is good.’ (Mosel 1992: 267)
ona / ina + deranked clause

Remark: Ona is used for core-arguments, ina for adjuncts.
Functional distribution: Rig: Ref Head ≠ PoS (contentives, adverbs)
Example:
Ref Head:
ua taga [ona inu ava malosi tatou]
pfv allowed comp drink ‘kava’ strong 1.incl.pl
‘It is allowed that we drink alcohol.’ (Mosel & Hovdhaugen 1992: 599)
-e + clause
Functional distribution: Rig: Ref Mod ≠ PoS (contentives, adverbs)
Nominal categories: (DET)
Argument encoding: SENT - Ø / Ø - SENT (gapping) Gap can be filled with anaphoric element.
Example:
Ref Mod:
‘O lea ‘o le tama’ita’i l-[ē na tatou ō ‘i ai.]
pres that pres art woman art-rel pst 1.inc.pl go(pl) ld anaph
‘She is the woman we went to find.’ (Mosel & Hovdhaugen 1992: 635)
Unmarked clause
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (contentives, adverbs)

Examples:
Ref Head:
Na iloa e Tigilau [úa sau Sina]
past know erg Tigilau pfv comes Sina
‘T knew that S had come.’ (Mosel & Hovdhaugen 1992: 589)
Ref Mod:
Ua tu le alii lea [na ua e Popi]
pfv stand.up art man that pst bite erg Popi
‘The man who was bitten by Popi stood up.’ (Mosel & Hovdhaugen 1992: 635)
Guaraní
clause + há / hagwé / Ø
Functional distribution: Rig: Ref Head ≠ PoS (contentives, verbs)
Examples:
Ref Head:
Rey-anú [šé še-rasɨ̀ há]
you-hear I I-be.sick comp
‘You heard that I was sick.’ (Gregores & Suárez 1967:158)
ai-kwaá la [n o-ù moɁã́ i há]

I-know def(det) neg he-go mod neg comp
‘I know that he does not intend to go.’ (Gregores & Suárez 1967:158)
Rei-moɁã́ [še-tavɨ́]
You-think I-be.silly
‘You think that I am silly.’ (Gregores & Suárez 1967: 157)
Pred-va + clause 4
Functional distribution: Rig: Ref Mod ≠ PoS (contentives, verbs)
The relativized item is gapped, but there is a person prefix on the
dependent predicate.
Example:
Ref Mod:
A-hechal a karai [o-jagua-va-ekue ka angu’a].
I-see def man 3-buy-rel-pst def mortar
‘I saw the man who bought the mortar.’ (Velázquez-Castillo 2002: 162)
Clause + vo:
Functional distribution: Rig: Pred mod ≠ PoS (contentives, verbs)
Pred Mod:
H-asẽ́ [o-sẽ́ vo]
she-cry she-go.out adv
‘She goes out crying.’ (Gregores & Suárez 1967: 180)
Santali
Unmarked clause; pred without -a (IND)
Remark: In Ref Head subject function, the construction shows no subject marking, while middle voice
markers and TAM can be expressed. In Ref Head object function subject marking and TAM are lost, while
object markers are retained. Only with verbs of perception subject marking and TAM can be expressed. In
Ref Mod function, subject pronominals are omitted, but all TAM suffixes can be expressed.
4
The status of the relativizer is not entirely clear: it does not seem to change the internal syntax of the dependent clause, but
does attach to the predicate, preceding the tense marker.

The relativized element is gapped. In Pred Mod function, with case-marking, there is neither person
marking (subject / object) nor TAM marking, but middle voice can be expressed. In combination with a
postpostion (with similative semantics), all verbal categories are expressed, except for the indicative marker.
Functional distribution: Flex: Ref Head, Ref Mod (+ case (LOC / INSTR) / postposition also Pred Mod)
≠ PoS (contentives, verbs)
Verbal categories: Variable (see above)
Nominal categories: CASE
Argument encoding: SENT - Ø / Ø - SENT (co-referentiality / gapping)
Examples:
Ref Head:
Subject:
[ber həsur-kate ḍera-k’-dɔ] baŋ bes-a
sun set-conv camp-m-top neg good-ind
‘it is not good to camp after sunset.’ (Neukom 2001: 181)
Object:
[onko əgu-ko] mana-ko-m
those:pl bring-3pl.obj forbid-3pl.obj-2sg.sbj
‘Forbid them to bring those.’ (Neukom 2001: 182)
Perception predicate:
[mɔ̃ṛɛ̃-gɔṭɛn əiməi dak’ lo=ko hij-uk’kan-e] ɲel-gɔt’-ket’-ko-a
five-cl woman water fetch-3pl.sbj come-m-ipfv-3sg.sbj see-v2-pst:act-3pl.obj-ind
‘He saw five women come to fetch water.’ (Neukom 2001: 183)
Ref Mod:
Uni-y-[e [bujhəu-ɲɔ̃k’-ket’] hɔṛ-e sərat-gɔt’-ad-e-a.
that-(anim)y-3sg.sbj understand.-little-pst:act person-3sg.sbj beckon-v2-appl:pst:act-3sg.obj-ind
‘He beckoned the man who had understood a little.’ (Neukom 2001: 197)
Pred Mod:
Case-marked:
[cala-k’-calak’-te] mit’-ṭaŋ toyo-ko ɲɛl-tiok’-ked-e-a
go-m-rdp-instr one-cl jackal-3pl.sbj see-reach-pst:act-3sg.obj-ind
‘While they were walking along, they caught sight of a jackal.’ (Neukom 2001: 187)
uni gidrə-dɔ [hud̠iɲ-ge-y-e tahɛ̃kan-re]

that child-top small-foc-y-3sg.sbj cop:pst-loc
eŋga-t-tɛt’-dɔ-e gɔc’-en-a
mother-3poss-3poss-top-3sg.sbj die-pst:m-ind
‘While the child was (still quite) young, his mother died.’ (Neukom 2001: 189)
With postposition (similative):

[ɔnɛ-y-e met’-at’-ko leka-ge] jɔtɔ hɔṛ-ko cala-k’-kan-a
that-y-3sg.sbj say-appl:pst:act-3pl.obj like-foc all person-3pl.sbj go-m-ipfv-ind
‘They all went along as he had told him.’ (Neukom 2001: 195)
Pred-kate 5
Functional distribution: Rig: Pred Mod ≠ PoS (contentives, verbs)
Verbal categories: No Tense / Mood, no Person 6, voice can be expressed
Example:
Pred Mod:
[[nonka hudis-kate] bɛbhɔrsa-kate-ko] cala-k’-kan-a
like.this think-conv be.hopeless-conv-3pl go-m-ipfv-ind
‘Thinking so and being hopeless, they walked on.’ (Neukom 2001: 186)
5
This construction is described as a converb but its status is not completely unambiguous: “-kate occurs elsewhere in
isolated position as ‘then’ or together with deictic elements such a nit ‘now’ (cf. nit-kate ‘nowadays’) or ona ‘that’ (cf. ona-kate
‘thereupon’), or in combination with numerals, e.g. ponea-kate ‘(give them) four each’.” (Neukom 2001: 185)
6
Note that the subject pronominal is nevertheless retained in the second converbal form of the example.

Correlative construction
Remark: Correlative constructions are probably an influence from Indo-Aryan languages. These
constructions make use of various types of pronouns: interrogative, indefinite, demonstrative. The dependent
predicate often lacks the indicative marker -a, but the pronominal subject clitic is always present.
Functional distribution: Rig: Ref Mod ≠ PoS (contentives, verbs)
Verbal categories: Allmost all retained (see above)
Example:
Ref Mod:
ona dare [oka-m mak’-akat’]
that(inanim) tree which-2sg.sbj cut-pfv:act
‘the three which you have cut’ (Neukom 2001: 200)
Clause + mɛ nte
Remark: The complementizer / quotative is a lexicalized instrumental case-marked form of ‘to say’.
Used for complements of predicates of utterance, thought, and mental perception.
Functional distribution: Rig: Ref Head ≠ PoS (contentives, verbs)
Example:
Ref Head:
Ba-kin ɲɛl-ṭhik-e-kan-a, nui-dɔ
neg-3dual.sg see-correct-3sg.obj-ipfv-ind this(an)-top
[əkin-ren apa-t kan-a-e mɛnte]
they(dual) father-3poss cop-ind-3sg.sbj comp / quot
‘They did not recognize that he was their father.’ (Neukom 2001: 183)
Warao
Pred-kitane (‘infinitive’):
Remark: According to Romero-Figeroa (1997) this construction is used for same-subject complements and
for purpose-clauses, but no example is available of the former use.
Functional distribution: Rig: Pred Head ≠ PoS (non-verbs)
Verbal categories: None
Argument encoding: Ø - SENT (co-referentiality)
Clause + kotai:
Functional distribution: Rig: Ref Mod ≠ PoS (non-verbs)
Example:
Ref Mod:
Ima-ya domu [nari-te kotai] mi-kitane nao-kotu
night-all bird fly-n.pst rel see-inf come-2pl.imp
‘You all, come to see the bird that flies at night.’ (Romero-Figeroa 1997: 42)
Turkish
Pred-DIK / -(y)AcAK
Remark: In combination with the postposition gibi this construction can be used as a similative adverbial clause.
Functional distribution: F lex: Ref Head, Ref Mod (non-subject / possessor clauses), (+ postposition also
Pred Mod, similative). ≠ PoS (non-verbs, derived modifiers)
Verbal categories: No Aspect and Mood marking, (relative) tense is expressed by the choice of
marker: -DIK for past and present, -(y)AcAK for future.
Nominal categories: CASE, nominal agreement
Argument encoding: POSS - SENT / Ø - POSS
(In Ref Mod function, the relativized element is gapped; the subject is POSS.)

Examples
Ref Head:
(ben) [Ahmed-in öl-düg-ün]-ü duy-du-m
‘I heard that Ahmed died.’ (Kornfilt 1997: 50)
[Orhan-ɪn bir şey yap-ma-yacağ-ı] belliydi.

Orhan-gen anything do-neg-nmlz-3sg.poss it.was.obvious
‘It was obvious that Orhan wouldn’t do / wasn’t going to do anything.’ (Göksel & Kerslake 2005: 423)
Ref Mod:
[adam-ın git-tiğ-i] okul
‘the school that the man goes / went to’ (Kornfilt 1997: 50)
[Fatma-‘nın yarın gör-eceğ-i] film

‘the film that Fatma is going to / will be seeing tomorrow’ (Göksel & Kerslake 2005: 442)
Pred Mod:
Pastayı [anne-m-in analat-tɪğ-ɪ gibi] yapmaya çalıştım
mother-1sg.poss-gen describe-nmlz-3sg.poss like
‘I tried to make the cake [as my mother had described].’ (Göksel & Kerslake 2005: 477)
Pred-mAK
Remark: “The crucial difference between -mAK clauses and those with -mA is that -mA clauses in the
majority of cases contain their own subject, whereas -mAK clauses do not.” (Göksel & Kerslake 2005: 413)
Verbal categories: No TAM and Person agreement
Nominal categories: CASE (usually when functioning as a direct object complement, except with the
verb iste- ‘to want’, see third example)
Examples:
Ref Head:
[Lütfen pencere-yi aç-mağ]-ı unut-ma
please window-acc open-inf-acc forget-neg
‘Please, don’t forget to open the window!’ (Kornfilt 1997: 51)
[(ben) Ahmed-ı kaç-mağ]-a zorla-dı-m

I Ahmed-acc flee-inf-dat force-pst-1sg
‘I forced Ahmet to flee.’ (Kornfilt 1997: 51)
[Sokaǧa çik-mak] isti-yir-um

go out-INF want-IPFV-1SG
‘I want to go out.’ (Göksel & Kerslake 2005: 413)
Pred-mA
Remark: In general terms noun clauses formed with -mA are less abstract in meaning than those
formed with -mAK.
Verbal categories: No TAM and Person agreement
Argument encoding: POSS - SENT (co-referentiality)
Examples:
Ref Head:
[Kerkes-in birier kikaye anlat-ma-sɪ] iste-n-iyor-muş
everyone-gen one.each story tell-nmlz-3sg.poss want-pass-ipfv-ev.cop
‘It seems they want [everyone to tell a story].’ (Göksel & Kerslake 2005: 420)
Ahmed-e [ben-i bekle-me-sin]-i söyle-di-m

a.-dat I-acc wait-nmlz-3sg.poss-acc tell-pst-1sg
‘I told Ahmet to wait for me.’ (Kornfilt 1997: 53)
Pred-An
Functional distribution: Rig: Pred Mod (subject / possessor clauses) ≠ PoS (non-verbs, derived modifiers)

Verbal categories: No TAM, no Person Agreement (verbal)
Nominal categories: None.
Examples:
Ref Mod:
[burada sat-ɩl-an] kitap-lar
here sell-pass-ptc book-pl
‘the books that are sold here’ (Göksel & Kerslake 2005: 440)
[ögretmen ol-an] haydar

‘Haydar, who is a teacher’ (Göksel & Kerslake 2005: 440)
[araba-sɩ çal-ɩn-an] komşu-muz

car-3sg.poss steal-pass-ptc neighbour-1pl.poss
’our neighbour, whose car was stolen’ (Göksel & Kerslake 2005: 440)
Pred-(y)ArAK
Remark: Normally, the subject is unexpressed under co-referentiality.
Functional distribution: Rig: Pred Mod ≠ PoS (non-verbs, derived modifiers)
Verbal categories: No TAM / Person agreement
Example:
Pred Mod:
Ben [etraf-ım-a bak-arak] yür-ür-üm
I around-1sg-dat look-conv(manner) walk-aor-1sg
‘I walk looking around (myself ).’ (Kornfilt 1997: 73)
Pred-(y)A… Pred-(y)A
Remark: “This construction occurs either with identical verb stems or with different ones. Its use is less
widespread than that of -(y)ArAk, and its meaning is more emphatic, stressing the continuous or repeated
nature of the action it expresses. The forms involving two different verb stems are for the most part
lexicalized items.” (Göksel & Kerslake 2005: 476)
Functional distribution: Rig: Pred Mod ≠ PoS (non-verbs, derived modifiers)
Verbal categories: No TAM / Person agreement
Argument encoding: Ø - Ø ( / Ø - SENT?)
Pred Mod:
Genç kadɩn [ağala-ya agala-ya] hikayesini anlattɩ
The young woman told her story [continuously weeping]’ (Göksel & Kerslake 2005: 476)
Adam [it-e kak-a] öne geçmeye çalɩşɩyordu

man [pushing shoving] was trying to get to the front
‘[Pushing and shoving,] the man was trying to get to the front.’ (Göksel & Kerslake 2005: 476)
ki + clause
Remarks: This construction is borrowed from Persian.
In relative clause function, the construction is mostly non-restrictive (the head is almost always the subject
of the main clause, and 3rd person singular or plural).
Normally, the relativized item is gapped, but under certain circumstances, it may or must be reiterated in
the dependent clause, by means of a resumptive pronoun.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (non-verbs, derived modifiers)
Argument encoding: SENT / Ø - SENT / Ø (gapping in Ref Mod function)
Examples:
Ref Head:
Isti-yor-um [ki yarin ben-imle sinema-ya gel-esin]
want-pres.progr.1sg comp tomorrow I-gen-with cinema-dat come-sg.opt
‘I want you to come to the movies with me tomorrow. (Kornfilt 1997: 46)

Sanɩyorum [ki iş-in-i bɩrak-mak isti-yor]
I.think comp job-sg.poss-acc leave-nmlz want-ipfv
‘I think that s / he wants to quit his / her job.’ (Göksel & Kerslake 2005: 409)
Ref Mod:
Bir adam [ki çocuk-lar-ın sev-me-z] yalniz yaşa-malı-dır
a man rel child-pl.3sg-acc love-neg-aor alone live-neg-ep.cop
‘A man who does not love his children must live alone.’ (Kornfilt 1997: 60)
bi ahçɩ [ki baklava yap-may-ɩ- bil-me-sin]

a cook rel baklava make-nmlz-acc know-neg-3sg.opt
‘a cook who doesn’t know how to make baklava’ (Göksel & Kerslake 2005: 459)
Clause + diye
Remark: diye is the converbal (-(y)A) form of the verb de ‘to say’. The construction is used for complements of
predicates of speech other than de, and for complements of predicates of cognition, perception, and emotion.
Functional distribution: Rig: Ref Head ≠ PoS (non-verbs)
Example:
Ref Head:
Meral [Turgut onu Selim’le gör-ür-se diye] kork-uyor-du
Meral Turgut with Selim see-aor-con.cop comp / quot be.afraid-ipfv-pst.cop
‘Meral was afraid that Turgut would see her with Selim.’ (Göksel & Kerslake 2005: 409)
Unmarked clause
Remark: Used for complements of de ‘to say’, and for complements of predicates of believe and desire.
Functional distribution: Rig: Ref Head ≠ PoS (non-verbs)
Example:
Ref Head:
Herkes [sen sinema-ya git-ti] san-ıyor
everybody you(nom) cinema-dat go-pst-2sg believe-prs.progr
‘Everybody believes that you went to the movies.’ (Kornfilt 1997: 47)
Kayardild
Pred-n(-marri) (‘plain / privative nominalization’)
Remark: Active plain nominalizaions may, apart from their dependent uses, function as main clauses
describing ongoing, uncompleted actions (Evans 1995: 472).
In Ref Head function, the construction can occur only as the complement of a perception predicate.
In Ref Mod function, it is used when the subject of the dependent clause is relativized. This subject is gapped.
Functional distribution: F lex: Ref Head (perception complements), Ref Mod (subject clauses), Pred Mod =
PoS (non-verbs).
Verbal categories: No TAM
Nominal categories: Nominal agreement
Argument encoding: All overt arguments take the associating oblique case (A.OBL) or
proprietive / locative modal case (MPROP / MLOC).
Examples:
Ref Head:
Ngada kurri-ja [ki-l-wan-ji dalwani-n-ki thawal-urrk]
1sg.nom see-act 2-pl-poss-mloc dig.up-nmlz-mloc yam-mloc:a.obl
‘I saw you digging up yams.’ (Evans 1995: 472)
Ngada kurri-ja [niwan-ji budii-n-marri]

1SG.nom see-act 3sg.poss-mloc run-nmlz-priv
‘I saw that he was not running.’ (Evans 1995: 476)

Ref Mod 7:
Nga-ku-l-da [wirr-n-ku] dangka-wu kurri-ju
1-inc-pl-nom dance-nmlz-mprop man-mprop see-pot
‘We will watch the dancing man.’ (Evans 1995: 474)
Pred Mod:
[Bilaangka-nurru kari-i-n-da] ngada warra-j
blanket-assoc cover-m-nmlz-nom 1sg.nom go-act
‘I went along, covering myself in a blanket.’ (Evans 1995: 474)
Diya-ja wuran-ki [kinaa-n-marri]
eat-act food-m.loc tell-nmlz.priv
‘(He) eats food without telling (anyone).’ (Evans 1995: 475)
Pred-Thirri-n (‘resultative nominalization’)

Remark: This construction can also be used as a main clause, and as an adverbial clause expressing
temporal sequence.
Functional distribution: Rig: Ref Mod = PoS (adjectives)
Argument encoding: Ergative alignment: Objects of transitive verbs and subjects of intransitives (S and P)
take nominative case; (demoted) subjects of transitive verbs (A) take oblique case.
Examples:
Ref Mod:
Bath-in-ki bal-umban-ji [niwan-jiyarrng-niaba-ya
west-from-mloc west-orig-mloc 3sg.poss-du-abl-mloc
jibarna-yarrng-kinaba-ya bidiru-thirri-n-ji]
uncle.in.law-du-abl-mloc miss-res-nmlz-mloc
‘One coming from the west, that had been missed by his two uncles-in-law.’ (Evans 1995: 480)
Nyingka kada buru-tharra [wungi-jirri-n-jina] mala-na

2sg.nom again get=pst steal-res-nmlz-abl beer-m.abl
‘Did you get some stolen beer again?’ (Evans 1995: 479)
Pred-n-ngarrba (‘consequential nominalization’)

Remark: The construction is rarely used as a main clause describing actions preceding the temporal
reference point. (Evans 1995: 481) All arguments are marked with a consequential suffix. This kind of
marking resembles so-called ‘complete concord’ (all elements of a constituent are marked for case; see
Dench 2006). This means that the consequential form looks like a case-marker with the function of a
complementizer. Therefore, the construction is classified as a D-SENT, rather than a D-ALT construction.
Nominal categories: Resultative case (see above)
Argument encoding: Resultative case (see above)
Example:
Ref Mod:
Nyinka kamburi-ja dathin-a dangka-a [yarbu-nyarrba balangkali-ngarrba ba-yii-n-ngarrb!]
2sg.nom speak-imp that-nom man-nom snake-cons brown.snake-cons bite-m-nmlz-cons
‘You speak to that man who was bitten by a brown snake!’ (Evans 1995: 481)
Clause + -ntha (‘oblique complementizer case’)

Remarks: This construction “closely resembles normal finite clauses, permitting almost the full range of
verb inflections 8. (…) Commonly a Complementizing Oblique or Locative case appears after all other
inflections, usually on all constituents.” (Evans 1995: 488)
In Ref Head function, this construction is used for complements of predicates of perception,
7
There are no examples available of the privative nominalization in Ref Mod function, but according to Evans (p.c.) this is
possible: “I’m sure you can say it, but it’s a gap – probably accidental – in my data.”
8
The tense-system deviates slightly from independent clauses:
Independent clause: - ACT(ual), which covers present, past and immediate future, the
latter two of which can be marked if extra precision is desired.
- POT(ential)
Dependent clause: - PAST
- IMMED (= present and immediate past)
- POT(ential)
- ACT can not be expressed
Modal case marking is the same in independent and dependent clauses.

knowledge and speech.
The morpheme -(u)rrka is a special portmanteau for LOC + C.OBL. It is used for locative complement
clauses (Evans, p.c.).
In Ref Mod function, this construction is used only when the relativized item is not the subject of
both clauses.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (non-verbs, adjectives)
Verbal categories: Largely retained (see above)
Nominal categories: Complementizer case (see above)
Argument encoding: Complementizer case (see above)
Examples:
Ref Head:
Ngada mungurru [(ngijuwa) kada-ntha thaa-thuu-nth]
1sg.nom know.nom 1sg.subj:c.obl again-c.obl return-pot-c.obl
‘I know that I will come back again.’ (Evans 1995: 490 / 491)
Ngada kamburri-ja niwan-ji [walbu-ntha dathin-inja barji-nyarra-nth]

1sg.nom say-act 3sg-mloc raft-c.obl that-c.obl capsize-appr-c.obl
‘I told him the raft would capsize.’ (Evans: 516)
Ngada marin-marri-i-jarr [dathin-kurrka thungal-urrka kamburri-jurrk]

1sg.nom self-hear-m-pst that-loc:c.obl thing-loc:c.obl speak-immed:c.obl
‘I heard myself speaking on that thing (the radio).’ (Evans 1995: 491)
Ref Mod:
nyinka kurri-jarra dathin-kina dangka-na
2sg.nom see-pst that.mabl man-mabl
[thawurr-inaa-ntha raa-jarra-ntha niwan-jinaa-nth]
throat-mabl-c.obl spear-pst-c.obl 3sg-mabl-c.obl
‘Did you see the man whom (he) speared in the throat?’ (Evans 1995: 490)
Unmarked clause
Remark: This construction is used in Ref Mod function, in cases where no complementizer case appears,
i.e. when the relativized element is the subject of the relative clause. Usually, the relativized element is
gapped, but it may also be retained.
Verbal categories: Largely retained (see above)
Argument encoding: SENT- SENT / Ø - SENT (gapping)
Example:
Ref Mod:
Jina-a dathin-a dangk-a, [dan-kina yii-jarrma-tharra wangal-kina]
where-nom that-nom man-nom, here-m.abl put-caus-pst boomerang-m.abl
‘Where is the man, who left the boomerang here?’ (Evans 1995: 489)
Paiwan
tu(a) / tjai + clause
Remark: The oblique marker tu(a)/tjai is also used for non-clausal arguments. Semantically, it is used for patients,
beneficiaries, instruments, goals, objects of comparison etc. With DCs it marks “less integrated” complements.
Functional distribution: Rig: Ref Head = PoS (nouns)
Verbal categories: Retained
Examples:
Ref Head:
ma˙rhekutj ti-naju [tu laq˙laq-en ni-a-maju]
fear foc-he obl tickle-pat defoc-ag-pl-he
‘He is afraid that she will tickle him.’ (Egli 1990: 177)
na ma˙saru a rhavats [tu rigu ti Yohan].

pfv believe lk very obl prophet foc John
‘He believed strongly that John was a prophet.’ (Egli 1990: 192)

pa˙telip [tua ini pu˙valav-an ni-maju tua qatjuvi]
regret obl not marry-loc defoc.ag-he obl snake
‘She regretted that she hadn’t married the snake.’ (Egli 1990: 202)
a + clause
Remark: The focus marker / linking element a is also used for non-clausal constituents. This construction is
used for subject clauses, and for “more integrated”object complements, such as with modal predicates.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, small / derived adjectives)
Argument encoding: SENT / Ø - SENT / Ø (gapping)
Examples:
Ref Head:
na˙nguaq [a ma-ngetjez sun]
good foc pass-come you
‘It is good that you have come.’ (Egli 1990: 230)
ini˙ka ma˙qati [a tja-parh-pa˙tsun-en]

not can lk we-one.another-see-pat
‘We cannot see one another.’ (Egli 1990: 230)
Ref Mod:
qala [a na tem-ker tua vaua]
stranger lk pfv drink-ag obl wine
‘the stranger, who has drunk wine’ (Egli 1990: 178)
[a˙ zu’ a i-vetsik a] kai

foc those lk pfv-write lk word
‘the word that I have written’ (Egli 1990: 271)
[pin-˙atsay˙ an nua valau a] vavay˙an

die-pfv loc defoc.ag husband lk woman
‘the woman whose husband has died’ (Egli 1990: 183)
a parhu + clause
Functional distribution: Rig: Pred Mod ≠ PoS (no manner adverbs)
Example:
Pred Mod
sa ringul-˙i sun [a parhu qemlev]
and be.around-pron.pat you lk like put.a.cover.on
‘And she will be around you as if she would want to cover you.’ (Egli 1990: 209)
((-)in- Pred -an + a

Remark: This is a perfective participle construction, formed with the perfective affix (-)in (infix for verbs
with a consonant in the Anlaut, prefix for verbs with a vowel in the Anlaut), which is also used with
independent verb forms, and the participial suffix -an9. Since there is no inflectional verbal morphology, it
is hard to say whether this is a deranked construction. However, participles are restricted to the extent that
they cannot occur with other tenses (marked with particles), nor with focus markers and they cannot be
transitive. The participal form must be combined with the linker a.
Functional distribution: Rig: Ref Mod = PoS (small / derived adjectives)
Verbal categories: Restricted tense
Argument encoding: Ø (gapping, intransitive)
9
Paiwan also has a present participle form, with –an and without the perfective marker, but this form is hardly ever used.
Like the perfective participle, it can be used as a modifier in a referential phrase. In addition, it can be used for manner
expressions, i.e. as a modifier in a predicate phrase:
rhemaketj-an a masengseng
to.do.all.day.long-ptc lk work
‘to work steadily.’ (Egli 1990: 124)

Example:
Ref Mod:
k-in-a’aluj’-an a impits
?-pfv-point(pass)-ptc lk pencil
‘the pointed pencil’ (Egli 1990: 122)
Imbabura Quechua
Pred-j / -shka / -na
Functional distribution: Flex: Ref Head (different-subject), Ref Mod = PoS (nominals)
Verbal categories: The different forms indicate different relative tense values: -j for present, -shka for
past, -na for future. Progressive aspect is retained. No subject agreement.
Nominal categories: CASE in Ref Head function; in Ref Mod function only when the relative clause is
extraposed.
Argument encoding: SENT / Ø - SENT / Ø (gapping in Ref Mod function) The object can remain
without accusative case (noun-stripping).
Examples:
Ref Head:
Marya nin-n [Juzi jatun wasi-ta chari-j]-ta
María say-3 José big house-acc have-nmlz:prs-acc
‘Maria says that José has a big house.’ (Cole 1982: 14)
ñuka-ka [Juan kay-pi ka-shka]-ta ya-ni

I-top Juan this-in be-nmlz:pst-acc think-I
‘I think that Juan was here.’ (Cole 1982: 33)
Juzi-ka [ñuka kaya llama-tar andi-na]-ta kri-n

José-top I tomorrow sheep-acc buy-nmlz:fut-acc believe-3
‘José believes that I will buy a sheep tomorrow.’ (Cole 1982: 37)
With noun-stripping:
Juzi-ka [ñuka kaya llama-Ø randi-na]-ta kri-n
José-top I tomorrow sheep buy-nmlz:fut-acc believe-3
‘José believes that I will buy a sheep tomorrow.’ (Cole 1982: 37)
Ref Mod:
[Marya riku-j] runa
María see-ptc:prs man
‘the man whom Maria sees’ (Cole 1982: 47)
[Marya riku-shka] runa

María see-ptc:prs man
‘the man whom Maria saw’ (Cole 1982: 47)
[Juzi kulki-ta kara-na] warmi

José silver-acc give-ptc:fut woman
‘the woman to whom José gave money’ (Cole 1982: 54)
Extraposed:
Kwitsa-ta juya-ni [Juan-wan tushu-shka ka-shka]-ta
girl-acc love-1 Juan-with dance-ptc:pst be-ptc:pst-acc
‘I love the girl who had danced with Juan.’ (Cole 1982: 51)
Pred-ngapaj (‘subjunctive’)
Remark: Subjunctive forms are used for the complements of manipulative and desiderative predicates;
-ngapaj is used for same subject.
Functional distribution: Rig: Ref Head, same subject ≠ PoS (nominals)
Verbal categories: No tense and subject agreement. Aspect can be retained.
Argument encoding: Ø - SENT (coreferentiality)
The object can remain without accusative case (noun-stripping).
Example:
Ref Head, same subject:
muna-y-man [ñuka mama-ta riku-ngapaj]
want-1-cond my mother-acc see-sbjv
‘I want to see my mother.’ (Cole 1982: 37)

Pred-chun (‘subjunctive’)
Remark: Subjunctive forms are used for the complements of manipulative and desiderative predicates;
-chun is used for different subject.
Functional distribution: Rig: Ref Head, different subject ≠ PoS (nominals)
Verbal categories: No tense and subject agreement. Aspect can be retained.
Example:
Ref Head:
muna-ni [Juzi pay-paj mama-ta riku-chun]
want-1 José he-poss mother-acc see-sbjv
‘I want that José sees his mother / I want José to see his mother.’ (Cole 1982: 37)
Pred-y
Remark: In Pred Mod function, the form is reduplicated.
Functional distribution: Flex: Ref Head, Pred Mod ≠ PoS (nominals, small manner adverbs)
Verbal categories: No tense / aspect / subject agreement
Nominal categories: CASE (in Ref Head function)
Argument encoding: Ø - SENT (coreferentiality) The object can remain without accusative case (noun-
stripping).
Examples:
Ref Head:
Juzi-ka [llama-ta / Ø randi-y]-ta usha-n
José-top sheep-acc / -Ø buy-inf-acc can-3
‘José is able to buy sheep.’ (Cole 1982: 40)
ñuka-ka [shuj ali wagra-ta-mi randi-y]-ta muna-ni

I-top one good cow-acc-val buy-inf-acc want-1sg
‘I want to buy a good cow.’ (Cole 1982: 40)
Pred Mod:
[Kanda-y kanda-y] shamu-rka-ni
sing-inf sing-inf come-pst-1
‘I came singing.’ (Cole 1982: 62)
Pred-shpa:
Functional distribution: Rig: Pred Mod = PoS: small manner adverbs
Verbal categories: No tense / aspect / subject agreement
Argument encoding: Ø - SENT (coreferentiality)
Example:
Pred Mod:
[Kanda-shpa-mi] shamu-rka-ni
sing-conv-val come-pst-1
‘I came singing.’ (Cole 1982: 62)
Ma'di
Pred-lɛ̄
Remarks: Apart from the subordinating suffixes lɛ́, kā, rɛ̄, ɓá, and dʒɔ́, Ma’di dependent predicates can only
take a low-tone prefix, which in independent clauses expresses non-past tense. It is not clear whether the
prefix on dependent predicates is the same, since it is compatible with any tense interpretation
(Blackings & Fabb 2003: 192).
In Ref Head function this construction is used for complements of desiderative predicates. The subject is
unexpressed under co-referentiality; the object is SENT.
In Ref Mod function this construction is used for object relative clauses. The object is either gapped, or
expressed with the postposition nā, meaning ‘aforementioned’ (AFR), and interpreted as the possession of
the modified noun. The subject is either left unexpressed, or expressed with the possessive postposition.
Functional distribution: Flex: Ref Head (desiderative), Ref Mod (object) = PoS (nominals)
Verbal categories: No tense
Argument encoding: Ø - SENT / POSS - Ø / POSS - OBL

Examples:
Ref Head:
Má lɛ̀-ā [èɓí ˋɲā-lɛ́] rá
1sg (n)want-obj fish n-eat-nmlz aff
‘I certainly want to eat fish.’ (Blackings & Fabb 2003: 202)
Ref Mod:
àràbɨ́à [ɔ́pɨ ́ Ɂà dƷɨ̄-lɛ́] rɨ ̀ pá nā ādī rá.
car Opi poss (n)-take-ptc def leg afr deflate aff
‘The car which Opi took certainly has a flat tyre.’ (Blackings & Fabb 2003: 22)
ílí [ŋɔ̀-lɛ́] rɨ ̀ lɔ́tʃɨ ̄

knife (n)-break-ptc def sharp
‘The knife which was broken is sharp.’ (Blackings & Fabb 2003: 201)
ágɔ́ [tí nā ɓārá nà Ɂà ōgù-lɛ́] rɨ ̀

man cow afr child that poss (n)-steal-ptc def
‘The man whose cow that child stole…’ (Blackings & Fabb 2003: 201)
Pred-dƷɔ́
Remarks: In Ref Head function this construction is used for the complements of phrasal predicates.
The subject remains unexpressed under co-referentiality. In Ref Mod function this construction is used for
relative clauses, in which the relativized element is a source. This relativized argument is gapped;
the subject is either unexpressed or possessive, and the object is SENT.
Functional distribution: F lex: Ref Head (phasals), Ref Mod (source clauses) = PoS (nominals)
Argument encoding: Ø - SENT / POSS - SENT
Examples:
Ref Head:
ɔ́pɨ ́ ɛ̄ɗɔ́ ˋsī-dƷɔ́ rá
Opi start [n-build-nmlz] aff
‘Opi has certainly started to build (with) it.’ (Blackings & Fabb 2003: 22 / 207)
Ref Mod:
bélè ágɔ́ rɨ ̀ pɨ̄ ˋmgbā-dƷɔ́ rɨ ̀ dɨ̀ Ɂɨ ̄
stick [man def plpron n-beat-ptc] def this foc
‘The stick with which the man and his associates was beaten is this one.’ (Blackings & Fabb 2003: 22)
Pred-ka
Functional distribution: Rig: Ref Head ≠ PoS (nominals)
Argument encoding: POSS - SENT / Ø - SENT
Examples:
Ref Mod:
má ɓì mū-kā kʊ̄rʊ̀
1sg try [(n)-go-nmlz] neg(pst)
‘I have not tried / did not try to go / going.’ (Blackings & Fabb 2003: 22)
Má ndrɛ̄ àrɨ́áŋgwá rɨ̀ Ɂà ɛ̄-ɗɛ́-kā rá

1sg see [bird def poss (n)-ve-fall-nmlz] aff
‘I saw the bird’s falling / fall.’ (Blackings & Fabb 2003: 21)
ɔ́-lɛ̀ tʊ́bà sɛ̄-kā ɗɨ Ɂ́ ā kʊ̄

ind-want [cigarette (n)-smoke-nmlz] here neg
‘Smoking is not permitted here.’ (Blackings & Fabb 2003: 213)
Pred-rɛ̄ / ɓá
Remark: ɓá is the plural equivalent of -rɛ̄ Only in non-active cases can an overt subject appear, which is
then marked with a postposition (see second example).
Functional distribution: Rig: Ref Mod (subject / possessive clauses) ≠ PoS (nominals)

Examples:
Ref Mod:
ágɔ́́ àm-à dƷɔ́ ̀ nɨ̄ ˋsī-ɓá rɨ̀
man [1pl-poss house spec pron n-build-ptc(pl)] def
‘one of the men who built / are building our house’ (Blackings & Fabb 2003: 193)
ɗɨ ágɔ́ ti nā ēgwè-rē dì rɨ ̀ Ɂɨ̄

this man [cow afr (n)-get.lost-ptc com] def foc
‘This is the man whose cow got lost.’ (Blackings & Fabb 2003: 195)
Unmarked clause
Remark: Used for complements of utterance, propositional attitude, knowledge, manipulatives and desideratives.
Example:
Ref Head:
ɲɨ́ ʄō k-ē-mú ɔ̀ɓʊ́
2sg say [dir-ve-go tomorrow]
‘You said that she should come tomorrow.’/‘You told her to come tomorrow.’ (Blackings & Fabb 2003: 21)
Clause with zɨ ̄ + sɨ̀

Remark: This construction is marked by the grammatical verb zɨ̄ and the source propositiona sɨ̀. It
expresses simultaneous action.
The subject is co-referential but overtly expressed (SENT). (Blackings & Fabb 2003: 421).
Functional distribution: Rig: Pred Mod = PoS (small and derived manner adverbs)
Nominal categories: CASE (postposition)
Example:
Pred Mod:
ká sɨ́tà ɲā míndrā nā kóˋ-zɨ ̄ rā sɨ̀
3 pepper (n)eat [tears afr 3-n-sim leak] sr
‘She was eating pepper as her eyes were (continuing) running.’
(Blackings & Fabb 2003: 440)
Hungarian
Pred-ni
Remarks: When this construction has subject function, agreement (in person and number) with the subject
is optionally retained. In object function, subject agreement is always lost. Only one argument can be
overtly expressed. If the subject is overt, it takes DAT; if the object is overt, it is SENT.
Verbal categories: No tense / mood. Causative and frequentative affixes can be expressed. Agreement
is sometimes retained (see above).
Argument encoding: Ø - SENT / DAT - Ø
Examples:
Ref Head:
Fontos volt [Péter-nek olvas-ni(-a)]
important was Peter-dat read-inf(-3sg)
‘It was important for Peter to read.’ (Kenesei et al. 1998: 35)
Anna most akar [olvas-ni]
Anna now wants read-inf
‘Anna wants to read now.’ (Kenesei et al. 1998: 33)
Anna meg-próbál-t-a [meg-tanul-ni a verset]

Anna pref-try-pst-def.3sg pref-learn-inf the poem.acc
‘Anna tried to learn the poem.’ (Kenesei et al. 1998: 33)

Pred-ás / -és
Remark: The subject of a transitive nominalized verb is optionally expressed as an attributive adjectival
phrase involving the adjectival általi form of the postposition által ‘by’, or involving the postposition által
and the active participial form való of van ‘to be’. The subject of an intransitive predicate and the object of
a transitive predicate trigger nominal agreement on the dependent predicate.
Verbal categories: No tense / aspect / mood / agreement
Argument encoding: POSS / OBL - POSS
Examples:
Ref Head:
János [a kincs el-rejt-és-é]-t javasol-t-a
John the treasure pref-hide-nmlz-poss.3sg-acc suggest-pst-def.3sg
‘John suggested to hide the treasure.’ (Kenesei et al. 1998: 207)
Örül-ök [Pál váratlan meg-érkez-és-é]-nek

be.happy-indef.1sg Paul unexpected pref-arrive-nmlz-poss.3sg-dat
‘I am happy about Paul’s unexpected arrival.’ (Kenesei et al. 1998: 207)
Pred-ó (‘active present participle’):

Functional distribution: Rig: Ref Mod (subject clauses) ≠ PoS (nominals)
Example:
Ref Mod:
[A könyv-et a fiú-nak gyorsan olvas-ó] lány itt van
the book-acc the boy-dat fast read-prs.ptc girl here is
‘Here is the girl who reads the book to the boy fast.’ (Kenesei et al. 1998: 45)
Pred-ótt (‘passive, past participle’)

Remark: Used in prenominal relative clauses where the relativized item is the undergoer / patient of the
DC. This element is gapped; the agent is marked OBL with the agentive postposition által ‘by’.
Functional distribution: Rig: Ref Mod (patient clauses) ≠ PoS (nominals)
Argument encoding: OBL - Ø
Example:
Ref Mod:
Az [Anna által tegnap olvas-ott] könyv
the Anna by yesterday read-pst.ptc book
‘the book (that was) read by Anna yesterday.’ (Kenesei et al. 1998: 46)
Pred-andó / -endő (‘future participle’)

Functional distribution: Rig: Ref Mod (subject and object clauses) ≠ PoS (nominals)
Argument encoding: Ø-SENT / SENT-Ø / Ø-Ø
Examples:
Ref Mod:
az [el-jöv-endő] kor
the pref-come-fut.ptc age
‘the age to come’ (Kenesei et al. 1998: 319)
a [ki-javít-andó] dolgozat-ok
the pref-correct-fut.ptc paper-pl
‘the papers to be corrected’ (Kenesei et al. 1998: 320)
Pred-vá / -vé (‘simple converb’)

Functional distribution: Rig: Pred Mod = PoS ((derived) manner adverbs)

Example:
Pred Mod:
A gyerek-wk [kiabál-va] szalad-t-ak végig az utcá-n.
the child-pl shout-conv run-pst-indef.3pl along the street-superess
‘The children ran down the street shouting.’ (Kenesei et al. 1998: 320)
Pred-vén (‘perfective converb’):

Remark: This form is very infrequent in spoken Hungarian; it is used only in formal and ceremonious style
in writing. The perfective converb has historically been used to refer to an action preceding that of the
finite verb. Nowadays it is sometimes used synonymously with the simple converb.
Example:
Pred Mod:
Ez-t mond-t-a nek-em [az asztalfő-n ül-vén].
this-acc say-pst-def.3sg dat-1sg the table.head-superess sit-pfv.conv
‘Sitting at the head of the table s / he said this to me.’ (Kenesei et al. 1998: 321)
Clause + hogy
Remarks: The construction can be combined with the expletive pronominal az, which takes case according
to the function of the DC. In nominative and accusative function case can be omitted; in other functions
it cannot. When functioning as the complement of a maniopulative or evaluative predicate, the dependent
predicate takes the subjunctive marker -j- and a prefix meg-. In the case of a manipulative predicate the
complementizer can be omitted; in the case of an evaluative predicate it cannot.
Verbal categories: All retained (SBJV)
Nominal categories: (CASE)
Examples:
Ref Head:
Anna elmondta nekünk (azt), [hogy Péter beteg volt].
Anna told.def to.us it.acc comp Peter sick was
‘Anna told us that Peter had been sick.’ (Kenesei et al. 1998: 31)
Anna azt mondta, [(hogy) tanul-ja-d meg a verset]

Anna it.acc said.def comp learn-imp / sbjv-def.2sg pref the poem.acc
‘Anna told you to learn the poem.’ (Kenesei et al. 1998: 32)
Nem szükséges, [*(hogy) Péter meg-tanul-j-a a verset]

Not necessary comp Peter prefix-learn-sbjv-def the poem-acc
‘It is not necessary for Peter to learn the poem.’ (Kenesei et al. 1998: 32)
Clause úgy + hogy

Remark: úgy is the adverbial form of the pronominal az in the main clause (most often ‘thus’).
Optionally a relative pro-adverb is attached to the subordinator hogy.
Examples:
Pred Mod:
Péter úgy akludt el, [hogy olvasott]
Peter adv.pron slept pref sub read.3sg
‘Peter fell asleep in such a manner that he was reading.’ 2 ‘Peter fell asleep reading.’ (Kenesei et al. 1998: 50)

Péter úgy aludt, [a-hogy gyerekkorában szokott]
Peter adv.pron slept rel.pron-sub in.his.childhood used.3sg
‘Peter fell asleep the way he used to in his childhood.’ (Kenesei et al. 1998: 50)
REL.PRON + clause
Remarks: The relative pronoun takes case-marking according to the function of the head noun in the
dependent clause. Optionally, the construction can be combined with a clause-initial demonstrative along
with the lexical head noun.
Functional distribution: Rig: Ref Mod ≠ PoS (nominals)
Argument encoding: SENT - SENT / Ø - SENT / SENT - Ø
Examples:
Ref Mod:
(Az) a könyv, [amely-et Anna olvas-ott,] érdekes volt
dem the book which-acc Anna read-pst interesting was
‘The book that Anna was reading was interesting.’ (Kenesei et al. 1998: 38)
Itt van a fiú, [aki-nek Anna felolvasta a könyvet]

here is the boy who-dat Anna pref-read-def the book.acc
‘Here’s the boy to whom Anna read the book.’ (Kenesei et al. 1998: 42)
Japanese
Clause + no, mono, koto, wake, yoo
Remarks: The ‘nominalizers’ are in fact case-marked nouns meaning ‘thing’. The dependent predicate
is tensed, but the subject can optionally appear in the genitive. Therefore this construction has a double
classification 1 / 3.
Structural type: 1 (Balanced) / 3 (D-ALT)
Verbal categories: Tense retained
Argument encoding: POSS-SENT / SENT- SENT
Example:
Ref Head:
[Ano hito ga / no hon o kai-ta koto ga] yoku sira-re-te iru
that person nom / gen book acc write-past thing(nmlz) nom well know-pass-ger be
‘It is well known that that person wrote a book.’ (the fact that that person wrote a book is well-known.’/
the fact of that person’s having written a book is well known.’) (Lombardi Vallauri 1997: 497)
Unmarked REL clause

Remark: There is no relative marker, but resumptive pronouns (in the form of demonstrative, personal,
or reflexive pronouns) can be optionally used.
Functional distribution: Rig: Ref Mod ≠ PoS (nominals)
Verbal categories: Tense retained
Argument encoding: SENT- SENT / Ø - SENT / SENT- Ø (gapping)
Examples:
Ref Mod:
Dakara, [tamago motte-ru] hito mo ita shi,
so eggs hold-pst person too were and
[juusu mo motte-ru] hito mo ita shi…
juice too hold-pst person too was and
‘So, there were people holding eggs, and people holding juice, and…’ (Hinds 1986: 59)
With resumptive pronoun:

[Sono mae ni kuruma ga tomatte-iru] mise
dem front dir car nom stopped shop
‘the shop in front of which a car is parked’ (Hinds 1986: 61)
Pred-te / -de / -ite

Remark: Overt subjects can take the topic marker in stead of the nominative.
Functional distribution: Rig: Ref Mod = PoS (manner adverbs)
Verbal categories: No Tense

Argument encoding: SENT - SENT / Ø - SENT (co-referentiality)
Examples:
Pred Mod (see for more instances the next example):
Son san wa [hoka no nihonjin sutaffu o kun-de] shigoto o shi-te i-ru
Song Mr. top other attr Japanese staff obj unite-conv job obj do-conv be-prs
‘Mr Song is working together (in a united manner) with other Japanese staff.’
(Alpatov & Podlesskaya 1995: 469)
Different subject:
Yasuko wa juugoroku no koro ibiki no kuse ga
Yasuko top fifteen:sixteen attr time snore attr habit subj
at-te, oya wa kyoosei ni kushin shi-ta-soo da
be-conv parents top correction i.obj efforts do-pst-likely cop.prs
‘They say that Yasuko snored when [she was] fifteen or sixteen [and her] parents did their best to get rid
[of this habit]. (Alpatov & Podlesskaya 1995: 469)
Pred-i / -Ø (‘infinitive’)
Remark: Overt subjects can take the topic marker in stead of the nominative.
Functional distribution: Rig: Ref Mod = PoS (manner adverbs)
Verbal categories: No Tense
Argument encoding: SENT - SENT / Ø - SENT (co-referentiality)
Pred Mod:
Same subject:
Ogata Shingo wa [sukoshi mayu o yose-Ø] [sukoshi kuchi o
Ogatta Shingo top slightly eyebrow obj moved.together-inf slightly mouth obj
ake-te] [nanila kangae-te] i-ru fuu datta
open-conv something think-conv be-pres look cop.pst
‘Ogata Shingo looked as if he was thinking (about) something, bringing his eyebrows slightly together
and slightly opening his mouth.’ (Alpatov & Podlesskaya 1995: 468)
Different subject:
Shingo wa kao o shikame-Ø, Shuuichi wa yoi ga same-ta daroo
Shingo top face d.obj frown-inf Shuuichi top drunkenness subj abate-pst tent
‘Shingo frowned, [and] it seemed that Shuuichi got sober.’ (Alpatov & Podlesskaya 1995: 468)
Hmong Njua
qhov + clause
Remarks: This construction can be a complement clause or a nominalization, depending on the scope of
its structural coding, as shown by the position of the marker, which may either precede the dependent
predicate, or the whole DC. Since there is no difference in terms of verbal / nominal categories and / or
argument encoding, this construction is classified as a balanced one only.
This construction is used for subject clauses.
Verbal categories: Not applicable
Example:
Ref Head:
Qhov cov zej zog tsiv tawm ua kuv zoo sab.
comp cl neighbour move out do 1sg happy
‘It makes me happy that the neighbour moves out.’ (Harriehausen 1990: 200)
kuam / (has)tas + clause

Remarks: The choice between the two forms depends on the type of matrix predicate. Kuam is used with
complements of desideratives and manipulatives; (has)tas with complements of perception, knowledge,
propositional attitude, and utterance. Both forms are used for object complements only.

Examples:
Ref Head:
Peter xaav [kuam Paul yuav lub fais]
Peter want comp Paul buy cl car
‘Peter wants Paul to buy a car.’ (Harriehausen 1990: 220)
Kuv paub [(has)tas nwg yuav npaaj]

1sg know comp 3sg buy flowers
‘I know that he has bought flowers.’ (Harriehausen 1990: 22)
Unmarked clause
Example:
Ref Head:
Kuv xaav [kuv moog tsev sai]
1sg hope 1sg go house soon
‘I hope to go home soon.’ (Harriehausen 1990: 221)
kws (REL) + clause:

Example:
Ref Mod:
Tug txiv neej [kws kuv saib sab]
cl man rel 1sg see big
‘The man that I saw was tall.’ (Harriehausen 1990:141)
Lango
Pred-(kk)ɔ̀ (‘infinitive / nominalization’)
Remarks: Overt subjects are marked as attributive modifiers.
The construction is used for complements of phasal, modal, desiderative, and achievement predicates.
Verbal categories: No aspect; no subject agreement. Object agreement, voice and
(benefactive / ventive) valency retained.
Argument encoding: Ø - SENT / POSS - SENT
Example:
Ref Head:
ámɪ̀ttò [gwɛ̀ɛ̀yòò dìlò]
1sg.want.progr kick.inf ball
‘I want to kick the ball.’ (Noonan 1992: 213)
àpwóɲ òtɪ̀tòwá à cècèk [bɛ̀ɛ̀r à kwân]

teacher 3sg.tell.pfv.1pl attr.prt little.bit good.nmlz attr.prt read-inf
‘The teacher told us briefly about the benefits of reading.’ (The teacher told us briefly the being good of
reading.) (Noonan 1992: 213)
[Nè-kkòg wôkk à dákô] ràc

kill-nmlz dog attr.prt woman 3sg-bad-hab
‘The killing of a dog by a woman is bad.’ (Noonan 1992: 214)
nî + clause
Remark: The dependent predicate is indicative when there is independent time reference and subjunctive
when there is dependent time reference.
Verbal categories: Retained (see remark)

Example:
Ref Head:
Àŋéô [nî lócǝ̀ dágô tìc]
1sg-know-hab comp man 3sg-hate-hab work
‘I know that the man hates work.’ (Noonan 1992: 191)
ámɪ̀ttò [nî dákô òwɪ̆llɪ̀ àtîn búk]

1sg-want-progr comp woman 3sg-buy-ben-sbjv child book
‘I want the woman to buy the child a book.’ (Noonan 1992: 191)
àmɛ̂ + clause
Remark: The marker àmɛ̂ is a combination of the attributive particle à and the relative particle mɛ̂. The
relativized item is gapped. A resumptive pronoun must be used when the relativized item is a benefactive,
associative, or object of preposition.
Functional distribution: Rig: Ref Mod ≠ PoS (modifiers)
Verbal categories: Retained (see remark)
Argument encoding: SENT - SENT / SENT - Ø / Ø - SENT (gapping)
Examples:
Ref Mod:
lócǝ̀ [àmɛ̂ márô gwôk]
man rel+attr.prt 3sg-like-hab dog
‘The man that likes the dog.’ (Noonan 1992: 215)
âtɪ̂nn [àmɛ̂ dákô òwɪ̀llɛ́ búk]

child rel+att.prt woman 3sg-buy-ben-pfv-3sg book
‘The child for whom the woman bought a book.’ (Noonan 1992: 215)
Ket
Bare infinitive
Remarks: The bare form is used in Ref Head function for the complements of phasal and modal (ability)
predicates. The object, if expressed, is incorporated. For the complement of desiderative or modal
(necessity) predicates, the infinitive is marked for translative case.
The bare infinitive is also used in Ref Mod function, but this may involve lexical derivation, since the
infinitive apparently cannot take any arguments.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, modifiers, small / derived adjectives)
Verbal categories: None (i.e. no tense / mood, no subject-object agreement)
Argument encoding: Ø - INC / POSS - INC
Examples:
Ref Head:
[ə́k-nà qús-bèt] bínùt
pl-anim.pl.gen tent-making.inf it.ended
‘We finished making the tent.’ (Lit: ‘Our tent-making ended.’) (Vajda 2004: 78).
[āt iɁl] ítpàɾam

1sg sing.inf I.know
‘I know how to sing / I can sing.’ (Vajda 2004: 78)
[Áb-àŋa ássȁno-esaŋ] nárá

1sg-dat hunt.inf-trl need
‘I need to hunt’ (Vajda 2004: 77)
Ref Mod:
bʌ̀ ɣ ságdì
find boot
‘A boot that is found’ (Vajda 2004: 79)
Unmarked clause
Remark: This construction is apparently infrequent. Werner (1997: 355): “It happens in Ket that a finite
verb stands before a noun in attributive position.” (emphasis added, EvL)
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, modifiers, small / derived adjectives)

Examples:
Ref Head:
dótàm-báàm íɣȍvilde [bísnìmin bə́jgìtnaŋtonoq]
dotam-old.woman she.heard brother.pl orphans.they.became
‘Old Dotam Woman (a forest witch) heard that the brothers had become orphans.’ (Vajda 2004: 93)
Ref Mod:
Tu˙r; [ital’em] kɛɁt
dem he.has.knowledge person
‘This knowledgeable person’ (lit: this person who has knowledge) (Werner 1997: 355)
Clause + ɛta qɔr’a

Functional distribution: Rig: Pred Mod = PoS (small / derived adjectives) ≠ PoS (modifiers)
Example:
Pred Mod
Tajɔbɔn, [ɛta qɔr’a bɛr’ɛta].
It has become cold, as if it cuts
‘It has become piercingly cold.’ (Werner 1997: 348)
Clause + ásqà
Example:
Pred Mod
bū toɁn dóldàq bɨ́ldȅ [deɁŋ dólìn ásqà].
3masc so he.lived all people they.lived like
‘He lived like everyone (else) lived.’ (Vajda 2004: 87)
Clause + PROS(ecutive) CASE

Example:
Pred Mod:
ū-k hɨɁp [dúrèn-bes] tséstè
2sg.-gen son he.cries-pros he.sits
‘Your son sits (there) crying.’ (Vajda 2004: 28 /Werner 1997: 354)
REL PRON + clause

Nominal categories: Class / number agreement
Examples:
Ref Mod:
āt qɨ̄m díɣaɾo [qó-ɾè sóòŋ də́ɣàɾaq]
1sg woman I.see.her who-fem there she.lives
‘I am looking at the woman who lives there.’ (Vadja 2004: 30)

bɨ́ldè̀ deɁŋ [qó-nè sóòŋ dólìn dúnòn]
all people who-anim.pl here they.lived they.died
‘All the people who lived here have died.’ (Vadja 2004: 30)
Pred-s / -bes + clause

Remark: The -s suffix, and its allomorph -bes (used when the object of the dependent clause stands
between the dependent predicate and the head noun) are called ‘nominalizers’ (and glossed as such),
but they appear on finite predicates, and arguments remain SENT.
Nominal categories: Class / number agreement
Examples:
Ref Mod:
[mámùl dóblà-s] dɨ̄l
milk he.drink.it-nmlz child
‘a child who drinks milk’ (Vajda 2004: 79)
[dóblà-bes mámùl] dɨ̄l

he.drink.it-nmlz milk child
‘a child who drinks milk’ (Vajda 2004: 79)
Itelmen
Infinitives (various forms)
Verbal categories: No agreement
Examples:
Pred-s (INF I) (Used for complements of modal, phasal and manipulative predicates.)
Ref Head:
Utu-z-en [əŋqa bol̦se l̦owal̦e-s.]
can.not-prs-3.sg something more say-inf.i
‘He cannot say anything anymore.’ (Georg & Volodin 1999: 168)
Kəmma t-iti-çen [p’e-çh no-ka-s.]

1.sg 1.sg-force-3.sg.pat child-dim eat-ccm-inf.i
‘I forced the child to eat.’ (Georg & Volodin 1999: 169)
Pred-kilh-kalh / -kilaʔn / -kalaʔn (PL) (INF II) (Used for complements of direct perception predicates.)
Ref Head:
Truk Sin ’aŋewt k-el̦çku-ʔin [̊t ’it’im es-kal̦h.]
suddenly S. inf.iii-see-inf.iii smoke come.out-inf.ii
‘Suddenly S. saw that smoke was coming out.’ (Georg & Volodin 1999: 173)
Pred-ki / ka (INF V): (Used for complements of phasal and modal predicates and some manipulative
predicates. Applies only to those verbs that take -k in Infinitive I.)
Ref Head:
A Sin’aŋewte-n nita əŋqan-kit k’-uzu-knen [əŋqsx-ki.]
interj s.-poss soul was-caus inf.iii-start-inf.iii hurt-inf.v
For some reason S. became sad. (lit.: S’s soul began to hurt.’) (Georg & Volodin 1999: 180)
Pred-l (INF VI) (Same function as Infinitive V, but it is used with those verbs that do NOT take k- in
Infinitive I.)
Ref Head:
T-utu-s-çen [Ememqut met’ele-l̦.]
1.sg-can.not-prs-3.sg.pat e. kill-inf.iv
‘I cannot kill E.’ (Georg & Volodin 1999: 182)
Unmarked clause
Remark: Used for subject clauses and sometimes for object complements of perception predicates.

Argument encoding: SENT - SENT (co-referentiality)
Examples:
Ref Head:
Qetew [çki-çiŋnen qéfq’ef].
good find-3.sg:3.sg strength
‘It is good that he has found the strength.’ (Georg & Volodin 1999: 204)
Utre n-esxl̦i-k t-laxl̦-kiçen [ən-k’esxl̦-kiçen muza-ʔn].

in.the.morning 1.pl-wake.up-1.pl 1.sg-see-1.sg 1.pl-be.dry-1.pl 1.pl-pl
‘In the morning we woke up, and I saw that we were dry.’ (Georg & Volodin 1999: 205)
min + gapped clause

Functional distribution: Rig: Ref Mod = PoS (derived adjectives)
Example:
Ref Mod:
T-çki-kiçen nu Nwilwejŋen [min k-çil-knen kulaka-ʔnk].
1sg-find-1sg dem N. rel inf.iii-choose-inf.iii Kulak-loc
‘I found this Nwilwejngen, whom the Kulaken had chosen.’ (Georg & Volodin 1999: 203)
qatz + clause
Functional distribution: Rig: Pred Mod = PoS (derived manner adverbs)
Example:
Pred Mod:
Ewun fçora celoj ql̦hɘl [qatz k-niŋ-ʔin] t-ļase-s-k.
thus yesterday whole day as.if inf.iii-loaded-inf.iii 1.sg-go-prs-1.sg
‘Thus I went around all day, heavily loaded.’ (Georg & Volodin 1999: 213)
Koasati
‘Nominalizations’ (various forms)
Remarks: For all verb classes, except one (‘class 2A’), the nominalization is formed form the 1st person plural
affirmative, without phrase-terminal markers. Other nominalizations are formed by replacing the element -li
with the element -ka. (Kimball 1991: 273-274). Nominalization cannot take any other verbal morphology
than categories related to voice / valency (reciprocal, reflexive, locative and instrumental prefixes).
Verbal categories: No TAM / phrase-terminal marker; only voice / valency can be retained.
Examples:
Ref Head:
[Icó sammí:ci-t inɬah-ó:li] sobáyli-l
deer do:how-conn to:shoot&hit-nmlz know-1sg.sbj
‘I know how to shoot deer.’ (Kimball 1991: 280)
[Aybacilká a;wáh-ka] ca-bán-ko-V

law ruin-nmlz 1sg.sbj.stat-want-neg-phtm
‘I don’t want to break the law.’ (Kimball 1991: 275)
Participles (various forms)

Remark: Participles can express subject/object cross-reference, but apparently with special forms. (Kimball
1991: 288): “It seems likely that the participle suffixes are added to an already nominalized verb. This
supposition is strengthened by the fact that all the participial suffixes (with the exception of the future
participle, which is transparently derived from the present participle) also appear as article suffixes on nouns.”

Functional distribution: Rig: Ref Mod = PoS (small adjectives)
Verbal categories: No TAM, no phrase-terminal marker; only voice / valency can be retained.
Nominal categories: CASE agreement (ACC is zero-marked)
Argument encoding: Ø - SENT / SENT - Ø (gapping)
Pred-:sáya (‘Present participle’) (Requires the focus form of the subject and object markers.)
Ref Mod:
Yilahá [í:pa-li-:sáy-ok] ká:n-á:ho:si-:s
orange eat-1ss-prs.ptc-ss.foc be.good-adv-i.pst
‘The orange that I am eating is very good.’ (Kimball 1991: 289)
Yilahá [am-híska-:sáy-on] í:pa-li-t
orange give.to.me-2sg.sbj-prs.ptc-obj:foc eat-1sg.sbj-pst
‘I ate the orange that you just gave me.’ (Kimball 1991: 289)
Pred-:yólli (Habitual participle)

Remark: Subjects of these participles are usually stripped of their case marking. The focus forms are used
for subject / object marking.
Ref Mod:
Akkó á:tí [í:pa-:yóll-ok] nì:há:ho:s
that person eat-hab-ptc-ss.foc be.fat-adv
‘A person who eats all the time is very fat.’ (Kimball 1991: 290)
Pred-:ka (‘Past participle’)

Remark: Focus forms are used for subject / object marking.
Ref mod:
Yilahá [nihtá-:k-on ám-híska-:k-on] í:pa-lit
orange day-art-obj.foc give.to.me-2sg.sbj-pst.ptc-obj.foc eat-1sg.sbj-pst
‘I ate the orange that you gave me yesterday.’ (Kimball 1991: 291)
Pred-:kítta (‘Imperfective participle’)
Ref Mod:
Á:ti [hí:ca-li-:kítta-p] hí:ca-li-t
person see-1sg.sbj-ipfv.ptc-top see-1sg.sbj-pst
‘I saw the person that I used to see.’ (Kimball 1991: 292)
Pred-laho:li:sáya (‘Future participle’)

Remark: This formation is a combination of the present participial suffix -:sáya with the verbal suffixes
-laho- (IRR), and -:li-, a hearsay suffix with the meaning ‘deduction’ (Kimball 1991: 292):
Ref Mod:
Yilahá [ám-híska-laho:li:sáy-on] í:pa-l-laha-V
orange give.to.me-2sg.sbj-fut.ptc-obj.foc eat-1sg.sbj-irr-phtm.
‘I intend to eat the orange that you will give me.’ (Kimball 1991: 292)
Pred (-o:si / -s:si)-n

Remark: -n is the different-subject switch-reference (SW) marker. It can be combined with the
diminutive / intensive suffix -o:si / -s:si.
Functional distribution: Rig: Pred Mod = PoS (small / derived manner adverbs)
Verbal categories: No TAM, no phrase-terminal marker
Argument encoding: SENT - SENT / Ø - SENT
Examples:
Pred Mod:
[iyyí-k atákka-n] wáyka-V́hci
foot-sbj hang(pl)-sw fly(sg)-progr
‘It flies with its legs hanging down.’ Lit.: ‘Its legs hang down and it flies.’ (Kimball 1991: 488)
[Talásba-:si-n] kóyli-li-t
be.thin-dim-sw cut(pl)-1sg.sbj-pst
‘I cut it up thinly.’ (‘as thinly as possible’) (Kimball 1991: 488)
Pred-t
Remark: The suffix is a connector. This construction is used when the verb modifier can be considered as

an action that takes place at the same time as the action of the matrix verb (= simultaneity clause).
Example:
Pred Mod:
[Fololohká:ci-t] cokkío:li-n hí:ca-li-:p im-ca-maɬátli-V́hco-k
be.coiled.up-conn sit(sg)-sw see-1ss-sbjv 3stat.obj-1sbj.stat-be.afraid(sg)-hab-ss
‘If I see one sitting coiled up, I am afraid of it.’ (Kimball 1991: 489)
Pred-k
Remark: The suffix is a same-subject marker
This construction is used when the adverbial action can be applied as much to the subject of the sentence
as to the verb (i.e. secondary predication).
Example:
Pred Mod:
[Wayóhka-k] ho-pálki-palámmi-n
fly(pl)-ss distr-be.fast-adv-sw
‘They all fly very fast.’ (Kimball 1991: 490)
Thai
Unmarked clause
Remarks: This construction is used for subject clauses and for object complements of desiderative and
achievement predicates. Under special conditions it also occurs in Ref Mod function, namely expressing a subject
relative clause that gives a general description of the head noun. In such cases, the relativized item is gapped.
Functional distribution: Flex: Ref Head, Ref Mod≠ PoS (nouns, adjectives, small modifiers)
Argument encoding: SENT - SENT / Ø - SENT (gapping)
Examples:
Ref Head:
[ɔ̀ɔk-kamlaŋ thúk wan] dii tɔ̀ɔ râaŋ-kaay
exercise every day good towards body
‘Doing exercise every day is good for your body.’ (Iwasaki & Ingkaphirom 2005: 253)
Kháw yàak [cɔ̀ hǔu khà]

3 want pierce ear prt
‘She wants to have her ears pierced.’ (Iwasaki & Ingkaphirom 2005: 231)
Ref Mod:
èk pen dèk [rian kèŋ]
Ek cop child study well
‘Ek is a child who studies well.’ (Iwasaki & Ingkaphirom 2005: 250)
thîi + clause:
Remark: This construction is used in Ref Head function, as the complement of predicates expressing
evaluation and emotion, and sometimes (optionally) of desiderative predicates. In Ref Mod function it is
used in combination with a gapping strategy or a resumptive pronoun.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, adjectives, small modifiers)
Argument encoding: SENT- SENT / Ø- SENT / SENT - Ø (gapping)

Examples:
Ref Head:
tɛ̀-wâa dii ná [thîi mây mii khay pen alay]
but good prt comp neg have who cop what
‘But it was good that no one was hurt.’ (Iwasaki & Ingkaphirom 2005: 255)
Ref Mod:
Khon [thîi duulɛɛ] nî pen pen acaan lə̌
person rel take.care prt cop cop teacher q
‘Is the person who takes care [of the students] a teacher?’ (Iwasaki & Ingkaphirom 2005: 243)
Khon [thîi kháw pay yùu kan taam roŋrian]

people rel rsp3 go stay rcp school
‘people who want to stay at school’ (Iwasaki & Ingkaphirom 2005: 245)
(thîi+) wâa / cà + clause

Remark: In some cases this construction combines with the flexible subordinator thîi, while in other cases
this element is optional and only wâa (‘say’) and / or the ‘challengeable marker’ (CM) cà remain.
Example:
Ref Head:
khít [wâa cà hǎa ŋaan tham thîi-nîI]
think say / comp cm look.for work do here
‘I think that I will look for work here.’ (Iwasaki & Preeya Ingkaphirom 2005: 262)
Basque
Pred-t(z)e
Verbal categories: No Tense, mood and agreement
Nominal categories: DET
Examples:
Ref Head:
Damu dut [zuri gezurra esa-te]-a
regret have you.dat lie say-nmlz-det
‘I regret telling you a lie.’ (Hualde & Ortiz de Urbina 2003: 656)
[haurrek etxean liburuak sari irakur-tze]-a

children.erg home.loc books often read-nmlz-det
‘children’s often reading books at home’ (Hualde & Ortiz de Urbina 2003: 666)
Pred-tu / du / -i / -Ø (‘perfective participle’)

Remark: this is the perfective counterpart of the nominalization with -tze (see above). In combination with
instrumental or partitive case 10, or with a postposition such as gabe ‘without’, this construction can also be
used in Pred Mod function.
Verbal categories: No Tense, mood and agreement
Examples:
Ref Head:
Damu dut [zu irain-du]-a
regret aux you offend-pfv.nmlz-det
‘I regret having offended you.’ (Hualde & Ortiz de Urbina 2003: 668)
10
In Eastern dialects, in stead of –rik, the morpheme –ta is used, which is probably related to the conjunction eta and as such
seems to form a dedicated adverbial construction in combination with the participial form (see Hualde & Ortiz de Urbina
2003: 745-746). In the classification of Chapter 6, this is not taken into account as a separate coding strategy.

[haurrek etxean liburuak sari irakur]-i-a
children.erg home.loc books often read-pfv.nmlz-det
‘children’s having often read books at home’ (Hualde & Ortiz de Urbina 2003: 666)
[Bilbon etxe bat eros-i] behar / nahi dut

Bilboa.loc house one buy-pfv.nmlz need / want aux
‘I need / want to buy a house in Bilboa.’ (Hualde & Ortiz de Urbina 2003: 694)
Pred Mod:
lan-a egi-n-az irabaz-ten d-a
work-sg.abs do-pfv.ptc.sg-instr earn-hab aux(3abs-pres)
diru-a ez alferkeria-n ego-n-az
money-sg.abs neg sloth-sg.loc be-pfv.ptc.sg.instr
‘Money is earned by working, not by being lazy.’ (Saltarelli 1988: 55)
Jon uneotan Ameriketan egon-ik, ezer gutxi egin dezagu

Jon moments.prox.loc Americas.loc be.ptc-part anything little do aux.pot
lana azurreratzeko
job advane
‘With Jon being in America right now, there’s very little we can do to make progress with the work.’
(Hualde & Ortiz de Urbina 2003: 746)
Buru-a buel-ta gabe segi z-u-en aurre-ra

head-sg.abs turn-pfv.ptc without follow 3sg.erg-aux2-pst front-sg.all
‘He continued forward without turning his head.’ (Saltarelli 1988: 55)
(subjunctive) clause -(e)n

Remarks: This construction is used for subjunctive complements of desiderative, manipulative, and emotive
predicates. Although -en is the most common complementizer in such cases, the complementizer -ela
seems to be more frequent in western dialects. In the western dialects -en-a (complementizer + determiner
-a) is also used to mark factive complements. The conjunction -enik is the counterpart of -(e)n used for the
factive complement of a negated matrix verb (Hualde & Ortiz de Urbina 2003: 646).
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, adjectives, small modifiers)
Nominal categories: (DET / CASE)
Examples:
Ref Head:
[Euskara Euskal Herri osoan ofiziala izan Dad]-in nahi dute
Basque Basque Country entire.loc official be aux-(sbjv).comp want aux
euskladun askok.
Basque.speaker many.erg
‘Many Basque speakers want that the Basque language be official in the entire Basque Country.’
(Hualde & Ortiz de Urbina 2003: 640).
Entzun dut [Amaiaren neba hil d]-en-a

‘I heard that Amaia’s brother died.’ (Hualde & Ortiz de Urbina 2003: 646)
Banekien [Mikel berandu etoriko z]-en-a

ba-knew Mikel late arrive.fut aux-comp-det
‘I knew that Mikel would arrive late.’ (Hualde & Ortiz de Urbina 2003: 646)
Bi udaltzainek ukatu dute [bidaian Rubioren bizkartzain izan zir]-enik

two policemen.erg deny aux trip.loc Rubio.gen bodyguard be were-comp
‘Two police officers have denied that they had been Rubio’s bodyguards during the trip.’
Ref Mod:
[Ez dakizkida]-n gaiez mintzatu nahi nuen
Not know-3abs.pl / 1erg-rel matters.instr speak will aux
‘I wanted to speak about matters that I don’t know.’ (Hualde & Ortiz de Urbina 2003: 763)
[Pellok ekarri du]-en dirua galdu dut

‘I lost the money that Peter brought.’ (Hualde & Ortiz de Urbina 2003:764)

Clause + -en bezala
Example:
Pred Mod:
[Zeuk esan didaz]-un bezala egin dut lana
you.emph say aux.adv do aux job
‘I did my job the way you told me.’ (Hualde & Ortiz de Urbina 2003: 722)
(subjunctive) clause -(e)la

Remarks: With desiderative and manipulative main predicates, the auxiliary in the dependent clause takes
subjunctive form. -(e)la can be used in Pred Mod function for adverbial manner clauses, either by itself
or combination with the partitive case -rik. In its basic form -ela has a modal / temporal (simultaneous)
meaning 11. The combination with -rik is a dialectal variant, separating the Bizkaian and Gipuzkoan area
from the eastern dialects.
Functional distribution: Flex: Ref Head, Pred Mod (+ partitive case) ≠ PoS (nouns, derived manner adverbs)
Examples:
Ref Head:
batzuek uste dute [hauk oro kazeten eta kazeta-egileen egitekoak dir]-ela
some.erg think aux these all journals.gen and journal-makers.gen duties are-comp
‘Some think that all these are duties of journals and journalists.’ (Hualde & Ortiz de Urbina 2003: 635)
Unibertsitateak [agiri guztiak euskaraz eta gaztelaniaz egin daitz]-ela

University.erg document all.det.pl Basque.instr and Spanish.instr do aux(subj)-comp
‘The University has demanded that all documents be written in Basque and Spanish.’
Pred Mod:
[Zer egin ez neki]-ela geratu nintzen
what do not knew-adv stay aux
‘I stood there not knowing what to do.’ (Hualde & Ortiz de Urbina 2003: 712)
Jaikitzen da, [jauzi egiten du]-ela-rik

rise.ipfv aux, jump do.ipfv aux-adv-part
‘(S)he gets up, jumping.’(Hualde & Ortiz de Urbina 2003: 713)
bait-clause
Remark: The conjunction bait can be used in Ref Head function, but this is uncommon. In Ref Mod
function it is used frequently, namely for extraposed relative clauses. In this function, a resumptive pronoun
can optionally be added.
Functional distribution: Flex: Ref Head, Pred Mod ≠ PoS (nouns, adjectives, small modifiers)
Argument encoding: SENT / Ø - Ø / SENT (gapping)
Examples:
Ref Head:
Hau da haren abanrailik handiena [ez baitu ainitz xahutzen]
this is this avantage.part biggest.det not conj.aux much spend.ipfv
‘That’s the main advantage, that he doesn’t spend much.’ (Hualde & Ortiz de Urbina 2004: 648)
Ref mod:
Landibarren badira lau kartier, [horiek bait-ira
Landibar.loc ba.are four neighbourhood those(rsp) conj-are
11
Saltarelli (1988) seems to analyze the suffix –(e)la in MP function as a real adverbializing suffix. Hualde & Ortiz de
Urbina (2003: 712) also allude to a difference between the complementizer and the adverbializer function: “Mitxelena points
out that this modal –ela is not exactly homophonous with the completive –ela since they have a different accentual pattern
(in some Gipuzkoan and Bizkaian dialects).”

Behaunem Dona Martine, Donoztia eta Azkonbegi]
Behaune Dona Martine, Donoztia and Azkonbegi
‘There are in Landibarre four neighbourhoods, which are Behaune, Dona Martine, Donoztia and
Azkonbegi.’ (Hualde & Ortiz de Urbina 2004: 816)
Abun
do / Ø + clause
Argument encoding: SENT / Ø - SENT
Examples:
Ref Head:
An jam [do an karowa ne nde]
3sg know comp 3sg close.to there neg
‘He knew not to [go] close to there.’ (Berry & Berry 1999: 189)
An syogat pa [jogru san].

3sg order child take.off clothes
‘He ordered the girl to take off (her) clothes.’ (Berry & Berry 1999: 187)
gato + clause
Remark: The relative clause can be followed by a determiner.
Functional distribution: Rig: Ref Mod ≠ PoS (modifiers)
Examples:
Ref Mod:
Men mu gu ye [gato man siri su men bi nggon].
1pl go kill person rel do wrong with 1pl poss woman
‘We will go and kill the person who committed adultery with our (clans) woman.’ (Berry & Berry 1999: 146)
Men mu dw syur mo syur wak [gato nje ben] ne

1pl go go.in water at water hole rel people make det
‘We went and washed at the well that people had made.’ (Berry & Berry 1999: 146)
sa gato + clause
Functional distribution: Rig: Pred Mod ≠ (modifiers)
Nominal categories: none
Example:
Pred Mod:
An da ben mó sarewo an yo ben kete
3sg actual do exist however 3sg neg do too.much
bado yo teker [sa gato nyim ne nde re.]
maybe neg too.much adv earlier det neg pfv
‘Although she does [these things] she does not do [them] very much, I mean, not like [she did them]
before.’ (Berry & Berry 1999: 158)
Bambara
Clause + ka
Remark: The exact coding details are unclear; the literal translation suggests that this is a balanced construction.
Verbal categories: Retained?

Examples:
Ref Head:
M’b’a fè i ka taga
I want that you go 2 I want you to go.
M’b’a fè au k’a ka foro cikè

I want that you cultivate your field 2 I want you to cultivate your field. (Brauner 1974: 80)
Pred-le / -ne (‘perfective participle’):

Remark: Apparently, the dependent predicate is non-finite and the relativized item is gapped.
It is not clear how other argument(s) are coded.
Argument encoding: Ø - SENT / Ø?
Example:
Ref Mod:
Mògò [pasa-le]
person loose.weight-pfv.ptc
‘a thin person’ (‘a person who has lost weight’) (Brauner 1974: 73)
min(u) / mun(u) + clause

Example:
Ref Mod:
Dunan [min nana Kulikòrò] oye tubabu ye
Stranger rel come K. cop European cop
‘The stranger who has come to K is a European.’ (Brauner 1974: 82)
Pred-tò (‘present participle’ 2 converb)

Remark: Apparently, the dependent predicate is non-finite, and the co-referential subject is coded twice.
It is not clear how other argument(s) are coded.
Functional distribution: Rig: Pred Mod = PoS (manner adverbs)
Example:
Pred Mod:
[A kasi-tò sègina] a ka dugu
3sg cry-prs.ptc. go.back.? 3sg pst? place / village
‘He went back to his village, crying.’ (Brauner 1974: 72)
Georgian
Pred-a:
Remark: Argument coding is in ergative alignment: subjects of instransitive and objects of transitive verbs
(S and P) are POSS. Transitive subjects (A) are typically OBL: accompanied by mier ‘by’ (or -gan ‘from,
by’) (Hewitt 1995: 542).
Verbal categories: No tense / mood / person-number agreement; aspect is retained.
Examples:
Ref Head:
minda [st’at’iis ts’er-a]
1.3.want.prs article.gen write-nmlz(-nom)
‘I want to write an article.’ (Vamling 1989: 35)

virtʃev [amis gak’eteb-a]-s
1.3.prefer.prs this-gen do-nmlz-dat
‘I prefer to do this.’ (Vamling 1989: 99)
[p’rezident’-is gada-dg-om-a]
president-gen prev-stand.down-ths-nmlz
‘the standing down of the president’ (Hewitt 1995: 542)
[mokalake-ta mier upleb-eb-is ga-mo-q’en-eb-a]

citizen-pl(gen) by rights-pl-gen prev-prev-use-ths-nmlz
‘the making use of their rights by the citizens’ (Hewitt 1995: 542)
m-Pred(-a-)(-el / al) (‘active participle’):

Remarks: Indicates the actor of the relative clause predicate. The negative (privative) counterpart of this
participle is formed with the prefix u-.
Verbal categories: No tense / mood / person-number agreement
Nominal categories: Case agreement
Argument encoding: Ø - POSS (gapping)
Examples:
Ref Mod:
Ševardnaje-m [tvitmprinav-is ga-m-t’ac-eb-el]-i axalgazrd-eb-i
Shevarnadze-erg plane-gen prev-act.ptc-seize-ths-act.ptc-nom youth-pl-nom
da-a-xvret’-in-a
prev-lv-execute-caus-he(aor)
‘Shevarnadze had the young ones, who hijacked the plane executed.’ (Hewitt 1995: 608-609).
Privative:
[k’ont’rol-s da-u-kvemdebar-eb-el]-i birtvul-i energia
control-dat prev-priv.ptc-subordinate-ths-priv.ptc-agr nuclear-agr energy(nom)
‘Nuclear energy, which is subordinate to no control…’ (Hewitt 1995: 609)
Pred-ul / il / m-Pred-ar / al (‘past participle’)

Remark: When derived from a transitive verb, this construction has a passive interpretation; the agent is
marked oblique and the patient is gapped. When derived from an intransitive verb, the constructions has
perfective interpretation and the relativized argument is gapped.
Argument encoding: POSS - Ø (gapping)
Examples:
Ref Mod, transitive:
Sakartvelo [ara-eb-is mier da-p’q’r-ob-il]-i kveq’ana i-q’o
Georgia(nom) Arab-pl-gen by prev-grab-ths-pst.ptc-nom country(nom) sv-cop(3.aor)
‘Georgia was a country (that had been) occupied by the Arabs.’ (Hewitt 1995: 609)
[razik’ašvil-is ča-c’er-il]-i varian’-i

Razikashvili-gen prev-record-pst-ptc-agr variant-nom
‘the variant recorded by Razikashvili’ (Hewitt 1996: 611-612)
Intransitive:
[okro-s verj-s da-k’avšir-eb-ul-i] legend-eb-i
gold-gen ram-dat prev-connect-ths-pst.ptc-agr legend-pl-nom
‘legends linked to the golden ram’ (Hewitt 1996: 609)
sa-Pred(-el / -al / -r) (‘future participle’)

Remark: The meaning of this construction is ‘that which is to be V-ed’. The agent is demoted and as
remains unexpressed; the patient is gapped. It is not clear whether this construction can take an oblique
agent argument.
Argument encoding: Ø - Ø (gapping)

Example:
Ref Mod:
Ager m-i-k’av-i-a [xel-mo-sa-c’er]-i kagald-eb-i
here I-ov-hold-prs.stat-3 hand-prev-fut.ptc-sign]-nom paper-pl-nom
‘I am holding here the papers which are to be signed.’ (Hewitt 1995: 609)
rom + (subjunctive) clause

Remarks: In Ref Head function, the strategy can be combined with a suitable correlative in the main
clause. This is obligatory when the clause is dependent on a postposition or functions obliquely (i.e. follows
a verb that governs instrumental case). (Hewitt 1987: 218)
Complements of modal and desiderative predicates combine with the subjunctive mood (optative when the
matrix verb is present or future tense and pluperfect when the matrix verb is past). The complementizer is
often omitted with subjunctive complements, whereas it is usually obligatory with indicative complements
(Vamling 1989: 32 / 33). With subjunctives, the complementizer rom may also be replaced by titko(s) or any
of its synonyms vitom(c) and vitom(c)da. (Hewitt 1995: 623)
In Ref Mod function either the head or the co-referential noun can be deleted, or both may be retained.
(Hewitt 1995: 606-607) When the co-referential NP is deleted and it does not function as a subject or direct
object in the DC, a resumptive pronoun is often used. In the Ref Mod function, the relativizer itself avoids
clause-initial position and usually appears between the first constituent and the verb. (Hewitt 1987: 187)
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, adjectives)
Examples:
Ref Head:
(is) uk’ve še-v-a-mčn-i-e
(that(nom)) already prev-i-nv-notice-ths-aor.ind
[rom es xalx-i sando ar ar-i-s]
comp this people-nom trustworthy(nom) not be-prs-it
‘I have already noticed that this people is not trustworthy.’ (Hewitt 1995: 613)
Minda [gamovatsxo namtsxvari]

1.3.want.prs 1.3.bake-opt cake.nom
‘I want to bake a cake.’ (Vamling 1989: 33)
vinme-m ar i-pikr-o-s, [titko(s) ingoroq’va-s

anyone-erg not sv-think-aor.subj-he as.if Ingoroq’va-dat
v-a-c’er-de-et im mosazreba-s…]
1-lv-ascribe-ipf-prs.sbjv.pl that view-dat
‘Lest anyone think that we are ascribing to Ingoroq’va that view…’ (Hewitt 1995: 623)
Ref Mod:
[gušin rom beč’ed-i m-a-čuk-e,] is (beč’ed-i) sad ar-i-s?
yesterday rel ring-nom me-lv-present-aor.ind that(nom) (ring-nom) where be-prs-it
‘Where is that ring which you presented to me yesterday?’ (Hewitt 1995: 607)
[K’ino-ši rom (ma-s) e-lap’arak’-eb-o-i] is megrel-i sad

cinema-in rel rsp-dat ipfv-speak.to-ths-ipfv-ind that Mingrelian-nom where
ga-i-can-i?
prev-sv-get.to.know-aor.ind
‘Where did you get to know that Mingrelian to whom you where speaking in the cinema?’ (Hewitt 1995: 607)
REL PRON + clause

Remark: The relative pronouns are vin-c ‘who’, ra-c ‘which’, and ro-mel-i-c ‘who, which’.
Nominal categories: Case agreement on the REL.PRON
Ref Mod:
q’vela, [vin-c nacionalizm-s
all(nom) who(nom)-rel nationalism-dat
a-h-q’v-eb-a] samšoblo-s u-galat’eb-s
prev-it-follow-ths(fut)-he homeland-dat ov-betray-ths(fut)he
‘Everyone who follows nationalism will betray this country.’ (Hewitt 1995: 601)

Avt’or-I [romel-is c’ign-s(a-c) c’v-av-en…]
Author-nom who-gen book-dat(-rel) burn-ths-they
‘the author, whose books they are burning…’ (Hewitt 1995: 600)
ra in clause
Examples:
Pred Mod:
[e-xmar-eb-i-an ra saxalxp armi-is šet’eva-s]
ov-help-ths-prs-they adv people’s army-gen attack-dat
p’art’izan-eb-i a-jlier-eb-en…
partisan-pl-nom nv-strenghten-ths-they(prs)
‘Helping the people’s army to attack, the partisans strengthen …’ (Hewitt 1995: 600)
[mo-h-q’av-s ra sxvadasxva mk’vlevar-ta

prev-he-bring-3(prs) adv different investigator-pl(gen)
azr-eb-i,] marr-i a-sk’vn-i-s..
opinion-pl-nom Marr-nom nv-conclude-ths-he
‘Adducing the opinions of different investigators, Marr concludes…’ (Hewitt 1995: 600)
Rogorc (‘as, like’) + clause

Example:
Pred Mod:
[Rogorc šen g-e-p’rian-eb-a], ise mo-i-kec-i
as you(dat) you-iov-appeal-ths-it so prev-pass-act-aor.ind(.imp)
‘Act as the fancy takes you.’ (Act as you are appealed to.’) (Hewitt 1995: 589)
Bukiyip
Unmarked clause
Examples:
Ref Head:
Énan n-a-kli [yek i-wich umu énaniny moul]
He he-real-say I I-irr-enter ben his work
‘He said that I would have his job.’ (Conrad & Wogiga 1991: 179)
ch-a-kli [ch-e-geik mamawegasibel]

pl.mix-real-say / want pl.mix-irr-build parent.wood.poss.fence
‘They wanted to build a parent type (= very strong) fence. (Conrad & Wogiga 1991: 182)
Clause + (u)li
Remark: The relativized item is gapped, but it is cross-referenced (with a prefix denoting class and
number) on the dependent predicate.

Example:
Ref Mod:
H-a-gab-we-yagú agú-dak
3pl-masc.subj-real-fix-cl3.sg-.obj cl3.sg.dem-this
nebe-gali trag [g-a-lahe-mu bbensin úli]
large-cl3.sg truck cl3.sg.subj-real-travel-ben gasoline rel
‘They repaired this big truck which transports gasoline.’ (Conrad & Wogiga 1991: 103)
bwidouk + clause + -(u)mu

Functional distribution: Rig: Pred Mod = PoS ((small) manner adverbs)
Example:
Pred Mod:
[Bwidouk ecech ch-a-kli-mu]
like they.mix 3pl.mix.subj-real-say-like
‘Like they said’ (Conrad & Wogiga 1991: 968)
Abkhaz
Pred-ra
Verbal categories: No TAM / agreement
Examples:
Ref Head:
[Ara sə-q’a-zaa-ra] (Ø-)r-dər-we-yt’
here 1sg.poss-be-ths-nmlz it-2pl-know-dyn-fin
‘They know that I was here.’ (Hewitt 1979: 31)
[darà rə̀-la wac˚’ə̀ Čerkèssk’-qá ħà-št-ra] (Ø-)ǰà-s-šo-yt’

them them-by(instr) today Čerkessk-to our-send-nmlz it-prev-1sg-be.surpised.at-fin
‘I am surprised at their sending us to Čerkessk today.’ (Hewitt 1979: 31)
Pred-N.FIN
Remark: This construction makes use of a special non-finite paradigm. It expresses nearly all verbal
categories that are also expressed in independent clauses, but in a different form.
Verbal categories: Almost all retained, but dependent paradigm
Example:
Ref Head:
[d-şaa-wà] (Ø-)xà-s-c’a-wa-yt’.
he-come-dyn(n.fin.prs) it-head.in-I-put-dyn-fin(prs)
‘I believe that he will come.’ (Hewitt 1987: 238)
REL.PRON + Pred-N.FIN
Remarks: This construction makes use of a special non-finite paradigm. It expresses nearly all verbal
The person affix that is used in independent clauses is replaced with one of two relative affixes. -y(ə) is used
in stead of all person affix of the first declination, regardless of the person, class and number of the head
noun, whereas -ɀ(ə) performs this function for all person affixes of declinations 2 and 3. (Hewitt 1987: 200)
Nominal categories: Class agreement of REL.PRON

Examples:
Ref Mod:
[a-pħ̊ ə̀s də-z-šə̀-z] a-xàc’ d-aa-yt’
det-woman her-who(rel)-kill-n.fin(pst.indf ) det-man he-come-fin
‘The man who killed the woman came.’ (Hewitt 1987: 201)
[r-c’a-y˚ə̀s-s yə̀-q’a-w] a-pħ̊ ə̀s də-y-də̀r-wa-yt’

teacher=adv who(rel)-be-n.fin the-woman her-he-know-dyn-fin
‘He knows the woman who is a teacher.’ (Hewitt 1987: 201)
[a-xà’a yə̀-y-ba(k˚a-)-z] à-ħ˚sa (Ø)-z-də̀r-wa-yt’

the-man whom(rel)-he-see-pl-n.fin the-women them-I-know-dyn-fin
‘I know the women whom the man saw.’ (Hewitt 1987: 201)
ş-Pred-N.FIN
Remark: This construction makes use of a special non-finite paradigm. It expresses nearly all verbal
Example:
Pred Mod:
[Ye-ş-w-à-s-ħ̊a-z] yə-q’a-c’à
it-adv-you-to-I-say-nfin(pst.indf ) it-prev-do.imp
‘Do it as I told you!.’ (Hewitt 1987: 120)
Polish
Pred-nie
Remark: Occurs with A-argument unexpressed under co-referentiality and P-argument SENT, but also
with an A-argument oblique A and a P-argument POSS.
Verbal categories: No TAM / agreement
Argument encoding: OBL - POSS / Ø - SENT
Examples:
Ref Head:
Prosze̖ o [zwoln-ienie z pracy dwie pracownice]
request:1sg about dismiss-nmlz:loc from work:gen two:acc woman.worker:acc.pl
‘I request the dismissal from work of the two women workers.’ (Comrie 1976: 191)
[kupowa-nie mie̖s-a prez Hanke̖]

buy-nmlz meat-gen by Hanka
‘the buying of meat by Hanka’ (Koptjevskaja-Tamm: 293)
Pred-INF (various forms)

Remark: There is a number of different infinitival endings:
-ać -ąć
-eć -nąc
-ieć -ść
-ić -źć
-yć -ować
-uć -ywać
-c -iwać -awać
Verbal categories: No tense, aspect, agreement
Example:
Ref Head:
Postaniwiɫem [kupić dom].
I.decided buy.inf house
‘I decided to buy a house.’ (Bielec 1998: 19)

Pred-c- (‘present active participle’)
Verbal categories: No tense, aspect, valency, person marking
Nominal categories: Adjectival agreement (number, gender, and case)
Example:
Ref Mod:
Widzȩ chɫopca [sɫuchają-c-e-go radia]
I.see boy listen-ptc-sg.masc-case radio
‘I see a boy (who is) listening to the radio.’ (Bielec 1998: 170)
Pred-any / -ana / -ane(SG) / -ane / -ani(PL) (‘present passive participle’)

Remarks: Used with imperfective verbs only. Can possibly be regarded as derived adjectives, since there are
no overt arguments: The relativized item is gapped, and the demoted agent argument remains unexpressed.
Example:
Ref Mod:
Odzież [sprzedaw-ana] tam jest tania.
clothes (being) sold-ptc(pass.prs) there are cheap
The clothes (being) sold there are cheap.’ (Bielec 1998: 171)
Pred-PASS.PST.PTC (various forms)

Remarks: There are several formation strategies, depending on the form of the infinitive.
This construction is used with perfective verbs only.
These participles can possibly be regarded as derived adjectives, since they take no overt arguments: the
relativized item is gapped, and the demoted agent argument remains unexpressed.
Example:
Ref Mod:
Mikoɫaj ma [zɫamaną] nogȩ
N. has broken(pass.pst.ptc) leg
‘Nicolas has a broken leg.’ (Bielec 1998: 171)
Pred-PRS.3rdPL + -c (‘adverbial participle’)

Remark: The (present) converb (called adverbial participle in Bielec 1998) is formed by adding the suffix -c
to the 3rd person plural of the present tense.
Verbal categories: No tense, aspect, person marking
Example:
Pred Mod
[Sɫuchają-c muzyki,] ubraɫem siȩ.
listen.prs.3.pl-conv music dress.I (my)self
‘Listening to music, I got dressed.’ (Bielec 1998: 71)
że + clause

Example:
Ref Head:
Myśle, [że ona jest mȩżatką].
‘I.think that she is married.’ (Bielec 1998: 239)
kktóry / która / które / co (REL.PRON) + clause

Nominal categories: Adjectival declination of relative pronoun
Example:
Ref Mod:
Chata, [gdzie mieskali,] byɫa maɫa. [= w której]
Cottage [in.which they.lived] be small
‘The cottage in which they lived was small.’ (Bielec 1998: 155)
Jak (gdy)by + clause

Functional distribution: Rig: PredMod = PoS (small / derived manner adverbs)
Example:
Pred Mod:
Agata spojrzala na mnie, [jak gdyby chciaɫa o coś zapytać].
Agatha looked at me as if she wanted to say something.’ (Bielec 1998: 238)
Burushaski
Pred-(á)as (INF)
Remarks: Used as the complement of modal, phasal, desiderative, and manipulative predicates. With phasal
and desiderative predicates, the dependent verb form it is declined like a regular noun.
When combined with an inherent (dative, inessive, adessive, superessive or locative) case marker, this
construction can be used as an adverbial simultaneity clause.
Functional distribution: Flex: Ref Head, Ref Mod, + CASE also Pred Mod. ≠ PoS (nouns, adjectives,
small / derived manner adverbs)
Verbal categories: Tense, mood and subject agreement are lost. Aspect and object-agreement are retained.
Nominal categories: (CASE), see above
Examples:
Ref Head:
buṭ muškíl bilá [góo-ltir-as]
very difficult be.IV 2-show-inf
‘It is very difficult to show (it) to you.’ (Anderson 2002: 545)
Ref Mod:
[Chá-aṭe oóo-rvγ-as] hvk
post-superess neg-sit-inf dog
‘a dog which doesn’t sit at its post’ (Anderson 2002: 545 / Berger 1998: 171)
Pred Mod:
[sén-as]-ar
say-inf-dat
‘when he said’ (Anderson 2002: 547 / Berger 1998: 190)
[du-ús-as]-ulo
subj.vers-come.out-inf-iness
‘when he came out’ (Anderson 2002: 547 / Berger 1998: 190)
Pred-im / -um / -am (‘aorist participle’)

Remark: In combination with a dative, inessive, adessive or superessive (2 privative) case marker this
construction can be used as an adverbial simulteneity clause. The durative form of the aorist participle can
also function adverbially, in combination with locative or superessive case, or with the comitative marker
káa and the genitive case. These constructions seem to have ‘true’ manner semantics.

Functional distribution: Rig: Ref Mod (+ CASE also Pred Mod) = PoS (adjectives)
Argument encoding: SENT - SENT Ø- SENT
Examples:
Ref Mod:
[i-ŋí burúm-man-úm] mapéer-an
he-beard white be(come)-aor.ptc elder-sg.art
‘an old man with a white beard.’ (Anderson 2002: 546 / Berger 1998: 166)
Pred Mod:
[ɣaliz bay-a-m]-ulu Khudá-e-re duá et-a-m
ill be-1-aor.ptc-iness God-obl-dat prayer aux-tr-1.aor.ptc
‘When I was ill I prayed to God.’ (Anderson 2002: 549)
[oó-ar-um]-aṭe
neg-cry-aor.ptc-superess
‘without crying / having cried’ (Anderson 2002: 549 /Tikkanen 1995: 493)
Khos chigír-an-e [i-súmal phíphil éc̆-um]-e a-yár man-ím-I

This.III goat-sg.art-erg III-tail wag aux.tr.dur-aor.ptc-loc 1-side become-aor-iii
‘The goat came before me wagging its tail.’ (Anderson 2002: 550 / Berger 1998: 171)
[duró ay-éc̆-um]-e
work neg-aor.ptc-loc
‘without working’ (Anderson 2002: 551 / Berger 1998: 172)
in [ɣar éc̆-um-e káa] girát-im-i

he song aux.tr.dur-aor.ptc-gen com dance-aor.ptc-i
‘He danced singing’ (Anderson 2002: 551 / Berger 1998: 140)
n-Pred-(a)n (RDP)
Remark: This is the only primary converbal form, i.e. one that is not derived from another non-finite
form by case-marking. It normally expresses anteriority but it may also express manner of action or an
accompanying circumstance, in which case the conjunctive participle is often reduplicated.
Examples:
Pred Mod:
Aya máma [já-ar taklíiff n-á-či-n] uyóŋ-ko u-yú-ar
father(erg) mother(erg) i-dat trouble conv-1sg.io-give-conv big-pl 3cl.poss-son(pl)-dat
šuá n-e gar-íŋ ét-um-an.
good cp-do marriage-pl do-stat-pl.subj
‘Father and mother arranged good marriages for their big sons, giving me a lot of trouble.’ (Tikkanen 1995: 502)
[Čáya n-ét-an n-ét-an] wazíir-e sén-im-i.

story conv-do-conv conv-do-conv vizier-erg say-stat-2masc:cl:sg.subj
‘Going on talking and talking, the vizier spoke.’ (Tikkanen 1995: 502)
Clause + ke / ki
Remark: This construction is borrowed from Persian / Urdu. It can be used as the object complement of
utterance and cognition predicates (no example available), as a relative clause (no example available), and as
a simultaneity clause.
Functional distribution: Flex: Ref Head, Ref mod (Simultaneity clause) ≠ PoS (nouns, adjectives,
small / derived manner adverbs)

Example simultaneity clause:
[Zuŋáatiŋ-e girám-ular asqúr-iŋ d-íi-m-i
z-gen village-ill flower-pl pref.-3cl.sg-(come)- 3cl.sg
ke] han dastúur-an b-il-úm
sub one:cl custom-sing be-3cl.subj-st
‘The apricots coming into blossom, when the flowers came to the village of Zungating, there was a custom
…’ (Tikkanen 1998: 498)
Clause + -sén / -ét (QUOT)

Remark: The quotative marker takes the form of the anterior converb of the quotative verb sén- ‘to say’ or
ét- ‘to do / to speak’. The construction can be used in Ref Head function for complements of utterance and
cognition predicates. No examples.
Lavukaleve
Pred-e / -i
Remark: The so-called “Agreement Suffix” (which is lost in this construction) marks gender and number
of a core argument (which argument that is depends on various factors, such as focus). The possessive
paradigm is identical to the verbal subject prefix paradigm except for one form: the first person singular
subject prefix is a-, while the possessive form is nga-. However, the fact that nominalizations take a- is
taken as evidence that the other forms are also subject prefixes rather than possessive markers.
The construction can take a determiner and can also be dependent on a postposition.
Verbal categories: No TAM, no Agreement Suffix
Nominal categories: (CASE / DET)
Example:
Ref Head
Ngai [nga-bo’rea la o-ma-e]
1sg 1sg.poss-arrow sg.fem.art 3sg.fem.obj-take-nmlz
e-lili-re ta a-lei
3sg.neut.obj-want-n.fin just 1sg.subj-exist
‘No! I just want to take my arrow.’ (Terrill 2003: 352)
Clause + AGR + DET

Remark: The Agreement Suffix is obligatorily used on the (final) dependent predicate to cross-reference
the head of the construction, which can be the subject, the object, or a postpositional object. Relative
clauses are internally headed.
Nominal categories: DET (+ adjectival gender and number agreement)
Examples:
Ref Mod:
[Ali nga-fo’sal a-u-m na] a-le-m fin.
man 1sg.poss-fish 3sg.masc.obj-eat-sg.masc art 1sg.subj-see-sg.masc 3sg.masc.foc
‘I saw the man who ate my fish.’ (Terrill 2003: 442)
[Ali a-na so-soko-ne a-ngooa-m na] o-foe

man 3sg.masc.obj-in rdp-laugh-ipfv 1sg.subj-stay-sg.masc sg.masc.art 3sg.poss-pig
na o-lufu-m
sg.masc.art 3sg.subj-leave-sg.masc
‘The man who I laughed at had lost his pig.’ (Terrill 2003: 448)
Alamblak
Pred-nef
Remark: The subject of intransitives may be incorporated or POSS. With transitive predicates, either the
subject or the object can be POSS.
Verbal categories: No agreement (actor and undergoer)

Nominal categories: Gender and number marking on predicate
Argument encoding: INC / POSS / SENT - SENT / POSS
Examples:
Ref Head:
Yifemr pëthawonmëanr nan-ho yi-nef-t
father talk.try.I.him [I-gen go-nmlz-3sg.fem]
‘I tried talking to Father (about) my going.’ (Bruce 1984: 124)
Yifemr pëthawonmëanr [rër-ho yak-ni-nef-t na]

father talk.try.I.him he-gen get-go-nmlz-3sg.fem me
‘I tried talking to Father (about) his taking me.’ (Bruce 1984: 125)
akfërafëwahn [yima-m bupa-r-oh yak-nef-t]

do.not.talk.forbid.you person-3pl water-3-gen.pl get-nmlz-3sf.fem
‘Don’t forbid the men(’s) getting of water.’ (Bruce: 125)
Pred-(kfë)t (INF)
Remark: Unlike the nominalization in -nef, the infinitive does not take gender / number markers.
Verbal categories: No tense / agreement (actor and undergoer)
Argument encoding: INC / POSS - INC / POSS
Examples:
Ref Head:
Na iñji wañfɨnahmëanr [baripat rhu-haku-t]
1sg thus hear.arrive.rem.pst.1sg.3sg.fem lake.only remain-always-inf
‘Thus I have heard it up to now (that) it is always only a lake.’ (Bruce 1984: 284)
nanho wurat kɨfiwt [yamkopɨntho nayay-kfët], nayayrhwa

1sg.gen foot speak.prs.3sg.fem Amongabi.gen come-inf come.fut.1sg
‘(if / when) my foot speaks (= when I like) to come to Amongabi, I will come.’ (Bruce 1984: 284)
(ɨnd -Pred) + clause

Remark: Relative clauses are optionally marked with ɨnd , but mostly unmarked. The function of the head
noun is not marked, except when a possessor is relativized.
Verbal categories: Aspect and mood are retained; agreement (subject and object) is lost
Examples:
Ref Mod:
[met-t maroha-haymë] yima-r
woman-3sg.fem money-gave person-3sg.masc
‘a man (who) gave money to a woman / a man (to) whom a woman gave money’ (Bruce 1984: 106)
[Ø na yawyr ɨnd-tat-më] mɨy-t

- I dog dem-hit-rem.pst stick-3sg.fem
‘a stick (with) which I hit a dog’ (Bruce 1984: 111)
[Ø yimar kuñt hingma-më-r-ho] met-t

- man house build-rem.pst-3sg.masc-gen woman-3sg.fem
‘a woman whose man built a house’ (Bruce 1984: 111)
ɨnd habhi kmi [na ɨnd-kfëmë-t]

dem small place I rel-said-3sg.fem
‘the small place (about) which I spoke’ (Bruce 1984: 108)
Pipil
ka(h) + clause
Functional distribution: Flex: Ref Head, Ref Mod (occasionally) ≠ PoS (nouns, adjectives)

Examples:
Ref Head:
ya klaroh k-ita-k [kah wi:ts ne siwa:-t].
he clear it-see-pret comp come the woman-abs
‘He saw clearly that the woman (was) come(ing).’ (Campbell 1985: 126)
tesu ki-mati katka [ka ne i-siwa:-w se bru:hah].

neg it-know before comp the his-wife-poss a witch
‘(He) didn’t know before that his wife (was) a witch.’ (Campbell 1985: 126)
Ref Mod:
Ni-k-miktih ne mistun [ka ki-kwah ne tu:tu:-t]
I-it-killed the cat rel it-ate the bird
‘I killed the cat which ate the bird.’ (Campbell 1985: 129)
kunih ahsi-ke-t i-chan ne ta:ka-t [kah ki-kutamin-ki]

then arrive-pret-pl his-house the man-abs rel it-throw-pret
‘Then they arrived (at) the house of the man who threw it down.’ (Campbell 1985: 129)
ne (REL. PRON) + clause

Example:
Ref Mod:
Naha ni-k-ita-k ne siwa:-t [ne ki-pa:k ne kwah-kwa:ch-ti]
I I-see-her the woman-abs rel it-washed the pl-clothes
‘I saw the woman who washed the clothes.’ (Campbell 1985: 129)
ke (REL) + clause
Remark: This construction is borrowed from Spanish (que).
Example:
Ref Mod:
Kunih ne ta:ka-t [ke ki-kutamin] k-its-ki ne chumpipi k-wi:ka ka i-chan
then the man-abs rel it-throw it-grab-pret the turkey it-take to his-house
‘Then the man who threw it down grabbed the turkey (and) took it to his house.’ (Campbell 198: 129)
ke:n-aken (‘just as’) + clause

Remark: This construction has similative semantics.
Functional distribution: Rig: Pred Mod ≠ PoS (no lexical strategy)
Example:
Pred Mod:
Ah, ni-pa:xa:lua [ke:n-aken taha ti-ki:s-tuk ti-pa:xa:lua]
Oh I-walk just-as you you-leave-pfv you-walk
‘Oh, I’m taking a walk just as you have come out to take a walk.’ (Campbell 1985: 289)
ADV + clause
Remark: This construction has simultaneity semantics.
Functional distribution: Rig: Pred Mod ≠ (no lexical strategy)

Example:
Pred Mod:
Ke:man yaha yawi katka derechoh ne kayeh, yah ki:s-ki i-i:xpan siwa:na:wal
when he go before straight the street then leave-pret his-before Siguanaba
‘When he was going (down) the street, he came out in front of (the) Shaunaba.’ (Campbell 1985: 131)
Wambon
-e (CONN) + clause
Remark: The connector -e links all preverbal NPs, irrespective of their functional or structural specification,
so this includes object clauses.
Argument encoding: SENT- SENT / Ø - SENT
Example:
Ref Head:
Kono heta-khe-n-[e lava-tmbo…]
and see-3sg.prs-tr-conn trap-3sg.pst
‘And he saw that he had trapped it (…)’ (De Vries 1986: 29)
-a + clause-o
Remark: The suffix -o links all types of pre-nominal modifiers to the head noun, including relative clauses,
which are additionally marked by the relativizer -a, which delineates the relative clause by cliticizing to
its first constituent. Relative clauses can also be head-internal. It that case, the construction is not pre-
nominal, so the connector -o does not appear.
Argument encoding: Ø / SENT - Ø / SENT
Examples:
Ref Mod:
Nuk [oy-a temke-n-o] kev-e hetak-nok-ndep
I pig-rel shoot-3sg.prs-tr-conn man-conn see-neg-1sg.prs
‘I do not see the man who shoots the pig.’ (De Vries & De Vries-Wiersma 1992: 56)
[Aliv-a ndu-n-e takhima-lepo-n-eve] setelep-ke-khe

yesterday-rel sago-tr-conn buy-1sg.pst-tr-top delicious-superl-3sg.prs
‘The sago I bought yesterday is delicious.’ (De Vries & De Vries-Wiersma 1992: 57)
Pred stem-mo
Remark: This is a “medial verb form”, which takes switch-reference marking for same subject (SS).
The interpretation is temporally neutral (but conceptually close).
Verbal categories: No tense, mood, and subject agreement
Example:
Pred Mod:
Jakhov-e [matet-mo] ka-lembo?
they-conn good-ss go-3pl.pst
‘Did they travel without problems?’ (lit.: ‘Did they go well?’) (De Vries & De Vries-Wiersma 1992: 19)
Pred-o
Remark: This is also a “medial verb form” with same-subject switch reference marking. The interpretation is
not temporally neutral: it expresses simultaneity.
Verbal categories: No tense, mood, and subject agreement. Transitivity markers can be retained.

Example:
Pred Mod:
Jakhov-e [ko-n-o] nggom li-knde
they-conn go-tr-ss song sing-3pl-prs
‘While travelling they are singing.’ (De Vries & De Vries-Wiersma 1992:20)
clause-ka
Remark: The suffix -ka (and its allomorphs) is a circumstantial suffix that also occurs with inanimate NPs that
specify the circumstances of the event denoted by the verb, including time, location, instrument, and manner.
Example:
[Nukh-e ande-lepo-ngga] ev-o kav-e nde-tmbo
I-conn eat-1sg.pst-adv that-conn man-conn come-3sg.pst
‘When I ate, that man came.’ (de Vries 1986: 41)
Dhaasanac
Clause (+DEM) + DET
Remarks: Used for the complements of desiderative and perception predicates, and for all types of relative
clauses. With subject relative clauses the subject is repeated in the matrix clause with a pronoun.
The construction further takes the determiner ka or its clitic form =a; in the Ref Mod function often
preceded by a deictic.
There is only one difference with independent clauses: for constructions in non-past tense, imperfective
aspect is excluded; instead, the dependent form must be used.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, (small) adjectives)
Argument encoding: Ø / SENT - Ø / SENT
Example:
Ref Head:
Yú [kúun kosolonni=a] ku Ɂargi
I you laugh.pfv=det you.obj see.pfv
‘I saw that you were laughing.’ (Tosco 2001: 287)
yú [kúun hí fuɗɗu=a] feɗe

I [you 3(verbal pron) open.dep=det want.ipfv
‘I want you to open it.’ (Tosco 2001: 288)
Ref Mod:
Máa [ɠuoɗ ɠoɠɠoɗo=ti=a] hé ye Ɂóg
Man trees dig.rdp.dep=that=det 3.subj me know
‘That man who is digging trees (he) knows me.’ (Tosco 2001: 283)
Pred Mod:
Yú [ɠúo ɠii ka] máa ɲigéɲ=ati=a ɗu seõI
I cattle have.pfv det person young=that=det upon go.pfv
‘I, having the cattle, went to the boy.’ (2001: 286)
Pred-ɲ / -an
Remarks: The suffix -ɲ is used for the basic and causative / factitive paradigms, -an for the middle and
inchoative paradigms. The dependent predicate has feminine nominal gender.
The object is unmarked (noun-stripping).
Verbal categories: Tense and aspect are lost
Nominal categories: (DET), gender marking
Argument encoding: Ø / POSS - unmarked
Examples:
Ref Head:
[rúb mummur-an] hé beyyi
sorghum cut.rdp.m-nmlz 3.subj finish.pfv
‘The harvesting of the sorghum is finished.’ (Tosco 2001: 121)

Yú [ʔáɲ ɠali-ɲ ci]=a feɗe
I animal skin-nmlz 1.poss=det want.ipfv
‘I want to skin the animal myself.’ (lit. ‘I want my animal skinning.’) (Tosco 2001: 121)
Berbice Dutch Creole

(fu / fi)) + deranked clause
Remark: The complementizer can be deleted.
Examples:
Ref Head:
Eke timi [(Ø) tan l:anggi ka]
1sg be.able stand long neg
‘I am not able to stand up for very long.’ (Kouwenberg 1994: 249)
Governor ma suk-a [fi nimi hosa ju krik-te-ke.]

Governor irr want-ipfv comp know how.foc 2sg get-pfv.1sg
‘The governor will want to know how you got me.’ (Kouwenberg 1994: 249)
(bi(fi)) (QUOT) (+ dati) (COMP) + clause

Remark: The complementizer dati can be deleted. Dati-clauses can also be preceded by the quotative
marker bi(fi) (from the verb for ‘to say’), resulting in a ‘double’ complementizer sequence.
Examples:
Ref Head:
Bat nou eke horo [dat-o mw-a]
but now 1sg hear comp-3sg go-ipfv
‘But now I hear that he is leaving.’ (Kouwenberg 1994:244)
Bateni nim-te [dat-ek ma ku-mona-ngga]

but.3pl acquire.knowledge.pfv comp-1sg irr come-Mona-loc
‘But they knew that I would come to Mona’s.’ (Kouwenberg 1994: 241)
Eke pan-te ju [date ju mu bifi ka]

1sg tell-pfv 2sg comp 2sg must speak neg
‘I told you that you must not talk.’ (Kouwenberg 1994: 238)
Ek glof ka [(Ø) o nin-te musu]

1sg believe neg 3sg know-pfv much
‘I don’t believe he knows much.’ (Kouwenberg 1994: 242)
Eni pama-te-ke [bifi date-ke ma mu danga ka]

3pl tell.pfv.1sg quot comp-1sg irr go there neg
‘They told me that I should not go there.’ (Kouwenberg 1994: 238)
(Wh-word) + clause:
Remark: The Wh-element can be deleted (only in restrictive clauses).
Functional distribution: Rig: Ref Mod = (adjectives)
Example:
Ref Mod:
Di gugu manj-ap [wat fol me hari]
the big=big man-pl what full with hair
‘the giants that are covered with hair’ (Kouwenberg 1994: 268)

Di bwa:, di en [wati wa siki,] ori doto-te esti
The leg the one what pst ill, 3sg die-pfv first
The leg, the one that was ill, it became numb first.’ (Kouwenberg 1994: 271)
Di sem jermatoko [(Ø) eke pan-te ju abot]

the same woman.child 1sg tell-pfv 2sg about
‘the same girl I told you about’ (Kouwenberg 1994: 268)
Babungo
lāa + clause:
Remark: The subject of the matrix clause is repeated in the dependent clause as a logophoric pronoun.
Example:
Ref Head:
Làmbí gì [lāa yì zɨ́zāsə̄]
Lambi say-pfv that he-l.pron be.sick.progr
‘Lambi said that he was sick.’ (Schaub 1985: 31)
fáŋ / yúu + clause

Remarks: In Ref Mod function, if the head noun is the subject of the main clause, it may be reintroduced
into the matrix clause with a pronoun after the relative clause, or the head noun may be repeated together
with an anaphoric demonstrative after the relative clause (especially after ‘heavy’ ones). If the relativized
item is an instrumental, locative, or time adverbial, a resumptive pronoun is obligatory. In subject and
object clauses resumptive pronouns are also possible, but usually not expressed.
In Pred Mod function, the construction has similative semantics.
Functional distribution: Flex: Ref Mod, Pred Mod (similative) ≠ PoS (small / derived adjectives,
small manner adverbs)
Examples:
Ref Mod:
mə̀ yè wə́ ntɨ ə́ [fáŋ / yúu (ŋwə́) shàw ngú yê
I see.pfv person that rel he steal.pfv fowl your
‘I have seen the man who has stolen your fowl.’ (Schaub 1985: 32)
lwàŋ ngwə́ [fáŋ tĭ̆i ŋwāa nə̀ jwì nə́

hammer that [REL father my pst come.pfv with
ŋwə́nə̂,] lwàŋ ghɔ̄́ lùu táa ŋȉI
it(anaph)] hammer that.anaph be in house
‘That hammer which my father came with (it), that hammer is in the house. (Schaub 1985: 33)
Pred Mod:
ŋwə́ sɔ̀ sāy [fáŋ nshú wī nàysə́ tɨ́ ŋwə́]
she plant.pfv corn adv mother her tell.pfv to her
‘She has planted the corn like her mother told her.’ (Schaub 1985: 39)
ŋwə́ gwìa tɨ́ kî [yúu vɨ́ nàysə́ tɨ́ ŋwə́]

he cut-pfv tree that adv they tell-pfv to him
‘He has cut that tree how / when he was told.’ (Schaub 1985: 39)
kɨ̀(i) / Ø + clause
Remark: There are two types of these “simultaneous aspect constructions”: One relating to a preceding verb
(anaphoric), and the other relating to a following verb (kataphoric). With stative verbs the anaphoric
construction is marked by the simultaneity particle kɨ̀. With other verbs, there is no overt coding. In both
cases, the verb following the particle can only be in the imperfective aspect.
The other type (kataphoric) is used when event B occurs while A is still in process. It is marked by the
marker kìi in the first verb phrase, which expresses the event in process. The verb with which the marker
kìi occurs, is in the imperfective aspect, while the verb of the following clause is in the perfective aspect.
Functional distribution: Rig: Pred Mod = PoS (small manner adverbs)
Verbal categories: Restricted aspect

Argument encoding: Ø-SENT
Examples:
Pred Mod:
Anaphoric:
ŋwə́ bwéy [kɨ̀ shə́ŋ ŋwáa]
he sleep-impf sim rest-ipfv body
‘He was asleep, resting his body.’ (Schaub 1985: 220)
ŋwə́ nə́ tó’ fɨ́ fúu jɨ ɨ̀ , [zwá nyáysə̄]

he pst walk-ipfv from on road sing-ipfv songs
‘He was walking on the road, singing songs.’ (Schaub 1985: 220)
Kataphoric:
ŋwə́ [kìi tó’ fɨ́ fúu jɨ ɨ̀ ], gwə̀kə̀ gǔ nsí
he sim walk-ipfv from on road suddenly fall-pfv ground
‘When he was walking on the road, he suddenly fell down.’ (Schaub 1985: 221)
Nama
Clause without ke (= DECL) + !xáis-a
Remark: This construction is marked by the nominal class marker !xáis (SG. FEM), which can be
shortened to -s, and the object marker -à.
Apart from the declarative marker ke, no verbal categories are lost.
Functional distribution: Rig: Ref Head = PoS (small manner adverbs)
Verbal categories: Retained except declarative marker
Nominal categories: Nominal class
Examples:
Ref Head:
Tsíí / / ’iípàk-kxm̀ ke kè míípa [!úu-kxm̀ ta !xáis-à]
And him-we decl told were.going-we sg.fem-obj
‘And we told him that we were going.’ (Hagman 1974: 257)
/ / ’íípke ‘aḿ’ase kèrè ≠óḿ [ / ’aé- / / amsà xuú-kxm / xií hàa !xáis-à]
he decl really believed we had come from Windhoek sg.fem-obj
‘He really believed that we had come from Windhoek.’ (Hagman 1974: 257)
Unmarked clause without ke (= DECL) (+ RSP)

Remark: When the head of a relative clause functions as an oblique argument, the construction is usually
marked with an appropriate postposition or associative particle, and a resumptive pronoun may be added
which agrees with the head in person / number / gender.
Verbal categories: Retained except declarative marker
Nominal categories: (Nominal agreement of RSP)
Examples:
Ref Mod:
[ / / naá !hùupà xuú kè / xií hàa ‘ií] kxòeǹ
from that land had come the.people
‘the people who had come from that land’ (Hagman 1974: 229)
Oblique argument:
[Tiíta / / ’íis tàpa síí nìí / / xáa / / xaa] !’áas
I which(rsp) at am going to teach town
‘the town I am going to teach (in).’ = ‘the town where I am going to teach’ (Hagman 1974: 231)
Deranked clause without ke (= DECL) + -se / !’aa / tsíí / Ø

Remark: The declarative marker ke cannot be expressed and the marking of aspect is restricted: the
dependent clause must have non-punctual aspect. Usually, the subject is unexpressed under co-
referentiality.
The adverbial manner suffix -se and the general adverbial subordinating conjunction !’aa are completely
interchangeable, but -se is by far the most frequently used. The construction with tsíí requires punctual
aspect in the dependent clause.
The unmarked variant has the same meaning, but can only be used with a special set of verbs in the

dependent clause. Tense cannot be expressed in this variant (but aspect is retained).
Verbal categories: No declarative marker, restricted tense / aspect
Examples:
Pred Mod:
[!ái ‘a≠áó] ra-se-p ke kè pèé
rejoicing-adv-he decl left
‘Rejoicing, he left.’ (Hagman 1974: 244)
!xóókxoeku kè !nàeí’úhè [kxétehǹ / xáa !áéhè tsíí]

the.prisoners were transported bound with chains adv (Hagman 1974: 246)
Hdi
Clause + ká (+ low tone subject pronouns)
Remark: All subject pronouns following the complementizer, except the 3rd person singular á have low
tone, whereas subject pronouns in independent clauses, except for the 3rd person plural, have high tone.
Verbal categories: Retained (special subject forms)
Examples:
Ref Head:
[m̀nd-á ráyá Mbákà ká] mbítsá mná-tà
man-gen hunt Mbaka comp Mbitsa say-ref
‘Mbitsa said that Mbaka is a hunter.’ (Frajzyngier & Shay 2002: 422)
Ndá sn-í [zá-ná-p-zá-xə̀n tá pìtsákw-á-tán ká-‘á]

stat know-1sg forget-dem-out-forget-3pl obj hoe-gen-3pl comp
‘I know that they forgot their hoe.’ (Frajzyngier & Shay 2002: 450)
tá (OBJ / COMMENT) + deranked clause:

Remarks: The marker tá is an object / comment marker: It is also used to mark lexical objects and
comments in focus constructions (cf. glosses in examples).
In Ref Head function the subject is POSS. In Ref Mod function it is gapped, but marked with a possessive
pronoun on the dependent predicate. In Ref Mod function, this construction is used only for perfective
subject relative clauses.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, small adjectives)
Verbal categories: No person marking
Nominal categories: Possessive person marking
Examples:
Ref Head:
Sí tà ɗv-áy-xə̀n [tá hliy-á-mú]
pst ipfv want-po-3pl obj leave-gen-1pl.incl
‘They wanted us to leave.’ (lit. They wanted our leaving.) (Frajzyngier & Shay 2002: 480)
Snà-n-sn-í [tá ghw-ày-tán tá ngúrldùŋ-á hlá.]

hear-3-hear obj slaughter-po-3pl.poss obj neck-gen cow
‘I heard them slaughter cattle.’ (Frajzyngier & Shay 2002: 470) (lit. I heard their slaughtering the cow’s neck?)
Ref Mod:
ghùrúm [tá lá-ghw-í ndá mà xàɗík]
hole com go-d:so-ref assoc in ground
‘a hole that went deep into the ground’
màmú sàn m̀ghám [tá kl-áf-tá màràkw xìs]

exist certain chief com take-up-ref wife two
‘There was once a chief who married two wifes.’ (Frajzyngier & Shay 2002: 470 406)
tà (IPFV) + verbal root form (+ ka (COMP)-subject pronoun)

Remark: The predicate takes dependent aspect marking, but is otherwise balanced.

This construction is used for (imperfective) object relative clauses only.
Verbal categories: Retained (dependent imperfective aspect)
Argument encoding: SENT - Ø (gapping)
Example:
Ref Mod:
Wúyás kwì [tà klá-ghá-tà-ŋnì, ká-xə̀ŋ]
here thing ipfv take-2sg-ref-1pl.excl comp-3pl
‘Here is the thing that we give you.’ (Frajzyngier & Shay 2002: 409)
tà (IPFV) + clause with nominalized verb

Remark: Used for subject relative clauses only.
Verbal categories: Only imperfective aspect retained, no other affixes indicating role / number of
argument(s).
Example:
Ref Mod:
Màmú tsə̀mə́k-xà kɗérí [tà ìrí ndà tsí]
exist enemy-pl:gen Kderi ipfv envy.nmlz assoc 3sg
‘There were enemies of Kderi who envied him.’ (Frajzyngier & Shay 2002: 404)
Pred-a (dependent perfective aspect):

Remark: Used for (perfective) object relative clauses only.
Verbal categories: Retained (dependent perfective aspect)
Argument encoding: SENT - Ø (gapping)
Example:
Ref Mod:
Gítà kàm ná dzà’á plá-ghá-m-plá-ŋní tá dá-ghá [dzá-xə̀ŋ]
today then dem fut return-2sg-in-return-1pl.excl obj father-2sg kill:dep.pfv-3pl
Today, we will avenge your father whom they have killed. (Frajzyngier & Shay 2002: 410)
Mandarin Chinese
Unmarked clause
Remarks: In Ref Mod function, the relativized element is gapped; in Pred Mod function it can be left
unexpressed under co-referentiality.
In Ref Mod function, this construction is called a “descriptive clause”. It is semantically similar to a relative
clause marked with de (see below), although according to Li & Thompson it constitutes a separate
assertion: “Semantically, a descriptive clause simply adds another assertion to the first one. A relative
clause, on the other hand, is a part of the noun phrase naming the item in question, and as such expresses a
pre-established class of items with the property it names.” (Li & Thompson 1981: 618)
Functional distribution: Flex: Ref Head, Ref Mod, Pred Mod ≠ PoS (nouns, small adjectives,
small manner adverbs)
Argument encoding: SENT / Ø - SENT / Ø (gapping / co-referentiality)
Examples:
Ref Head:
Wŏ pànwàng [nĭ kuài yidiăn bìyè]
I hope you soon a.little graduate
‘I hope you’ll graduate a bit sooner.’ (Li & Thompson 1981: 599)
Ref Mod:
wŏ măi-le yi-jiàn yīfu [tài dà]
I buy-perf one-cl outfit too big
‘I bought an outfit that turned out to be too big.’ (Li & Thompson 1981: 614)

tā yŏu yi-ge mèimei [hĕn xĭhuān kàn diànyĭng]
3sg exist one-cl younger.sister very like see movie
‘S / he has a younger sister who like to see movies.’ (Li & Thompson 1981: 611)
Pred Mod:
Tāmen [yòng shŏu] chī-fàn
they use hand eat-food
‘they eat with their hands.’ (Li & Thompson 1981: 597)
Clause + de
Functional distribution: Flex: Ref Mod, Pred Mod ≠ PoS (small adjectives, small manner adverbs)
Examples:
Ref Mod:
[Jīntiān yíng de] qián fù fang-zū
today win conn money pay house-rent
‘The money we won today goes to pay the rent.’ (Li & Thompson 1981: 581)
Pred Mod:
nĭ păo [de hĕn kuài]
you run conn very quick
‘You run very quickly.’ (Li & Thompson 1981: 625)
Pred-zhe (DUR)
Remarks: The durative aspect marker -zhe can be used in the first of two clauses to signal that one event
provides a durative background for another event.
Since this construction occurs only with (unexpressed) co-referential subjects, it is analyzed as a Ø-SENT
construction, even though there is no independent evidence for deranking (since there is no verbal
inflectional morphology).
Functional distribution: Rig: Pred Mod = PoS (small manner adverbs)
Examples:
Pred Mod:
Xiăo gŏu [yáo-zhe wěiba] păo le
small dog shake-dur tail run crs
‘The small dog ran away wagging its tail.’ (Li & Thompson 1981: 223)
tā [guāng-zhe] jiăo shàng-kè

3sg bare-dur foot ascend-class
‘S / he goes to class barefooted.’ (Li & Thompson 1981: 23)
Tamil
Pred-atu
Remark: With locative case or postposition poola ‘like’ the construction can be used in Pred Mod function.
Functional distribution: Rig: Ref Head (+ case / postposition also Pred Mod: simultaneity and similative)
= PoS (nouns)
Verbal categories: Agreement is lost; tense is retained; some but not all aspect and mood distinctions
are retained.
Examples:
Ref Head:
[tiruʈan nakeyellaam eʈuttukkiʈʈirukkarate] naan patteen
thief jewellery-all take-progr-prs-nmlz:acc I see-pst-1sg
‘I saw the thief taking the jewels.’ (Asher 1982: 20)

Pred Mod:
Koȥanteŋka [biic-ile maɳalviiʈu kattar-atile] curucuruppaa iruntaaŋka
child-pl beach-loc sand-house build.prs-nmlz:loc busy be.pst.3pr
‘The children were busy building a sand-house on the beach.’ (Asher 1982: 21)
[[Eŋka ammaa kaappi pooʈar-atu] poola]-vee en mandevi pooʈaraa

our mother cofee put:pres-nmlz like-emph my wife put-prs-3sg.fem
‘My wife makes coffee just as my mother makes it.’ (Asher 1982: 48)
Pred-(kk)a
Verbal categories: Agreement is lost; tense is retained.
Example:
Ref Head:
Raaman [avaru viiʈʈukku enne [var-a] connaaru
Raman his house.dat i.acc come-inf say.pst.3sg
‘Raman told me to come to his house.’ (Asher 1982: 22)
Pred-a (PTC):
Verbal categories: Agreement is lost; TAM is retained.
Examples:
Ref Mod:
[vaɳɳaane aticc-a] taccan
carpenter-acc beat.pst-ptc washerman
‘the washerman who beat the carpenter’ (Asher 1982: 28)
[taccan aticc-a] vaɳɳaan

carpenter beat.pst-ptc washerman
‘the washerman whom the carpenter beat’ (Asher 1982: 28)
[taccan paɳam kuʈutt-a] vaɳɳaan

carpenter money give.pst-ptc washerman
‘the washerman to whom the carpenter gave money’ (Asher 1982: 228)
[akkaa taŋkaccikki caata pooʈʈ-a] karanʈi

elder.sister younger.sister-dat rice put.pst-ptc spoon
‘the spoon with which elder sister gave rice to younger sister’ (Asher 1982: 228)
Kisi
((m)àà) + clause
Remark: The complementizer can be omitted.
Examples:
Ref Head:
À dìmì sùɛ̀í kéŋ làcòò wá lé [àà wànà nùmndó
they say matter that lie cop neg comp person your
‘They say and it’s no lie, that your people are your people.’ (Childs 1995: 278)
ò cà kó bɛ́ɛ́ [màà tè ò cǎŋ pɛ́ ò fùlá lènǐŋ]

he see only indeed comp if he cry if he go.out inside
‘He really thinks that if he cries he will be free.’ (Childs 1995: 280)
Á dímúl ndú [Ø ò hìŋ wɔ̀ cóŋcóŋ]

you tell him he come prt ideoph
‘You should tell him (that) he should come quickly!’ (Childs 1995: 281)

Unmarked deranked clause:
Examples:
Ref Head:
í ló [kìsìé pɛ̀ɛ̀kùò].
I stay Kisi study
‘I continued to study Kisi.’ (Childs 1995: 281)
ò dèmál [sàá wàlló tòsàl] kpɔ̂ŋ

he fail Saa work do ideoph
‘He failed completely to do the work for Saa.’ (Childs 1995: 281)
Noun class pronoun + clause (+ RSP) + noun class suffix

Remarks: The head noun loses its noun class suffix, which is supplanted by a noun class pronoun (unless
the noun belongs to the o-class) and the noun class suffix appears at the end of the relative clause. The
relativized item is gapped, but when the syntactic role of the head noun is oblique, or when the head noun
is distant, a resumptive pronoun can appear in the relative clause.
Examples:
Ref Mod:
mɛ̀ŋ mà [ó kól]-áŋ mà nǐŋ lé
water pron he drink-suff pron good neg
‘The water that he drank wasn’t good.’ (Childs 1995: 286)
When the head noun belongs to the o-class, there is no pronoun, only a suffix:
Wàná [tóóf cíóóŋ]-ó ò cò lɛ̀ hùnɔ̀ɔ̀

person inspect towns-suff pron aux again come
‘The person who inspects towns will return.’ (Childs 1995: 286)
wálí [ŋ̀ cò hóò tòsàá]-ó tàmbà ndóó ké núm tòòlúlá

work you aux rsp do-suff Tamba he-ipfv give you support
‘Did Tamba give you any help with this work that you were doing?’ (Childs 1995: 288)
Yàú wílɛ̀ [yá póóɲál núm kóŋ ó pùùlùéí nǐŋ á ó Kìsìé nǐŋ]-ó í tósá
letter long [I write you that to English in and to Kisi in]-suff I make
háwɛ̀íyó ò pìlɛ́ kɛ́ɛ̀tɛ̀tɛ̀ lé ndú cîɩ pòòɲìáá
hour pron one ideoph for rsp finish write
‘The long letter that I wrote to you in English and Kisi, it took me a whole hour to finish writing it.’
(Childs 1995: 290)
Nung
Unmarked clause
Functional distribution: Flex: Ref Head, Ref Mod, Pred Mod ≠ PoS (nouns, small adjectives, no adverbs)
Examples:
Ref Head:
Mu’hng [càng slóng vam] đày lēo
you speak two sentences be.able all
‘You can speak both sentences.’ (Saul & Wilson 1980: 55)
lēo mu’hn ca tẹo slò’ng [heht vú]

then he about again desire do king
‘Then he wanted to be king again.’ (Saul & Wilson 1980: 52)

Lēo hảhn mạhn [kíhm mi lái]
then see pl gold be much
‘Then they saw that there was much gold.’ (Saul & Wilson 1980: 111)
Mu’hn vạ [pị bạo mu’hn ma]

he say older brother his come
‘He said that his older brother came.’ (Saul & Wilson 1980: 115)
Ref Mod:
mu’hn tẹo pển pehn tú đéhc-sláo [đáy-sláo lái]
she again become like cl child-girl good-girl much
‘She then became a very pretty girl.’ (Saul & Wilson 1980: 33)
Pred Mod:
Tú mā páy [chũhm lái]
cl horse go slow very
‘The horse goes very slowly.’ (Saul & Wilson 1980: 95)
Mu’hn càng cồ [heng lái] cả vahn

He speak story strong very all day
‘He spoke loudly all day.’ (Saul & Wilson 1980: 95)
(REL) + gapped clause + DEM / FOC

Remarks: Relative clauses are formed with a demonstrative or a focus marker or both added to
the end of the clause and / or a relativizer at the beginning of the clause. When the function of
subject / source / locative-directional is relativized, the relativizer is optional. When the object is relativized,
the DEM / FOC marking is optional. When the indirect object / beneficiary are relativized, the marker hù’
‘to, for’ is added at the end of the relative clause. With all types of relative clauses, an anaphoric mu’hn is
often inserted before the main predicate.
Examples:
Ref Mod:
Lēo vang [khi bề tế] chihng áu tú bề hù’ kề Hong [khi͈ mā tế]
Then boy ride goat dem then take cl goat give man Hong ride horse dem
‘Then that boy riding the goat gave the goat to that man Hongh riding the horse.’ (Saul & Wilson 1980: 16)
kề [tị-vạ tóhc cộn thihn tề ni]

man rel lost cl stone dem foc
‘The man who had lost the stone.’ (Saul & Wilson 1980: 78)
Áhn ho’n [cáu-ní-ọc-khọi ni]

cl house I.run.out dem
‘The house (that) I was running from.’ (Saul & Wilson 1980: 80)
kề [tị-vạ cáu áu sẹc slú’ hù’ ni] mu’hn

man rel I.gave cl book to foc he
‘The man I have the book to, he ….’ (Saul & Wilson 1980: 79)
Óhng kề [cáu slú’ áhn tu hù’ ni]

cl man I.bought cl hat for foc
‘The man for whom I bought the hat.’ (Saul & Wilson 1980: 79)
bạt + clause
Remark: This construction is a simultaneity clause.
The subject is expressed in both the matrix and the dependent clause.
Argument encoding: SENT - SENT / Ø

Example:
Pred Mod:
Lēo [mu’hn bạt hảhn cáh slú’] mu’hn láo lái
then he when see cl tiger he fear much
‘Then when he saw the tiger he was very afraid.’ (Saul & Wilson 1980: 110).
Garo
Clause + in-e
Examples:
Ref Head:
[Pir-ga-cha song ni-to-a in-e] ang-a chanchi-a
Pirgacha village beautiful-neut comp I-nom think-neut
‘I think Pirgacha village is beautiful.’ (Burling 2004: 318)
[Bi-a sok-ba-ku-ja in-e] ang-a chanchia

he-nom arrive-here-not-yet comp I-nom believe
‘I believe that he has not yet arrived here.’ (Burling 2004: 319)
Pred-a
Remark: With the suffix git-a ‘like’, this construction can be used as a similative clause.
Functional distribution: Rig: Ref Head (+ extra suffix also similative) = PoS (nouns)
Verbal categories: No tense / aspect
Examples:
Ref Head:
ang-a [bi-ni giit ring͘ -a]-ko kin-a-a
I-nom he-gen song sing-nmlz-acc hear-neut
‘I hear him singing songs.’ (Burling 2004: 294)
Similative:
Jal dong-kan, [bi-ni ha͘ -sik-a-gita].
whatever be.at-imp he-gen wish-nmlz-like
‘Let (him) be according to his wishes.’ (Burling 2004: 297)
Pred-a-ni
Remark: Even though the suffix -ni is homophonouns with the genitive case marker, the construction
takes (a second) case-marker according to function.
Argument encoding: Ø - Ø?
Example:
Ref Head:
[agan-a-ni]-ko seng-ja, an͘ -tang dra-emina kam-ko dak-a
talk-nmlz-gen-acc wait-neg own force-adv work-acc do-neut
‘Not waiting for talk (instructions), (he) does the work forcefully himself.’ (Burling 2004: 296)
Pred-na (INF) (+ in-e)

Remark: Used as the same-subject complement of modals, desideratives and predicates of achievement and
emotion. With achievement and emotion predicates, the complementizer in-e can optionally be added.
Argument encoding: Ø-SENT (co-referentiality)

Examples:
Ref Head:
Ang-a [mi cha͘ -na] man͘ -a
I-nom rice eat-inf can-neut
‘I can eat rice.’ (Burling 2004: 320)
Bi-a [kat-ang-na] ha͘ -sik-ing-a

he-nom run-away-inf want-progr-neut
‘He wants to run away.’ (Burling 2004: 320)
Ang-a [bik-ko nam-et-na (in-e)] ches-ta dak-no-a

I-nom you-acc good-caus-inf comp try-fut
‘I wil try to improve you.’ (Burling 2004: 321)
Ang-a [nang͘ -ko nik-na (in-e)] ken-a

I-nom you-acc look.at-inf comp fear-neut
‘I an afraid to look at you.’ (Burling 2004: 321)
Pred-kan / ka-na (IMP-INF) (+ in-e)

Remark: -ka is the 3rd person imperative suffix, used here as a subordinating suffix with manipulative
predicates. The complementizier in-e is optional with this construction.
Example:
Ref Head:
Nang͘ -ko [boi bre-kan-na (in-e)] ang-a hit-a
you-acc book buy-comp comp I-nom order
‘I order you to buy a book.’ (Burling 2004: 322)
Pred-gip-a
Functional distribution: Rig: Ref Mod ≠ PoS (no lexical strategy)
Argument encoding: Ø / POSS - Ø / SENT (gapping)
Examples:
Ref Mod:
[nok-o pi͘ -sa-ko nik-gip-a] me-tra
‘the woman who saw the child at the house.’ (Burling 2004: 301)
[me͘ chik-ni skang-o den͘ -gip-a] a͘ -bol

women-gen previously-loc cut-ptc firewood
‘firewood that the women chopped previously’ (Burling 2004: 299)
Pred-e / -e-ming / -e-min-a

Example:
Pred Mod:
Na͘ -tok-ko [nam-e] [ni-e] cha͘ -ja-od-de, (…)
fish-acc good-conv watch-conv eat-neg-if
‘If you don’t watch out well when eating the fish, (…)’ (Burling 2004: 314)
lit.: ‘If you don’t eat the fish in a good and watchful manner, ….’
[Klemen-ko rim-jol-e] ha͘ ti-cha re͘ -ang-a

Clement-acc take-along-conv market-loc go-neut
‘(I) go to the market, taking Clement along.’ (Burling 2004: 315)

Krongo
àní (QUOT) - tíŋ + clause
Remark: This construction is used only with utterance verbs.
The quotatative marker is part of the matrix clause; the particle tíŋ is suffixed to the first word of the
dependent clause.
Example:
Ref Head:
ŋ-áa Lúwáalà ótó-ŋ ànáày àní
conn:masc-cop Luwaala inf:say-tr dat:3.pl quot
[k-óc-ìiní-ttíŋ àay fókkìrì ŋ-àdéelá]
pl-pst-make-prt you.pl plan conn:masc-ipfv:be.good
‘And Luwaala says to them, that they have made a good plan.’ (Reh 1985: 354)
(t-)Pred (“nominalization/infinitive”)
Remark: For class II verbs, the nominalization and the infinitive have the same form; for class I verbs, the
infinitive is formed without the prefix.
The construction (NMLZ and INF) takes case marking when not in object function (the accusative is
zero-marked). Infinitives in object function with subject-to-object-raising are locative case-marked. When
an overt subject is a full noun, the subject takes a possessive prefix. If not, it is expressed by a dependent
person suffix.
Verbal categories: No agreement / mood, tense, aspect and valency / voice are retained
Examples:
Ref Head:
n-átàasà àʔàŋ [t-óshó-ókò-n-tú ɲáamà àʔàŋ]
1 / 2-impf:want I nmlz-ipfv: cook-ben-tr-2.sg(dep) thing dat:i
‘I wanted you to cook something for me.’ (I wanted your cooking something for me.’) (Reh 1985: 333)
n-átàasà àʔàŋ [t-úmúnó kà- Sárrà àʔàŋ]

1 / 2-ipfv:want I nmlz-ipfv:help poss-Sarah me
‘I want Sarah to help me.’ (lit. I wanted Sarah’s helping me.’) (Reh 1985: 333)
Ácáamà-ŋ káaw àʔàŋ [má-t-ákká-ttɪ́nɪ ́ kí-cáaw à-Tábáaɲ]

masc:ipfv:speak-ben person dat:i gen-nmlz-fut-3sg(dep) loc-inf:go dat-Tabanya
‘The man says to me that he will go to Tabanya.’ (lit. The man speaks me of his (future) going to
Tabanya.’) (Reh 1985: 334)
n-átàasà àʔàŋ [àkù úudà] n-átàasà àʔàŋ

1 / 2-impf.want I inf:eat meat 1 / 2-impf.want I
‘I want to eat meat.’ (Reh 1985: 335)
ùʔùŋ [k-áaláaná àʔàŋ kí-nìinò mó-di]

you(acc) loc-inf:teach me loc-language gen-homeland
‘I want you to learn me Krongo.’ (Reh 1985: 335)
CONN-Pred (+ RSP)
Remarks: The construction is marked by a connector-prefix, which is fused with morphemes encoding
agreement.
In Ref Mod function, if the relativized item is not the subject of the DC, an anaphoric pronoun is used,
which takes case according to function.
In Pred Mod function, the DC obligatorily has imperfective aspect.
Functional distribution: Flex: Ref Mod, Pred Mod ≠ PoS (no adjectives, small mAdverbs)
Verbal categories: No agreement / mood, tense, aspect and valency / voice are retained.
Nominal categories: Gender and number agreement (+ case agreement of RSP)

Examples:
Ref Mod:
káaw [m-ísò]
person conn:fem-ipfv.walk
‘the woman, who walks’, ‘the walking woman’ (Reh 1985)
n-úllà àʔàŋ kí-ǹt-àndiŋ [n-úufò-ŋ kò-nìimò kàti]

1 / 2-ipfv:love I loc-sg-clothes conn:neut-ipfv:sew-tr poss-mother my
‘I love the dress that my mother is sewing.’ (Reh 1985:256)
Káaw [m-àasàlàa-tɪ́ àakù]

person conn:fem-ipfv:look.at-1sg(dep) she.fem
‘the woman that I looked at.’ (Reh 1985)
nóoni àʔàŋ káaw [ŋ-áfàrà kò-níimò kàtɪ́ m-ɪ ʔ́ ɪ ŋ̀ ]

know I person conn-ipfv:cry poss-mother my gen-he
‘I know the man about whom my mother cries.’ (Reh 1985: 258)
Pred Mod:
n-óocó-óní àʔàŋ [ŋ-ásàŋ kí-tùlùnkwáaná]
1 / 2-impf:laugh-dtr I conn-ipfv:neg.have loc-joy
‘I laugh joylessly.’ (Reh 1985: 302)
n-áa t-ánkwà-ànì [n-úrùná-ŋ úuní kànáày]

conn-neut-cop inf-go.round-dtr conn:neut-ipfv:pay.attention.to footprints poss:3.pl
‘She goes round, paying attention to their footprints.’ (Reh 1985: 333)
ŋ-áa árící ádìyà kítáccɪ̀-mày [ɲ-íisò túkkúrú kúbú]

conn:masc-cop man inf:come there-ref conn:masc-ipfv-walk adv:with.lowered.head
‘The man comes right there walking with a lowered head.’ (Reh 1985: 345)
Hixkaryana
Pred-nɨ / -thɨ- / -hɨto + - rɨ
Remarks: The construction is formed with the nominalizer-nɨ-, followed by the inflectional suffix -rɨ ‘possessed
item’, glossed as POSSD. For past tense, -nɨ- is replaced by -thɨ-; for negation, the form -hɨto- is used.
The arguments are marked ergatively: the first argument of a transitive verb (A) is marked by a
postpositional phrase with wya (‘to, by’, normally the indirect object marker). The only argument of
an intransitive (S) and the second argument of a transitive verb (P) are marked as possessives. Since
Hixkaryana is head-marking, the possessive appears on the predicate.
In combination with extra elements, the nominalization construction is used in adverbial functions.
The relevant elements are the “de-nominalizing relator” me for manner, and the postposition wyaro ‘like’
for simile constructions.
Functional distribution: Rig: Ref Head (+ extra element me, wyaro ‘like’ also Pred Mod) = PoS (nouns)
Verbal categories: Aspect is lost; 3 out of 7 tense distinctions (non-past, simple past, remote past) are
retained.
Nominal categories: Possessive prefixes
Argument encoding: POSS - OBL
Examples:
Ref Head:
ro-wanota-nɨ-rɨ
1.poss-sing-nmlz-possd
‘my singing’ (Derbyshire 1979: 165)
Amna-y-omokɨ--thɨ-rɨ
1-3.poss-come-pst.nmlz-possd
‘our coming (in the past)’ (Derbyshire 1979: 165)
ɨ-nɨkɨ-hto-rɨ-komo
3.poss-sleep-neg.nmlz-possd-coll
‘their not going to sleep’ (Derbyshire 1979: 166)
Karyhe kamara Ø-to-thɨ-rɨ

quickly jaguar 3.poss-go-pst.nmlz-possd
‘the going quickly of the jaguar.’ (Derbyshire 1979: 23)

oxhe nɨ-hananɨh-nɨ-rɨ komo wosɨ wya
well 3.poss-teach-nmlz-possd coll woman by
‘the good teaching of them by the woman.’ (Derbyshire 1979: 24)
Koseryehyaha, [r-ehurka-nɨ-rɨ] hona

I.am.afraid, 1.poss-fall-nmlz-possd to
‘I am afraid of falling.’ (lit. I am afraid to my falling.’) (Derbyshire 1979: 24)
Pred Mod:
[Teryweryero ro-wanota-nɨ me] kewehyaha
loudly my-singing-nmlz dnmlz I.take.a.bath
‘I take a bath singing loudly.’ (Derbyshire 1979: 28)
Uro ehtxemako, [owhya Waraka yohtxeman-tɨ-rɨ wyaro]

me treat.with.medicine by.you Waraka treat-pst.nmlz-possd like
‘Treat me with medicine just like you treat Waraka.’ (Derbyshire 1979: 31)
Slave
Clause + nį / Ø
With the verbs ‘want’ and ‘say’ the complementizer can be omitted.
The construction is used in Ref Mod function when the relativized item expresses a time or location.
Functional distribution: Flex: Ref Head, Ref Mod ≠ PoS (nouns, no adjectives)
Examples:
Ref Head:
[Ɂelá táhɫa į] kodeyihshá yíle
boat 3.is comp 1sg.knew neg
‘I didn’t know that the boat came in.’ (Rice 1989: 1245)
metá [Ɂekǫ́ Ɂahndeh gha] ndi

3.father there 1sg.go FUT 3.say
‘His dad said that he is going there.’ (Rice 1989: 1224)
Ref Mod:
hejǫ sįa [Jim Ɂeghálayeda nį] sį ́ Ɂ agǫt’e
here probably Jim 3.work rel foc area.exist
‘This must be the place where Jim works.’ (Rice 1989: 1317)
Clause + gha / gú
Examples:
Ref Head:
[líbalá líts’eɁa gha] kegǫ́fa
canvas one.folds comp is.difficult
‘It is difficult for me to fold canvas.’ (Rice 1989: 1246)
[sah hehlį gú] Ɂedeghǫ náwitį

bear 1sg.be comp refl.about 1sg.dreamed
‘I dreamed that I was a bear.’ (Rice 1989: 1246)
Clause + i / sį́i / líi

Functional distribution: Rig: Ref Mod ≠ PoS (no adjectives)
Ɂelá [hįshá i]
boat 3.is.big rel
‘the big boat’ (Rice 1989: 1309)

Ɂeyi dene [Ɂįts’é whehk’é sį́i]
the person moose 3.shot rel
‘the person who shot the moose’ (Rice 1989: 1313)
ts’élįa [thetį líi]

girl 3sg.sleep rel
‘the girl who is sleeping’ (Rice 1989: 1309)
Clause + gháré
Remark: The meaning of the adverbial subordinator is instrumental ‘by means of ’.
Functional distribution: Rig: Pred Mod ≠ PoS (no manner adverbs)
Example:
Pred Mod:
dene [gháts’eyeda gháré] yá gáɁets’ederetę
person one.watches adv top one.teaches.refl
‘One learns through watching.’ (1063)
Nivkh
Pred-vut / vur (‘retelling converb’):
Used only for the complements of utterance predicates.
The allomorphy is conditioned by person agreement: -t for 1st person (SG / PL), -r for 2nd or 3rd person.
Example:
Ref Head:
If [imņ-aχ als p’e-ny-vur] it-t’
He they-dat / acc berry pick-fut-comp say-fin
‘He said [that] they would pick berries.’ (Gruzdeva 1998: 57)
Unmarked clause
Argument encoding: SENT / Ø - SENT (co-referentiality)
Examples:
Ref Head:
Řaņķ [čo n’I-d] esķa-d
woman fish eat-fin dislike-fin ‘
The woman does not like to eat fish. (Gruzdeva 1998: 49)
[Parf n’aķi p’řy-i-d] n’I j-ajm-d

evening I-elder.brother come-fut-fin I obj-know-fin
‘I knew (that) my elder brother would come in the evening.’ (Gruzdeva 1998: 49)
Pred without FIN SUFFIX (‘participles’)

Functional distribution: Rig: Ref Mod ≠ PoS (no adjectives)
Verbal categories: Retained except for FIN suffix
Example:
Ref Mod:
[N’I zosķ] t’aķo tyr t’xy p’i-d’.
I break:ptc knife table on be-fin
‘The knife which I have broken is on the table.’ (Gruzdeva 1998: 50)

Pred -r / -t-ř / / -n
Remark: The allomorphy is triggered by person agreement: -t for 1st person (SG / PL) subject / agent, and -r
for 2nd or 3rd person. The other variants are dialectological.
Functional distribution: Rig: Pred Mod ≠ PoS (no manner adverb)
Verbal categories: No tense / agreement, aspect and mood can be retained
Examples:
Pred Mod:
Haimņař t’axkyř [n’aχ-kis nloņbloņ-d’i-ķavr-r] t’yr-d
old.man straight eye-instr blink-prt-neg-conv look-fin
‘The old man looked straight, not blinking [his] eyes.’ (Gruzdeva 1998: 54)
hoʁor [eʁ-gu-r] məγ-d̦

therefore fast-caus-conv:2 / 3SG descend-ind
‘Therefore he descended quickly.’ (Matissen & Drossard 1998: 44)
WEST GREENLANDIC
Pred-niq
Remark: With intransitive predicates, the S takes relational case (RELC) and triggers possessive
agreement. With transitive predicates, there are several possibilities. When only the P is present, the
dependent predicate can be unmarked or marked with the passive suffix. The P appears in relative case
and triggers possessive agreement. When both arguments are present, the dependent predicate is marked
with a semi-transitivizing affix (SEMTR). The A appears in the relative case, and the predicate bears the
corresponding possessive suffix, while the P is in the instrumental. Thus, the possessive argument in the
nominalization always corresponds to the absolutive argument in the corresponding finite clause 12.
Functional distribution: Rig: Ref Head = PoS: nouns
Verbal categories: No mood, person / number agreement. Tense can be retained, but is usually not
expressed.
Nominal categories: Case, nominal agreement
Argument encoding: RELC / POSS - Ø / INSTR
Examples:
Ref Head:
[umiarsu-up qassi-nut tikin-ni-ssa-a] nalunngil-ara
ship-rel how.many-all arrive-nmlz-fut-3sg.poss know-1sg.3sg:ind
‘I know when the ship will arrive.’ (Fortescue 1984: 115)
Angutip tuqunnira
man:relc kill:nmlz:3sg.poss
‘the killing of the man’ (i.e. he is killed) (Fortescue 1984: 46)
nalu-aa [qinnuta-ata qanuq naammassi-niqar-ni-ssa-a]

not.know-3sg:3sg.indic request-3sg.poss.relc how implement-pass-nmlz-fut-3sg.poss
‘She didn’t know how his request would be implemented.’ (Fortescue 1984: 45)
[ikinngum-mi-nik tuqut-si-nir-a] tusar-para

friend-his.refl-instr kill-semtr-nmlz-3sg.poss hear-1sg.3pl.ind
‘I heard of his killing his friend.’ (Fortescue 1984: 45)
Pred-ta / sa (‘passive participle’):

Remark: Overt subjects are in relative or ablative case, non-gapped objects remain SENT.
Functional distribution: Rig: Ref Mod ≠ PoS (no lexical strategy)
Verbal categories: No tense, mood, verbal agreement.
Nominal categories: Nominal agreement (number and case)
Argument encoding: RELC / ABL / Ø - SENT / Ø (gapping)
Examples:
Ref Mod:
angum-mut [ippassaq naapi-ta-n]-nut tunniup-para
man-all.sg yesterday meet-pass.ptc-1sg.poss-all.sg give-1sg.3sg.ind
‘I gave it to the man I met yesterday.’ (Fortescue 1984: 49)
12
Strictly speaking, therefore, West Greenlandic has no transitive nominalizations. (see Koptjevskaja-Tamm 1993).

angut [aallaasa-a tigu-sa-ni] nigur-niar-paa
man gun-his take-pass.ptc-3poss.refl avoid-try-3sg.3sg.ind
‘He tried to avoid the man whose gun he had taken.’ (Fortescue 1984: 53)
Participial mood clause

Remarks: The participial mood endings are parallel to the indicative ones, but built up on mood-marker su
in the intransitive and gi in the transitive.
In Ref Head function, the construction is used for different-subject complements. When it functions as
such, the dependent clause is morphologically marked as an object.
In Ref Mod function, the construction must be intransitive and can only take 3rd person inflection. Relative
clauses that modify the subject of the matrix clause agree in (relative) case with their head.
When marked for instrumental case the participial mood can also be used in Pred Mod function.
Functional distribution: Flex: Ref Head, Ref Mod, + CASE also Pred Mod ≠ PoS (nouns, no lexical
strategy for modifier functions)
Nominal categories: Nominal agreement in Ref Mod function
Examples:
Ref Head:
Ilisima-vaa [urni-ssa-giga]
know-3sg.3sg.ind come.to-fut-1sg.3sg.ptc
‘He1 knew I would come to him2.’ (Fortescue 1984: 36)
paasi-nngil-luinnar-para [ilaa-juma-sutit]
understand-not-completely-1sg:3sg:ind come.along-wany-2sg.ptc
‘I didn’t understand at all that you wanted to come along.’ (Fortescue 1984: 36)
Ref Mod:
Niviarsiaq [kalaallisut ilinnia-lir-suq]
girl Greenlandic learn-begin-intr.ptc
‘the / a girl who has begun learning Greenlandic.’ (Fortescue 1984: 49)
Pred Mod (with case):

[Kulturi-kkut sunniuti-qar-luar-tu]-mik
culture-pros effect-have-well-intr.ptc-instr
‘having considerable cultural effect’ (Fortescue 1984: 56)
Contemporative mood clause

Remark: The contemporative or conjunctive mood for same-subject clauses. The contemporative is formed
with the mood-marker lu / llu followed by person markers (1st, 2nd, 4th) or 3rd person object markers.
Functional distribution: Flex: Ref Head, Pred Mod ≠ PoS (nouns, no lexical strategy for Pred Mod function)
Nominal categories: Nominal agreement in Ref Mod function
Argument encoding: SENT / Ø - SENT
Examples:
Ref Head:
Uqar-sinnaa-vunga [tama-asa uuqattaar-sima-llugit]
say-can-1sg.ind all-3pl try-pfv-1sg-3pl.cont
‘I can say that I have tried them all.’ (Fortescue 1984: 40)
Pred Mod:
[Irnir-lunga] isir-punga
do.hurriedly-1sg.cont go.in-1sg.ind
‘He entered quickly / in a hurry.’ (Fortescue 1984: 55)
[uqa-nngua-rani] ini-mi-nut isir-puq

say-little-4sg.neg.cont room-his.refl go.in-3sg.ind
‘He entered his room without even speaking.’ (Fortescue 1984: 62)

Summary in English
Linguistic typology is concerned with cross-linguistic comparison. The basic

problem encountered by typologists is that, in order to compare language-
specific structures, a tertium comparationis is needed that is independent
of structure. Therefore, typologists choose a particular functional domain,
and then examine the formal structures that languages use to express
certain distinctions within that domain. When they have identified the
range of cross-linguistic formal variation, typologists search for universals:
dependency relations between certain structural traits that apply to all human
languages. Within the functionalist approach to typology, adhered to in the
present study, it is assumed that such universals can ultimately be explained
in terms of the function of languages, namely to encode communication
(Croft 2003).
Three basic communicative or propositional functions may be
distinguished: Reference, predication, and modification. To refer means to
identify a referent, i.e. an entity that one wants to talk about. To predicate
is to ascribe something to a referent, i.e. to report that this referent is
involved in some state of affairs. To modify is to enrich either reference or
predication, by means of expressing an additional feature of either a referent
or a predication (Searle 1969, Croft 2001: 66).
The basic propositional functions of reference, predication, and
modification may be expressed by linguistic forms or constructions of
various types. Example (1) shows that a referential expression such as the
453
object argument of the predicate saw can take the form of a simple lexical
construction (a car in (1a)), or of a more complex clause-like construction
(that John has bought a new car in (1b)):
(1) a. I saw a car.

b. I saw that John has bought a new car.
This study is concerned with the two construction types illustrated in

(1): lexical constructions or parts of speech and clause-like constructions or
dependent clauses.
Parts of Speech (henceforth PoS) is the traditional term used to refer to
the major classes of lexemes or content words that are distinguished in a
particular language (Schachter & Shopen 2007: 1). Typically, the class of
nouns is used to express the function of reference (cf. car in example (2a));
verbs are used for predication (cf. wash in (2b)); adjectives for modification
of referential expressions (cf. red in (2c)); and adverbs for modification of
predications (cf. quickly in (2d)).
(2) a. I wash the car.

b. I wash the car.
c. I wash the red car.
d. I quickly wash the car.
However, not all languages follow this particular pattern, in which every
PoS class is dedicated to the expression of a single function. Alternatively,
languages may display PoS classifications in which two or more functions
can be expressed by members of the same class. In Turkish, for instance,
there is one class of lexemes that may be used to express a referential unit (cf.
example (3a) below), or to modify a referential or a predicative expression (cf.
examples (3b) and (3c), respectively). The translations of examples (3b) and
(3c) show that in English the use of the lexical noun beautiful in modifying
functions requires derivation with the suffixes -ful and -ly.

(3) a. güzel-im
beauty-1poss
‘my beauty’

b. güzel bir kopek
beauty art dog
c. güzel konuştu
beauty s / he.spoke
‘S / he spoke beautifully’
Adopting a terminology proposed by Hengeveld (1992), I call parts of

speech that are functionally specialized rigid, and those that can express
multiple functions flexible.
As regards dependent clauses (henceforth DCs), three types are
traditionally distinguished: A complement clause functions as a referential
unit, expressing an arguments of the main clause predicate (cf. (1b) above);
A relative clause functions as a modifier of a main clause argument; and
an adverbial clause functions as a modifier of the main clause predicate
(Thompson et al. 2007: 238).
Like PoS classes however, DCs can be rigid, i.e. specialized for the
expression of a single function, or flexible, i.e. able to express more than
one function. For instance, the DC construction marked by do in the
West-Papuan language Abun can only be used in referential function,
i.e. as a complement clause (cf. example (4)). In contrast, the Basque DC
construction marked by -en can be used in both referential and modifying
function, i.e. as a complement clause and as a relative clause (cf. examples
(5a) and (5b), respectively).
Abun (Berry & Berry 1999: 167)

(4) Ji nut [do an bi obat ne nggi wa
1sg think comp 3sg poss magic det strong for
be an gu hi ré.]
later 3sg kill 1sg this
‘I thought that his magic was strong [enough] for him to kill me.’ (167)

knew Mikel late arrive.fut aux-comp-det
Summary in English | 455

Apart from their possibilities with regard to the expression of propositional

function(s), DCs – being complex constructions – can also be studied from
the perspective of their internal morpho-syntactic properties. Usually,
these properties are described in comparison to the properties of ordinary
independent clauses in the particular language under analysis. Consider the
difference between examples (6a) and (6b):

The dependent clause in (6a), marked with the subordinator that, expresses
future tense and the subject Paul is encoded in the same way as it would be
in an independent clause. This type of dependent clause, which structurally
resembles an independent clause, is called balanced. In contrast, the dependent
clause in (6b) is marked by the special -ing form of the predicate, it cannot
express tense, and it has a subject that is encoded as a possessor (Paul’s).
This second type of dependent clause, the structure of which deviates to a
certain extent from the structure of an independent clause, is called deranked
(Stassen 1985).
The aim of the present research is to investigate the relationship between
the functional flexibility or rigidity as displayed by a particular language’s PoS
classes on the one hand, and by its balanced / deranked DC constructions on
the other hand. More specifically, this study tries to discover to what extent
it is possible to predict, on the basis of a language’s PoS system, what the
functional potential of its DC constructions will be. This goal is approached
by means of an investigation of the PoS classes and DC constructions
of a genealogically and geographically balanced sample of 50 languages
(see Appendix i).
The first part of this book (Chapters 2 – 4) provides the theoretical and
methodological background of the study. Chapter 2 outlines a number of
theoretical issues related to the definition and comparison of PoS classes
within and across languages. It starts out with a characterization of possible
approaches to PoS typology in terms of the type of criteria on which PoS
class definitions are primarily based, distinguishing between semantic,

morphological, syntactic, and pragmatic criteria (Givón 2001: 49; Sasse
1993b:196).
Chapter 2 continues with a detailed discussion of two (relatively) recent
approaches to PoS typology: Hengeveld’s pragmatic-syntactic theory of
parts of speech and Croft’s universal-typological theory of parts of speech.
The former defines PoS exclusively in terms of the propositional function(s)
that they can express (Hengeveld 1992, Hengeveld et al. 2004, Hengeveld
& Van Lier 2008, 2009). The Hengeveldian model defines a functional
space in terms of two interacting dimensions. The first dimension involves
the distinction between predication and reference; the second concerns the
opposition between heads and modifiers. As regards the first parameter, the
function of predication is assumed to be privileged in relation to the function
of reference, since the communicative act of referring presupposes that
something is predicated of a referent. With respect to the second parameter,
the assumption is that obligatory heads are privileged in relation to optional
modifiers. In addition, these two hierarchical parameters are ranked with
respect to one another: Since all languages apparently have a predication-
reference distinction at some level of the grammar, while the function of
modification does not seems to be relevant in the grammar of all languages,
it is expected that the predication-reference parameter is primary in relation
to the head-modifier parameter. On the basis of this hierarchical model, a
number of predictions are made concerning possible and impossible types
of PoS classifications, in terms of the type and amount of propositional
functions that they can express.
The Hengeveldian approach to PoS typology can thus be characterized
as taking into account syntactic-pragmatic criteria only. This narrow approach
has been criticized by Croft (2000, 2001, 2005), who defines PoS in terms
of a broader set of criteria, including semantic meaning and the ability to
express certain morphological categories. In particular, Croft defines PoS
classes as prototypical combinations of semantic classes and propositional
functions: A prototypical noun denotes an object and functions referentially;
a prototypical adjective denotes a property and functions as a modifier; and
a prototypical verb denotes an action and functions predicatively. These
typological prototypes are claimed to correlate cross-linguistically with relative
degrees of morphological markedness: Categories that represent prototypical
combinations take less function-indicating markers and can express more
morphological distinctions than non-prototypical combinations.

Croft’s approach yields a typology in which PoS classes are not directly
comparable across languages. On the other hand, the cross-linguistic
comparability of PoS classes defined according to Hengeveld’s approach is
achieved at the cost of ignoring specific sets of linguistic facts. As such, the
difference between these two approaches touches upon a fundamental issue
in linguistic typology, namely the (non-)existence of universal categories
(see Haspelmath 2007; Newmeyer 2007).
In addition, the difference between the Hengeveldian and the Croftian
approaches has important repercussions for the debate about so-called
flexible languages, i.e. languages that presumably lack all parts of speech
distinctions, most significantly a basic distinction between nouns and verbs.
The most thorny issue in this discussion probably concerns semantic shift in
flexible lexemes, i.e. the phenomenon that such lexemes may have slightly
different meanings in each of the multiple functions in which they can
be used. According to some authors the interpretation of flexible lexemes
should be fully compositional (Evans & Osada 2005; Croft 2005), while
others contend that it may involve idiosyncrasies (Hengeveld & Rijkhoff
2005). At the end of Chapter 2 I propose that, in order to explain the
relevant linguistic facts, one must take into account the mismatch between
lexical and syntactic categorization in flexible languages: Such languages do
have lexical categorization, but of a nature that is irrelevant to the phrase-
structural distribution of lexical items (Himmelmann 2007; Don & Van
Lier, forthcoming).
Chapter 3 is concerned with the theory and typology of dependent
clauses. As already mentioned, DC typology can be approached from two
perspectives. First, like PoS, DCs can be defined in terms of the propositional
function(s) that they may express. In parallel with the PoS classes predicted
by the Hengeveldian model in the previous chapter, in the first part of
Chapter 3 I define a number of rigid and flexible DC constructions that are
predicted to be possible in actual languages.
Second, DCs can be classified according to their internal morpho-
syntactic properties. These properties can be characterized as involving
specific combinations of ‘verbal’ features, which are associated with
independent clauses, and ‘nominal’ features, which are typical of lexical
expressions (cf. example (6) above). Earlier functional-typological studies
have shown that both verbal and nominal features display certain hierarchical
relations (Bybee 1985; Dik 1989, 1997; Rijkhoff 2002). This is reflected,
among other things, by universal ordering patterns of these features relative

to their base unit; a verbal or nominal head. These patterns are presumably
iconically motivated, to the extent that they reflect the nature and degree of
relevance of the various features to the interpretation of the head: Features
that occur closest to the head – such as aspect, tense, and mood (TAM)
markers for verbs, and class and number markers for nouns – affect the
head’s semantics. In contrast, features that are expressed further away from
the stem – in particular verbal agreement, and nominal case and definiteness
– do not affect the meaning of the stem but are rather relevant to its syntactic
and / or pragmatic function.
A number of functional-typological studies have investigated the way
in which these verbal and nominal feature hierarchies are combined in the
morpho-syntactic expression of DCs (Comrie 1976; Noonan 1985 / 2007;
Lehmann 1988; Koptjevskaja-Tamm1993; Mackenzie 1996; Croft 1991,
2001; Dik 1997; Cristofaro 2003; Malchukov 2004, 2006; Dixon &
Aikhenvald 2006). In Chapter 3 I focus on two recent exponents of this
literature: Cristofaro’s (2003) typology of subordination and Malchukov’s
(2004) typology of nominalizations.
Cristofaro (2003) identifies a number of correlations between various
structural phenomena in DCs that can be summarized as follows: (i) the loss
TAM marking implies the loss of verbal agreement; and (ii) the expression
of nominal features implies the loss of verbal features. Cristofaro proposes
three types of functional factors underlying these dependency relations, two
of which are the well-known functional principles of Economy and Iconicity
(Haiman 1983). These principles account for the non-expression of verbal
features in DCs, including the non-expression of arguments. However, since
neither Economy nor Iconicity can account for the expression of nominal
features in DCs, Cristofaro proposes a third explanatory principle. In
particular, she claims that DCs are cognitively not construed as independent
processes, but rather as things or properties. Since verbal features are
relevant for processes but not for things and properties, Cristofaro argues
that cognitive thing / property-construal leads to loss of verbal features. In
addition, thing / property-construal may lead to the expression of features
associated with the grammatical entities that prototypically code things and
properties, i.e. nouns and adjectives.
Broadly speaking, Malchukov’s (2004) typological study of clausal
nominalizations identifies the same range of cross-linguistic generalizations
as Cristofaro's, but proposes a different functional explanation, which is
adopted in the present study. Malchukov shows that verbal and nominal

features at the external end of their respective hierarchies are affected in
DC constructions before internal features. According to Malchukov, the
functional explanation for this generalization must be sought in the fact
that external categories reflect the syntactic and / or pragmatic function
of the linguistic unit on which they operate, while internal categories
are relevant for the semantic interpretation of their base. DCs can have
referential or modifying pragmatic functions, but they do not semantically
denote things or properties (rather, they denote higher order entities such
as states of affairs or propositions). This explains why external nominal
(or adjectival) features, which are relevant to the function of reference (or
modification), are expressed in DCs before internal ones, which are relevant
to thing / property-denoting entities. In addition, this explains why external
verbal features, which are relevant to pragmatic and syntactic aspects of
(independent) predications, are the firsts to be lost in DCs.
At the end of Chapter 3, and taking the earlier typological studies as a
basis, I define three types of DC constructions, in terms of their particular
combination of verbal and nominal features. One type corresponds to
balanced DCs, the other two represent different types of deranked DCs
(see above). This three-way distinction is combined with the classification
of DCs in terms of their functional possibilities, as defined earlier in
the same chapter. Thus, I arrive at a typological framework for DC
constructions that takes into account both their functional and their internal
formal properties.
Chapter 4 rounds off the theoretical part of the book. In this chapter, I first
present the composition of the sample of 50 languages that are investigated.
Second, I formulate and operationalize the specific hypotheses that are
tested on this sample. These hypotheses take the form of predictions about
dependency relations between the functional patterns displayed by the PoS
classes of particular languages, and the functional patterns of their (different
structural types of ) DC constructions. Some predictions are formulated in
general terms, i.e. they make reference to languages with some flexibility
versus no flexibility in their PoS system, and the expected reflection of this
difference on the functional properties of DCs in these languages. Other
predictions are more specific; they aim at the identification of one-to-one
matches between the functional possibilities of particular types of PoS
classes and DC constructions. In addition, both types of predictions are
formulated first without differentiation for dependent clauses in terms of
their internal formal properties, and then in sets of sub-predictions that

make specific reference to balanced clauses, as opposed to (different types
of ) deranked clauses.
Finally, Chapter 4 presents the method that is used to test these
predictions. Each prediction involves two binary parameters; one concerning
PoS, the other DCs. In order to test the predictions, the observed frequencies
(the number of languages that display a particular value combination of the
PoS and the DC parameter) are compared to the expected frequencies (the
number of languages with this feature combination that would be expected if
the co-occurrence of the PoS and DC features would be purely coincidental).
Fischer’s Exact tests are used to calculate whether the deviations between
observed and expected frequencies are statistically significant. Whenever a
significant result is found, the particular nature of the dependency relation
between the PoS and DC traits is further assessed by means of a statistical
method developed by Maslova (2003).
The second part of the book (Chapters 5-7) contains the actual
typological data. First, Chapter 5 presents the typology of PoS classes as
displayed by the languages of the sample. The attested PoS systems are
evaluated in light of the predictions made in Chapter 2. It is shown that
most predicted PoS systems are found in actual languages, either in their
‘pure’ form, or in combination with another predicted system. In general,
these data suggest a reasonable typological adequacy of the implicational
map model of PoS as outlined in Chapter 2.
The remainder of Chapter 5 addresses a variety of issues regarding
the identification of PoS classes in individual languages, and their cross-
linguistic comparison. These issues, which are also touched upon in the
theoretical discussion on PoS typology in Chapter 2, include the problem
of fuzzy boundaries between PoS classes, the phenomenon of variable
distributional patterns displayed by sub-groups of lexical items, and the
occurrence of ‘restricted’ lexeme classes, i.e. small, closed classes and classes
consisting of derived lexemes. I also discuss the expression of non-verbal
predication in the sample languages. In the final section of Chapter 5 I apply
the theoretical notions about lexical flexibility, as proposed in Chapter 2, to
the relevant languages of the sample.
Chapter 6 presents the classification of the DC constructions attested
in the sample languages, in terms of the typological framework developed
in Chapter 3. First, the DCs are classified according to the propositional
function(s) that they can express. The attested DC types are compared with
the DC types predicted in Chapter 3. It is shown that all predicted types

occur in the languages of the sample, except those involving the functional
slot for head of a predicate phrase. I suggest that this latter finding may be
explained in terms of cognitive-semantic markedness of DCs with respect
to the function of independent predication, and / or a structural constraint
on the expression of finite verbal morphology on DCs. Second, the DCs
of the sample languages are classified according to their internal morpho-
syntactic properties. On the basis of these data, every DC construction is
assigned to one of the three structural DC types defined in Chapter 3. At
the end of Chapter 6 these data are combined with the DC classification
based on functional distribution into an integrated typology.
Subsequently, in Chapter 7, the data sets presented in Chapter 5 (PoS)
and Chapter 6 (DCs) are linked in order to test the predictions about their
functional interrelationship, as formulated in Chapter 4. The analyses reveal
significant dependency relations between flexible (rather than rigid) PoS
and DCs; between very flexible (rather than less flexible) PoS and DCs;
and between PoS and deranked (rather than balanced) DCs. The relevant
dependency relations are shown to be asymmetrical in nature, and can as
such be interpreted as statistical bases for implicational universals.
More specifically, it is shown that the presence of a pervasively flexible
deranked DC construction in a particular language requires the presence
of a pervasively flexible PoS class in that language. This does not mean,
however, that all languages with a large degree of flexibility in their PoS
system also display flexible deranked DC constructions. In fact, the results
make clear that pervasively flexible PoS classes and deranked DCs are
both, i.e. independently of each other, cross-linguistically rare phenomena.
Therefore, the generalization that maximal flexibility in the domain of
deranked DCs depends upon maximal flexibility in the lexical domain
does not have much explanatory power. More interestingly, it is shown that
whenever maximal lexical flexibility does occur in a language, this strongly
increases the chances of also finding the other rare phenomenon: maximally
flexible deranked DCs This shows that there is indeed a strong tendency for
the PoS and DC parameters to have the same value.
In contrast to the result obtained for pervasively flexible PoS and
deranked DCs, no significant dependency relations can be established
between the presence of less flexible and rigid PoS classes on the one hand
and deranked DCs with the same functional behaviour on the other hand.
In addition, it is shown that the functional patterns of balanced DCs are not
related to those of PoS classes.

In the third and final part of the book (Chapters 8 and 9) I further discuss
the results of the study and present its conclusions. First, In Chapter 8,
the findings of Chapter 7 are reconsidered, taking a specific functionalist
perspective that makes reference to the overall complexity of language
systems and the various ways in which this level of complexity can be attained.
More specifically, the proposed explanatory framework builds upon recent
functional-typological research, which has advanced the idea that flexibility
and rigidity are relative rather than absolute notions that may be applicable
in various degrees to specific construction types pertaining to different
levels of the grammar. These studies suggest that the categorial specificity of
linguistic constructions increases – or their flexibility decreases – when they
become structurally more complex. This generalization has been termed
the Principle of Increasing Categoriality or the Principle of Staggering Level-
dependent Categoriality (Haig 2006, Lehmann 2008). This principle can be
regarded as a specific instance of an even more general functional principle
regarding complexity in language systems, namely that flexibility or multi-
functionality in one area of the grammar must be counterbalanced or ‘traded
off ’ by rigidity or categorial specificity in another area, in order to guarantee
the identifiability of the function of any linguistic unit within an actual
utterance. This has been termed the Principle of Functional Transparency
(Frajzyngier & Shay 2003, cf. Hengeveld et al. 2004; Sinnemäki 2008).
In Chapter 8 I argue that deranked (but not balanced) DCs can be
regarded as secondary constructions, derived from primary underived
lexical constructions. Under this assumption, the Principle of Increasing
Categoriality predicts that the degree of functional flexibility of any
deranked DC construction should be either equal to or smaller than the
flexibility of a PoS class that can express at least one of the same function(s).
This prediction is born out for virtually all relevant constructions in the
sample languages. Thus, the Principle of Increasing Categoriality explains
the finding of Chapter 7 that flexibility in the domain of deranked DCs
is dependent on flexibility in the domain of PoS classes. This dependency
relation can now be interpreted as a constraint, imposed by the amount of
flexibility attested in the PoS system of a particular language, on the maximal
amount of flexibility (or minimal amount of categorial specificity) that can
be displayed by the deranked DC(s) of that language.
Furthermore, the Principle of Increasing Categoriality sheds light on
the unexpected findings of Chapter 7: (i) the lack of correlations between
the presence of weakly flexible PoS classes and deranked DCs with the

same functional possibilities in particular languages; and (ii) the lack of
correlations between various types of rigid PoS classes and the corresponding
rigid deranked DCs.
The first finding can be understood as involving a decrease in flexibility
on the part of the DC construction, as compared to the PoS class. In particular,
while the relevant PoS classes are weakly flexible to the extent that they can
express two propositional functions, the DCs are rigid constructions; they
can be used in just a single function. Notably, a similar explanation can
be offered for the fact that even languages with pervasively flexible PoS
classes display rigid deranked DC (alongside flexible deranked DCs). Both
findings are in accordance with the prediction that deranked DCs can be
either as flexible as or less flexible than PoS classes, but not more flexible.
Second, the lack of a correlation between rigid PoS classes and rigid
deranked DCs can be interpreted as follows. As expected, whenever a
language has a deranked DC construction that can appear in a function for
which a rigid PoS class is available, then the DC construction in question
is rigid as well. Again, this confirms that a deranked DC cannot exhibit
a greater degree of flexibility than a PoS class. However, instead of rigid
deranked DCs, languages may employ balanced DCs. The latter, which are
presumably not derived from lexical categories, were shown in Chapter 7 to
not display any functional relation with PoS classes.
Despite the fact that balanced DCs and PoS are thus neither formally
nor functionally related, I argue at the end of Chapter 8 that the distributional
properties of balanced DCs are in accordance with the more general Principle
of Functional Transparency. Whereas rigid balanced DCs are functionally
transparent by definition, flexible balanced DCs, like any other flexible
construction type, are expected to require additional morpho-syntactic
means in order to compensate for their inherent functional ambiguity. I
show that balanced flexible DC constructions may indeed combine with
several types of such disambiguating strategies. Some of these strategies,
such as the use of resumptive pronouns, are confined to DC constructions,
while others, such as constituent order constraints, are used more generally
to establish functional transparency in linguistic (sub-)systems.
Regarding this last point, a particularly interesting finding is that
flexible DCs (both balanced and deranked ones) in languages with very
flexible PoS systems often make use of the same morpho-syntactic strategies
that are used to disambiguate the functions of flexible lexical and phrasal
constructions in these languages. This suggests that languages with pervasive

lexical flexibility make use of a kind of ‘grid’ of morpho-syntactically marked
functional slots, which may be filled by formal units of any structural type:
not just single lexemes, but also complex phrasal and clausal constituents. All
these construction types can (but need not) retain maximal flexibility until
they are inserted into a slot corresponding to a particular phrase-structural
function.
Finally, Chapter 9 summarizes the main findings of the study. It
concludes with the observation that the explanatory principles of Increasing
Categoriality and Functional Transparency are ultimately motivated in
terms of Economy: Languages divide the task of establishing maximal
categorial specificity over the lexical, morphological, and syntactic devices
available in their grammatical systems, and complexity-increasing linguistic
processes typically produce output structures that are at least as categorially
specific as their input structures, in order to maximize processing ease for
the participants of a communicative situation.

Samenvatting in het Nederlands
Taaltypologie houdt zich bezig met taalvergelijking. Het primaire

probleem waarmee typologen zich geconfronteerd zien, is het volgende:
Om taalspecifieke structuren onderling te kunnen vergelijken, is een vorm-
onafhankelijk tertium comparationis nodig. Om die reden kiezen typologen
eerst een bepaald functioneel domein, waarna ze onderzoeken welke formele
structuren er in verschillende talen bestaan om bepaalde onderscheidingen
binnen dat domein uit te drukken. Nadat een typoloog het spectrum van
structurele variatie tussen talen binnen een functioneel domein in kaart heeft
gebracht, probeert hij of zij universalia te formuleren: afhankelijkheidsrelaties
tussen bepaalde structuurkenmerken die gelden voor alle talen. Binnen de
functionalistische benadering van taaltypologie, waaronder ook de onderhavige
studie valt, wordt aangenomen dat dergelijke universalia verklaard kunnen
worden in termen van de functie van menselijke taal, namelijk het coderen
van communicatie (Croft 2003).
Drie basale communicatieve of propositionele functies kunnen worden
onderscheiden: Referentie, predicatie, en modificatie. Refereren betekent:
het identificeren van een referent, i.e. een entiteit waarover je wilt praten.
Prediceren houdt in: iets toeschrijven aan een referent, i.e. vertellen dat deze
referent zich in een bepaalde situatie bevindt. Modificeren betekent het
uitbreiden van een referentiële of predicatieve uiting, door het uitdrukken
van een additionele eigenschap van ofwel een referent ofwel een situatie
(Searle 1969, Croft 2001: 66).
467
Deze drie propositionele functies – referentie, predicatie, en modificatie –
kunnen in talen worden uitgedrukt door verschillende soorten constructies.
Voorbeeld (1) laat zien dat een referentiële uitdrukking zoals het direct object
van het werkwoord zie de vorm kan hebben van een simpele lexicale constructie
(een auto (in 1a)), maar ook van een meer complexe, ingebedde zin (dat Jan een
nieuwe auto heeft gekocht in (1b)). Dit onderzoek heeft betrekking op de twee
constructietypes in voorbeeld (1): lexicale constructies of woordsoorten (parts
of speech) en ingebedde of afhankelijke zinsconstructies (dependent clauses).
(1) a. Ik zie een auto.

b. Ik zie dat Jan een nieuwe auto heeft gekocht.
Parts of speech is de traditionele term voor woordsoorten, dat wil zeggen de

klassen van inhoudswoorden die onderscheiden worden in de grammatica
van een bepaalde taal (Schachter & Shopen 2007:1). De klasse van
zelfstandig naamwoorden of nomina (nouns) wordt meestal gebruikt om te
refereren (cf. de auto in (2a)); werkwoorden of verba (verbs) worden gebruikt
voor predicatie (cf. wast in (2b)); bijvoeglijk naamwoorden of adjectieven
(adjectives) voor het modificeren van referentiële uitdrukkingen (cf. rode
in (2c)); en bijwoorden of adverbia (adverbs) voor het modificeren van
predicatieve uitdrukkingen (cf. grondig in (2d)).
(2) a. Henk wast de auto.

b. Henk wast de auto.
c. Henk wast de rode auto.
d. Henk wast de auto grondig.
Echter, niet alle talen van de wereld vertonen dit specifieke

woordsoortensysteem, waarbinnen elke klasse speciaal geschikt is voor het
uitdrukken van één bepaalde functie. Sommige talen hebben een ander
type woordsoortensysteem, waarbinnen twee of meer functies uitgedrukt
kunnen worden door één en dezelfde woordklasse. Het Turks, bijvoorbeeld,
heeft één klasse van woorden die zowel gebruikt kunnen worden voor
referentie (zoals in voorbeeld (3a)), als voor modificatie van een referentiële
of predicatieve uitdrukking (cf. respectievelijk voorbeeld (3b) en (3c)). De
Engelse vertalingen van de voorbeelden in (3b) en (3c) laten zien dat in het
Engels een nomen zoals beauty eerst afgeleid moet worden met de uitgangen
-ful en -ly, voordat het gebruikt kan worden in modificerende functies.

(3) a. güzel-im
beauty-1poss
‘my beauty’
b. güzel bir kopek

beauty art dog

c. güzel konuştu
beauty s / he.spoke
‘S / he spoke beautifully’
Deze verschillende types van woordsoorten worden in dit boek aangeduid

met twee termen die zijn voorgesteld door Hengeveld (1992): Woordklassen
die gespecialiseerd zijn in het uitdrukken van één functie worden rigide
genoemd, en woordklassen die zonder aanpassingen gebruikt kunnen
worden in twee of meer functies worden flexibel genoemd.
Met betrekking tot ingebedde zinnen of dependent clauses (verderop
afgekort als DCs) wordt meestal een drievoudig onderscheid gemaakt.
Complementzinnen (complement clauses) worden gebruikt om te refereren:
ze functioneren als het subject of object van een hoofdzin (zie voorbeeld
(1b)). Betrekkelijke of relatieve zinnen (relative clauses) modificeren het
subject of object van de hoofdzin. En bijwoordelijke of adverbiale zinnen
(adverbial clauses) modificeren het predicaat van de hoofdzin (Thompson et
al. 2007: 238).
Echter, net zoals er rigid en flexibele woordklassen zijn, bestaan er ook
rigide en flexibele ingebedde zinnen. Het Abun bijvoorbeeld (een taal uit
het westen van Papua) beschikt over een DC-constructie, gemarkeerd door
het onderschikkende voegwoord do, die uitsluitend referentieel gebruikt kan
worden, dat wil zeggen als complementzin (cf. voorbeeld (4)).
Abun (Berry & Berry 1999: 167)

(4) Ji nut [do an bi obat ne nggi wa
1sg think comp 3sg poss magic det strong for
be an gu hi ré.]
later 3sg kill 1sg this
‘I thought that his magic was strong [enough] for him to kill me.’
Samenvatting in het Nederlands | 469

De Baskische DC-constructie gemarkeerd door -en kan daarentegen zowel
een referentiële als een modificerende functie vervullen. Dit blijkt uit
voorbeeld (5), waar de constructie als complementzin wordt gebruikt in (5a)
en als relatieve zin in (5b).

knew Mikel late arrive.fut aux-comp-det

Behalve in termen van hun functionele mogelijkheden kunnen ingebedde

zinnen – aangezien het complexe constructies zijn – ook worden
bestudeerd vanuit het perspectief van hun interne morfo-syntactische
eigenschappen. Meestal worden deze eigenschappen beschreven in relatie
tot de eigenschappen van ‘gewone’, niet-ingebedde hoofdzinnen in de taal
in kwestie. Dit kan worden geïllustreerd aan de hand van de verschillen
tussen de Engelse DCs in voorbeeld (6). In de DC in (6a), gemarkeerd door
het onderschikkende voegwoord dat, wordt de toekomende tijd uitgedrukt
(will move) en ziet het subject (Paul) er net zo als in een hoofdzin. In de
DC in (6b) daarentegen, staat het werkwoord in de speciale ing-vorm,
kan er geen tijdsmarkering worden uitgedrukt, en heeft het onderwerp de
bezittelijke vorm (Paul’s). Het eerste type DC, dat structureel gezien lijkt op
een gewone hoofdzin, wordt in de literatuur balanced genoemd; het tweede
type, dat in bepaalde opzichten afwijkt van de structuur van een hoofdzin,
wordt deranked genoemd (Stassen 1985).

In dit proefschrift onderzoek ik of er een relatie bestaat tussen de functionele

flexibiliteit versus rigiditeit van de woordklassen in een bepaalde taal en de
ingebedde zinnen in die taal. Meer in het bijzonder richt deze studie zich op
de vraag in hoeverre het mogelijk is om op basis van het woordsoortensysteem
van een taal te voorspellen wat de functionele mogelijkheden zullen zijn van

verschillende typen ingebedde zinsconstructies in die taal. Het antwoord op
deze vraag is gebaseerd op een systematische analyse van de woordklassen en
ingebedde zinnen in een groep van 50 talen met een maximale genealogische
en geografische diversiteit (zie Appendix 1).
In het eerste gedeelte van dit boek (Hoofdstuk 2-4) wordt de theoretische
en methodologische basis gelegd voor het eigenlijke onderzoek. Hoofdstuk
2 bespreekt een aantal vraagstukken met betrekking tot de definitie van
woordklassen in talen, en de vergelijking van woordklassen tussen talen. Dit
hoofdstuk begint met een kort overzicht van benaderingen in de typologie
van woordklassen, aan de hand van verschillende soorten criteria die relevant
zijn voor de definitie van woordklassen, te weten semantische, morfologische,
syntactische, en pragmatische criteria (Givón 2001: 49; Sasse 1993b:196).
Vervolgens worden twee specifieke benaderingen binnen de typologie
van woordklassen uitvoeriger besproken: de pragmatisch-syntactische
theorie van Hengeveld en de universeel-typologische theorie van Croft.
Binnen de eerstgenoemde benadering worden woordsoorten uitsluitend
gedefinieerd in termen van de propositionele functie(s) die ze kunnen
uitdrukken (Hengeveld 1992, Hengeveld et al. 2004, Hengeveld & Van Lier
2008, 2009). Hierbij wordt gebruik gemaakt van een functioneel domein
bestaande uit twee dimensies. De eerste dimensie betreft de functionele
oppositie tussen predicatie en referentie. Aangenomen wordt dat de functie
van predicatie geprivilegieerd is ten opzichte van referentie, aangezien het
tot stand brengen van referentie vereist dat een bepaalde eigenschap van die
referent geprediceerd wordt. De tweede dimensie betreft het onderscheid
tussen hoofden en modificeerders binnen constituenten. Hier is de aanname dat
hoofden belangrijker zijn dan modificeerders, omdat hoofden de verplichte
kern van een constituent vormen, terwijl modificeerders optioneel kunnen
worden toegevoegd. De twee bovengenoemde functionele dimensies staan
bovendien in een hiërarchische relatie tot elkaar. Terwijl alle talen op een
zeker grammaticaal niveau het onderscheid tussen predicatie en referentie
maken, lijkt het hoofd-modificeerder onderscheid niet altijd grammaticaal
relevant te zijn. Om die reden wordt aangenomen dan de predicatie-referentie
parameter een primaire rol heeft ten opzichte van de hoofd-modificeerder
parameter. Deze drie dominantierelaties vormen samen een hiërarchische
model, op basis waarvan een aantal specifieke voorspellingen gedaan wordt
ten aanzien van mogelijke en onmogelijke woordsoortensystemen, in termen
van de hoeveelheid en het soort functies dat door bepaalde woordklassen
wordt uitgedrukt.

Uit het bovenstaande blijkt dat de Hengeveldiaanse benadering van
woordsoortentypologie uitsluitend gebruik maakt van pragmatisch-
syntactische criteria. Deze benaderingswijze is bekritiseerd door Croft (2000,
2001, 2005), volgens wie woordklassen gedefinieerd moeten worden op basis
van een uitgebreidere set van criteria, waarbij ook de semantische betekenis
van woorden en hun morfologische gedrag in aanmerking worden genomen.
Binnen Croft’s theorie worden woordsoorten gezien als prototypische
combinaties van een semantische betekenis en een propositionele functie.
Een prototypisch nomen verwijst naar een object (een ding) en heeft een
referentiële functie; een prototypisch verbum verwijst naar een actie en heeft
een predicatieve functie; en een prototypisch adjectief verwijst naar een
eigenschap en heeft een modificerende functie. In crosslinguïstisch perspectief
correleren deze typologische prototypen met bepaalde maten van morfo-
syntactische markering: Als een woordklasse een prototypische combinatie
van semantiek en pragmatiek representeert, dan gaat dit samen met relatief
minder morfo-syntactische functie-indicatoren, en met meer mogelijkheden
om morfologische onderscheidingen uit te drukken dan wanneer er sprake
is van een niet-prototypische combinatie.
Croft’s benadering resulteert in een typologie waarbij woordklassen
niet direct vergelijkbaar zijn tussen talen. Aan de andere kant, voor
de crosslinguïstische vergelijkbaarheid van de woordklassen in de
Hengeveldiaanse benadering wordt ook een prijs betaald, namelijk het
buiten beschouwing laten van bepaalde taalkundige feiten. Dit onderscheid
tussen de twee benaderingen raakt aan een fundamentele kwestie binnen de
taaltypologie, namelijk de vraag of er zoiets bestaat als een set van universele
categorieën in menselijke taal (zie Haspelmath 2007; Newmeyer 2007).
Bovendien hebben de verschillen tussen de Hengeveldiaanse en de
Croftiaanse benaderingen belangrijke gevolgen voor de discussie over
zogenaamde flexibele talen, dat wil zeggen talen waarvan gezegd wordt dat
ze geen woordklassen hebben, met name geen onderscheid tussen nomina
en verba. Het meest heikele punt in deze discussie betreft semantic shift:
kleine verschillen in de interpretatie van flexibele woorden of lexemen, die
samenhangen met de verschillende functies waarin zulke lexemen kunnen
voorkomen. Volgens sommige onderzoekers vereist flexibiliteit dat de
interpretatie van lexemen in verschillende functies volledig compositioneel
is (Evans & Osada 2005; Croft 2005), terwijl anderen juist beweren dat
die interpretatie idiosyncratische kenmerken kan vertonen (Hengeveld &
Rijkhoff 2005). In het laatste gedeelte van Hoofdstuk 2 betoog ik dat de

relevante taalkundige feiten verklaard kunnen worden door te onderkennen
dat er in flexibele talen geen één-op-één-relatie bestaat tussen lexicale en
syntactische categorisatie. Zulke talen hebben wel degelijk woordklassen,
maar van een zodanige aard dat de klasse waartoe een woord behoort
irrelevant is voor de functionele distributie van dat woord (Himmelmann
2007; Don & Van Lier, forthcoming).
Hoofdstuk 3 bespreekt een aantal theoretische en typologische aspecten
van ingebedde zinnen (DCs). Zoals eerder aangegeven kan de typologie van
DCs vanuit twee perspectieven worden bezien. Ten eerste kunnen DCs, net
als woordklassen, worden geclassificeerd in termen van de propositionele
functie(s) die ze kunnen uitdrukken. In het eerste deel van Hoofdstuk 3
gebruik ik het Hengeveldiaanse functionele model uit het vorige hoofdstuk
om voorspellingen te doen over welk soort rigide en flexibele DC constructies
wel en niet in talen zouden moeten kunnen voorkomen.
De tweede manier waarop DCs kunnen worden geclassificeerd
is op basis van hun interne morfo-syntactische eigenschappen. Deze
eigenschappen kunnen worden gedefinieerd als specifieke combinaties van
verbale features, die geassocieerd worden met hoofdzinnen, en nominale
features, die typisch zijn voor lexicale uitdrukkingen (zie voorbeeld (6)
hierboven). Functioneel-typologisch onderzoek laat zien dat zowel verbale
als nominale features bepaalde hiërarchische patronen vertonen, wat onder
meer tot uitdrukking komt in de volgorde waarin zulke features staan ten
opzichte van het hoofd waar ze bij horen (een verbum of nomen). Deze
vaste volgorde is iconisch gemotiveerd; het is een reflectie is van de manier
waarop en de mate waarin de afzonderlijke features de interpretatie van
het hoofd beïnvloeden. Features die dicht bij het hoofd worden uitgedrukt
hebben betrekking op de semantiek van het hoofd. Dit zijn met name de
verbale categorieën tijd, aspect, en modaliteit (tense, aspect, mood, afgekort
als TAM) en de nominale categorieën getal en geslacht. Features die verder
van het hoofd verwijderd zijn hebben daarentegen geen directe invloed op
de betekenis, maar zijn relevant voor de syntactische en pragmatische functie
van de constructie waarin ze verschijnen. Dit geldt voor verbale congruentie
en voor nominale (on)bepaaldheid en casusmarkering.
In een aanzienlijk aantal studies is onderzocht wat de invloed van deze
verbale en nominale feature hiërarchieën is op de morfo-syntactische expressie
van ingebedde zinnen (Comrie 1976; Noonan 1985 / 2007; Lehmann 1988;
Koptjevskaja-Tamm1993; Mackenzie 1996; Croft 1991, 2001; Dik 1997;
Cristofaro 2003; Malchukov 2004, 2006; Dixon & Aikhenvald 2006). In

Hoofdstuk 3 beperk ik mij tot het bespreken van twee recente studies op dit
gebied: Cristofaro’s typologische studie van subordinatie (Cristofaro 2003)
en Malchukov’s typologische studie van nominalizaties (Malchukov 2004).
Cristofaro (2003) stelt een aantal correlaties vast tussen structurele
aspecten van DCs. Deze correlaties kunnen als volgt worden samengevat: (i)
het niet kunnen uitdrukken van TAM features impliceert het niet kunnen
uidrukken van verbale congruentie; en (ii) het uitdrukken van nominale
features impliceert het verlies van verbale features. Cristofaro draagt drie
typen functionele verklaringen aan voor deze afhankelijkheidsrelaties. Twee
van de drie verklarende factoren zijn de fundamentele functionele principes
van Economie en Iconiciteit (Haiman 1983). Deze principes verklaren het
verlies van verbale features in DCs, inclusief het achterwege blijven van
subject en / of object argumenten. Echter, noch Economie noch Iconiciteit kan
verklaren waarom er ook nominale features in een DC kunnen verschijnen.
Daarom stelt Cristofaro een derde factor voor: Zij betoogt dat DCs in
cognitief opzicht niet worden beschouwd als onafhankelijke processen,
maar worden geconstrueerd als dingen of eigenschappen. Aangezien verbale
features niet relevant zijn voor processen, kunnen ze in DCs achterwege
blijven. Bovendien kan het cognitief construeren van een DC als een ding
of eigenschap leiden tot het uidrukken van features die normaal gesproken
horen bij lexicale constructies die dingen en eigenschappen uitdrukken:
nomina en adjectieven.
Malchukov’s (2004) studie levert in grote lijnen dezelfde generalisaties
op als Cristofaro’s onderzoek, maar geeft er een andere functionele
verklaring voor, waarbij ik mij in dit proefschrift aansluit. Malchukov’s
onderzoek laat zien dat er in DC constructies vaker iets aan de hand is met
verbale en nominale features die een externe positie innemen binnen hun
respectievelijke hiërarchieën dan features met een meer interne positie. Om
precies te zijn, het uitdrukken van een bepaalde nominale feature in een DC-
constructie impliceert dat alle nominale met een meer externe positie óók
kunnen worden uitgedrukt in die DC. Voor verbale features geldt dat het niet
kunnen uitdrukken van een bepaalde feature impliceert dat alle meer externe
features van de hiërarchie ook niet kunnen verschijnen. Volgens Malchukov
moet de verklaring voor deze bevindingen gezocht worden in het feit dat
externe features relevant zijn voor pragmatische en syntactische aspecten
van de constructie waarin ze verschijnen, terwijl interne features betrekking
hebben op de semantiek. Hoewel DCs de pragmatisch-syntactische functies
van referentie en modificatie vervullen, hebben ze niet de semantische

betekenis van een ding of eigenschap (integendeel, ze drukken situaties of
proposities uit). Dit verklaart waarom externe nominale (of adjectivische)
features, die betrekking hebben op deze pragmatisch-syntactische functies,
eerder uitgedrukt worden in DCs dan interne features. Bovendien verklaart
dit waarom externe verbale features, die te maken hebben met de pragmatiek
en syntaxis van predicaties, juist kunnen vervallen in DCs constructies.
Op basis van deze studies definieer ik in het laatste deel van Hoofdstuk
3 drie types van DC-constructies, in termen van hun specifieke combinaties
van verbale en nominale features. Eén van deze types is vergelijkbaar met
balanced DCs; de andere types zijn te beschouwen als twee verschillende
soorten deranked DCs (zie boven). Deze classificatie wordt vervolgens
gecombineerd met de DC-classificatie in termen van functionele
mogelijkheden, zoals gedefinieerd in het eerste gedeelte van Hoofdstuk
3. Dit resulteert in een typologisch raamwerk voor DCs dat zowel hun
functionele als hun vormelijke eigenschappen in aanmerking neemt.
Hoofdstuk 4 is het laatste hoofdstuk van het theoretische gedeelte
van het boek. In dit hoofdstuk ga ik eerst in op de samenstelling van het
sample van talen waarop mijn onderzoek gebaseerd is. Vervolgens formuleer
en operationaliseer ik de specifieke hypotheses die in de rest van de studie
worden getest. Deze hypotheses hebben de vorm van voorspellingen
over afhankelijkheidsrelaties tussen de functionele mogelijkheden van
woordklassen in specifieke talen en van verschillende structurele DC-types
in die talen. Een aantal van deze voorspellingen is relatief breed van opzet; ze
hebben betrekking op talen die enige flexibiliteit in hun woordsoortensysteem
vertonen versus talen die geen enkele flexibiliteit vertonen, en de manier
waarop dit verschil gereflecteerd wordt in de functionele eigenschappen
van DCs. Andere voorspellingen zijn specifieker: ze zijn gericht op het
identificeren van één-op-één relaties tussen woordsoorten met bepaalde
functionele eigenschappen enerzijds en DCs met dezelfde eigenschappen
anderzijds. Bovendien worden beide typen voorspellingen telkens op twee
manieren geformuleerd: eerst zonder te differentiëren tussen verschillende
structurele DC-types, en vervolgens in sets van van sub-predicties die
speciaal betrekking hebben op balanced DCs, dan wel op (verschillende
soorten) deranked DCs.
Tenslotte wordt in Hoofdstuk 4 de methode uitgelegd die ik gebruik
om deze hypotheses te toetsen. Elke voorspelling bevat twee parameters:
één met betrekking tot woordklassen, de ander met betrekking tot DCs.
Elke voorspelling wordt getoetst door de geobserveerde frequenties (het aantal

talen dat een bepaalde combinatie van woordklasse- en DC-kenmerken
vertoont) te vergelijken met de verwachte frequenties (het aantal talen
met deze combinatie dat je op basis van toeval zou verwachten). Om te
bepalen of het verschil tussen de geobserveerde en verwachte frequenties
statistisch significant is, wordt een Fischer’s Exact test gebruikt. Wanneer er
sprake is van een significante correlatie tussen een woordklasse-gerelateerd
en een DC-gerelateerd structuurkenmerk, dan wordt de aard van deze
afhankelijkheidsrelatie nader gespecificeerd door middel van een methode
ontwikkeld door Maslova (2003).
Het tweede deel van dit proefschrift (Hoofdstuk 5-7) bevat de
eigenlijke typologische gegevens. Om te beginnen wordt in Hoofdstuk 5 de
classificatie van de woordklassen in de talen van mijn sample gepresenteerd.
De aangetroffen woordsoortensystemen worden vergeleken met de
woordsoortensystemen zoals voorspeld in Hoofdstuk 2. Hieruit blijkt dat
de meeste voorspelde systemen ook inderdaad gevonden worden, in hun
‘pure’ vorm dan wel in een combinatie met een ander voorspeld systeem. In
het algemeen suggereren deze data dat het in Hoofdstuk 2 gepresenteerde
hiërarchische model in redelijke mate typologisch adequaat is.
De rest van Hoofdstuk 5 behandelt verschillende kwesties rondom
de identificatie en vergelijking van woordklassen binnen en tussen talen.
Deze problemen, die al in algemeen-theoretische zin aan orde kwamen
in Hoofdstuk 2, hebben betrekking op het voorkomen van vage grenzen
tussen woordklassen, van variabel distributioneel gedrag bij bepaalde sub-
klassen, en van ‘beperkte’ woordklassen (kleine, gesloten klassen en klassen
die bestaan uit afgeleide vormen). Ook wordt de uitdrukking van niet-
verbale predicatie in de talen van het sample besproken. De laatste sectie van
Hoofdstuk 5 betreft een toepassing van de theoretische noties over lexicale
flexibiliteit, zoals voorgesteld in Hoofdstuk 2, op de relevant talen in het
sample.
Hoofdstuk 6 beschrijft de classificatie van de DC-constructies in de
sample talen, in termen van het in Hoofdstuk 3 ontwikkelde typologische
raamwerk. Eerst worden de DC constructies geclassificeerd op basis van
de propositionele functie(s) die ze kunnen uitdrukken. Het resultaat wordt
vergeleken met de voorspellingen uit Hoofdstuk 3. De data laten zien dat
alle voorspelde DC-types voorkomen, behalve degenen die (onder andere)
de functie van hoofd van een predicatie zouden moeten kunnen uitdrukken.
Deze bevinding duidt op een cognitief-semantisch gemarkeerde status
van DCs ten opzichte van de functie van onafhankelijke predicatie, en / of

op een structurele beperking ten aanzien van het uitdrukken van finiete
werkwoordsmorfologie op DCs.
In het tweede deel van Hoofdstuk 6 worden de interne morfo-syntactische
eigenschappen van de DCs in de talen van het sample geïdentificeerd. Op
basis van deze gegevens kan elke DC-constructie worden geclassificeerd als
behorende tot één van de drie DC-types zoals gedefinieerd in Hoofdstuk 3.
Aan het einde van Hoofdstuk 6 wordt deze classificatie geïntegreerd met de
DC-classificatie op basis van functionele distributie.
Hoofdstuk 7 brengt de datasets uit Hoofdstuk 5 (woordklassen) en
Hoofdstuk 6 (ingebedde zinnen) op systematische wijze met elkaar in verband,
zodat de predicties uit Hoofdstuk 4 getest kunnen worden. De analyses
wijzen uit dat er significante afhankelijkheidsrelaties bestaan tussen flexibele
(in plaats van rigide) woordklassen en DCs; tussen zeer flexibele (in plaats van
minder flexibele) woordklassen en DCs; en tussen woordklassen en deranked
(in plaats van balanced) DCs. De correlaties in kwestie zijn asymmetrisch
van aard, en kunnen als zodanig worden geïnterpreteerd als statistische
fundamenten voor de formulering van implicationele universalia.
Meer in het bijzonder laten de resultaten in Hoofdstuk 7 zien dat de
aanwezigheid in een bepaalde taal van een deranked DC-constructie met
een hoge mate van flexibiliteit vereist dat die taal een zeer flexibel type
woordklasse heeft. Dit betekent echter niet dat alle talen met veel flexibiliteit
in hun woordklassensysteem ook flexibele deranked DC-constructies
hebben. Bovendien wijzen de data uit dat zowel zeer flexibele woordklassen
als zeer flexibele deranked DCs – onafhankelijk van elkaar – crosslinguïstisch
zeldzame fenomenen zijn. Om die reden heeft de generalisatie dat maximale
flexibiliteit bij deranked DCs afhankelijk is van maximale flexibiliteit in het
woordklassensysteem niet veel verklarende potentie. Veel interessanter is
de observatie dat, áls er sprake is van zeer flexibele woordklassen in een
taal, dit de kans aanzienlijk verhoogt dat die taal ook zoiets zeldzaams als
flexibele deranked DC-constructies heeft. Dit betekent dat er wel degelijk
een sterke tendens in talen bestaat om gelijke waarden te vertonen voor
structurele parameters die betrekking hebben op het functionele gedrag van
woordklassen en deranked DCs.
In tegenstelling tot de resultaten voor zeer flexibele woordklassen en
deranked DCs, kunnen er geen significante correlaties worden vastgesteld
tussen de aanwezigheid in talen van minder flexibele en rigide woordklassen
enerzijds, en de aanwezigheid van deranked DC met hetzelfde functionele
gedrag anderzijds. Bovendien wordt duidelijk uit de resultaten dat de

functionele patronen van balanced DCs in talen niet gerelateerd kunnen
worden aan die van woordklassen in dezelfde talen.
In het derde en laatste deel van dit proefschrift (Hoofdstuk 8 en 9)
worden de resultaten van het onderzoek nader becommentarieerd en worden
de conclusies geformuleerd. Hoofdstuk 8 heroverweegt de bevindingen
uit Hoofdstuk 7 vanuit een specifiek functionalistisch perspectief, dat
verband houdt met het algemene complexiteitsniveau van taalsystemen
en de verschillende manieren waarop dit niveau kan worden bereikt. Het
verklarende raamwerk dat ik voorstel, is gefundeerd op recent functioneel-
typologisch onderzoek, waaruit naar voren komt dat flexibiliteit en rigiditeit
relatieve noties zijn, die in verschillende mate van toepassing kunnen zijn op
bepaalde constructietypes, die weer behoren tot verschillende grammaticale
niveaus. Deze studies laten zien dat de categoriale specificiteit van
linguïstische constructies toeneemt – of hun flexibiliteit afneemt – naarmate
ze structureel complexer worden. Deze generalisatie wordt in de literatuur
Principle of Increasing Categoriality of Principle of Staggering Level-dependent
Categoriality genoemd (Haig 2006, Lehmann 2008). Dit principe kan
worden beschouwd als een specifieke invulling van een algemener functioneel
complexiteitsprincipe, namelijk dat flexibiliteit of multifunctionaliteit in een
bepaald deel van een grammaticaal systeem gecompenseerd moet worden
met rigiditeit of categoriale specificiteit in een ander deel, opdat de functie
van iedere constructie binnen een zin identificeerbaar is. Dit mechanisme
wordt wel Principle of Functional Transparency genoemd (Frajzyngier &
Shay 2003, cf. Hengeveld et al. 2004; Sinnemäki 2008).
In Hoofdstuk 8 bepleit ik dat deranked DCs, in tegenstelling tot
balanced DCs, beschouwd kunnen worden als een secundair constructietype
dat is afgeleid van een primaire lexicaal constructietype. Onder deze
aanname voorspelt het Principle of Increasing Categoriality dat de mate van
functionele flexibiliteit van een deranked DC-constructie in een bepaalde
taal kleiner of gelijk moet zijn aan de flexibiliteit van een woordklasse in
die taal die gebruikt wordt in tenminste één van dezelfde functies. Deze
hypothese wordt bevestigd voor vrijwel alle relevante constructies in
de sample talen. Dit betekent dat het Principle of Increasing Categoriality
verklaart waarom flexibiliteit bij deranked DCs afhankelijk is van flexibiliteit
in het woordsoortensysteem. Deze afhankelijkheidsrelatie, vastgesteld in
Hoofdstuk 7, kan nu worden geïnterpreteerd als een beperking, opgelegd
door de mate van flexibiliteit in het woordsoortensysteem van een taal, op
de mate van flexibiliteit die is ‘toegestaan’ bij deranked DCs in die taal.

Bovendien werpt het Principle of Increasing Categoriality licht op de
onverwachte resultaten uit Hoofdstuk 7, namelijk (i) het ontbreken van
correlaties tussen minder flexibele woordklassen en deranked DCs met dezelfde
functionele distributie, en (ii) het ontbreken van correlaties tussen verschillende
rigide woordsoorten en rigide deranked DCs van hetzelfde type.
De eerste bevinding kan worden geïnterpreteerd als een geval van
afnemende flexibiliteit bij DCs, in vergelijking met de woordklassen in
kwestie. De relevante woordklassen zijn matig flexibel omdat ze (slechts)
twee propositionele functies kunnen uitdrukken. Wanneer de DC-
constructies minder flexibel zijn dan de woordklassen, dan resulteert dit
automatisch in rigide constructies, die nog maar in één enkele functie
gebruikt kunnen worden. Een gelijksoortige mechanisme verklaart waarom
zelfs talen met zeer flexibele woordklassen toch rigide deranked DCs
hebben (naast eventuele flexibele deranked DCs): dit is in overeenstemming
met de voorspelling dat deranked DCs óf dezelfde flexibiliteit hebben als
woordklassen, óf minder flexibiliteit (maar nooit méér).
Ten tweede kan het ontbreken van correlaties tussen rigide woordklassen
en deranked DCs nu beter begrepen worden. Zoals verwacht blijkt in alle
gevallen waar een taal een deranked DC-constructie gebruikt in een functie
waarvoor een rigide woordklasse beschikbaar is, de DC-constructie óók
rigide te zijn. Echter, in plaats van een deranked constructie, kan een taal
ook een balanced DC gebruiken. Deze balanced DCs kunnen zowel rigide
als flexibel zijn; uit Hoofdstuk 7 blijkt immers dat balanced DCs, die formeel
niet afgeleid zijn van lexicale constructies, ook geen functionele verwantschap
met woordklassen vertonen.
Hoewel balanced DCs en woordklassen dus formeel en functioneel
ongerelateerd zijn, laat ik in het laatste deel van Hoofdstuk 8 zien dat de
distributionele eigenschappen van balanced DCs wel in overeenstemming
zijn met het meer algemene principe van Functional Transparency. Terwijl
rigide balanced DCs per definitie functioneel transparant zijn, zijn flexibele
balanced DCs – net als andere flexibele constructies – functioneel ambigu.
De verwachting is daarom dat het gebruik van balanced flexibele DCs vereist
dat er andere morfo-syntactische middelen in worden gezet om de inherente
ambiguïteit van deze constructies te compenseren. Ik laat zien dat balanced
flexibele DC constructies inderdaad samengaan met verschillende soorten
disambiguërende strategieën. Soms zijn deze strategieën speciaal toegespitst
op DC-constructies, zoals bij het gebruik van resumptieve pronomina, en in
andere gevallen, zoals bij woordvolgorde-restricties, gaat het om strategieën

die een bredere functie vervullen bij het bewerkstelligen van functionele
transparantie binnen het taalsysteem.
Met betrekking tot dit laatste punt is het interessant om te zien dat
flexibele DCs – zowel balanced als deranked – in talen met zeer flexibele
woordsoortensystemen vaak gebruik maken van dezelfde morfo-syntactische
strategieën als die ingezet worden bij het functioneel disambiguëren van
flexibele lexicale en andersoortige constructies in die talen. Dit duidt op een
patroon waarbij talen met een hoge mate van lexicale flexibiliteit gebruik
maken van een soort rasterwerk van morfo-syntactisch gemarkeerde slots, die
gevuld kunnen worden met constructies van allerlei aard: niet alleen simpele
woorden, maar ook complexere frases en ingebedde zinnen. Al deze meer
en minder complexe constructietypes hebben de mogelijkheid (maar niet de
noodzaak) om hun functionele flexibiliteit te behouden totdat ze worden
ingevoegd in een slot die correspondeert met een bepaalde zinsfunctie.
In Hoofdstuk 9, tenslotte, worden de belangrijkste resultaten van
het onderzoek samengevat. Dit hoofdstuk sluit af met de concluderende
observatie dat de verklarende principes van Increasing Categoriality en
Functional Transparency gefundeerd zijn op het algemene Economie principe.
De algemene taak om maximale categoriale specificiteit te bewerkstelligen
wordt door talen verdeeld over de verschillende lexicale, morfologische,
en syntactische middelen die ze tot hun beschikking hebben. Bovendien
produceren complexiteitverhogende grammaticale processen meestal output-
constructies die een minstens even hoge mate van categoriale specificiteit
vertonen als hun input-constructies. Zo kan het verwerkingsgemak voor de
taalgebruiker worden geoptimaliseerd.

Curriculum Vitae
Eva van Lier was born in The Hague (the Netherlands) in 1978. During
her secondary education at the Gymnasium Haganum (The Hague) she
studied classical Latin and Greek, as well as French, German, and English.
After finishing high school in 1996, she spent a year in Granada and Sevilla
(Spain) to learn Spanish and obtain a Diplóma Básico de Español como lengua
estranjera (basic degree of Spanish as a foreign language). In 1997, she started
studying Dutch Linguistics and Literature at the University of Groningen.
She continued the following year at the University of Amsterdam, studying
Spanish Linguistics and Literature and General Linguistics. She obtained
her MA degree in the latter subject with honours in 2002, with a thesis
on the relationship between linguistic typology and second language
acquisition. Subsequently, she completed a Research Master in Linguistics
and obtained her MPhil degree with honours in 2003. Her thesis was again
concerned with typology and its connections with other linguistic sub-
disciplines, including language contact and creolization, as well as first and
second language acquisition. During this period Eva was also employed as
a student assistant at the Steunpunt Nederlands as Vreemde Taal (Support
Center for Dutch as a Foreign Language) of the University of Amsterdam
and the Nederlandse Taalunie (Dutch Language Academy). After her
university studies, she worked for one year as a researcher at the Meertens
Institute for Dutch language and culture, on a project about multilingualism
in the Netherlands. In 2004 she started on her PhD research project at
481
the department of Theoretical Linguistics of the University of Amsterdam.
During this period she functioned occasionally as a teaching assistant
(BA and MA courses), and as a freelancer for the Taalstudio, contributing
material for linguistic education in Dutch high schools. She was a PhD
representative at the Advisory Board of the ACLC (Amsterdam Center for
Language and Communication) in 2006 and 2007. Eva has been involved
in several national and international collaborative projects, which resulted
in an international workshop in Amsterdam and a two-month stay at the
Max Planck Institute for Evolutionary Anthropology in Leipzig in 2007, as
well as several (co-authored) conference presentations, reviews, and journal
papers. A co-edited volume on lexical flexibility is in preparation. The
present thesis is the result of Eva’s doctoral studies.

Typology Eva Van Lier

Uploaded by

Copyright:

Available Formats

Typology Eva Van Lier

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Typology Eva Van Lier

Uploaded by

Copyright:

Available Formats

Parts of Speech and Dependent Clauses

LOT phone +31 (0)30 253 6006

Copyright © 2009 by Eva van Lier. All rights reserved.

ter verkrijging van de graad van doctor

Eva Helena van Lier

Faculteit der Geesteswetenschappen

5 Parts of Speech in the Languages of the Sample 131

6 Dependent Clauses in the Languages of the Sample 187

8 Discussion – Flexibility and Functional Transparency 313

9 Summary and Conclusion 349

Summary in English 453

VIII | Parts of Speech and Dependent Clauses

Very particularly, I want to thank my paranimf, Lotte, for adopting me

I want to thank my parents. Papa, thanks for being my father in a way

Amsterdam, June 2009

X | Parts of Speech and Dependent Clauses

Chapter 2 Parts of Speech

Chapter 3 Dependent Clauses

XII | Parts of Speech and Dependent Clauses

Chapter 3 Dependent Clauses

Chapter 4 Sampling, Research Questions, Method

Chapter 5 Parts of Speech in the Languages of the Sample

Chapter 6 Dependent Clauses in the Languages of the Sample

XIV | Parts of Speech and Dependent Clauses

XVI | Parts of Speech and Dependent Clauses

Chapter 8 Discussion: Flexibility and Functional Transparency

List of Tables | XVII

XX | Parts of Speech and Dependent Clauses

List of Abbreviations | XXI

XXII | Parts of Speech and Dependent Clauses

List of Abbreviations | XXIII

(1) a. I wash the car.

(2) Mary hopes [that John will come home soon].

Like parts of speech, certain types of dependent clause constructions are

(3) I gave her the book [that we bought together yesterday].

Apart from their possibilities regarding the expression of communicative

2 | Parts of Speech and Dependent Clauses

(4) a. John regrets [that Paul will move to Finland].

The aim of the present research is to investigate the relationship between

4 | Parts of Speech and Dependent Clauses

6 | Parts of Speech and Dependent Clauses

2.2. Functional-typological attempts at defining parts of speech

8 | Parts of Speech and Dependent Clauses

(i) Semantic criteria: the kind of meanings that tend to be coded by

To these, a fourth type may be added, namely:

(iv) Pragmatic criteria: the kind of discourse function that tends to be

Although most approaches acknowledge the relevance of all four types of

Chapter 2 – Parts of Speech | 9

10 | Parts of Speech and Dependent Clauses

Chapter 2 – Parts of Speech | 11

2.3.1 Slots and fillers

(i) The head of a predicate phrase and

(iii) A modifier within a predicate phrase and

The four slots can thus be tabulated as in Figure 2.1:

12 | Parts of Speech and Dependent Clauses

Chapter 2 – Parts of Speech | 13

14 | Parts of Speech and Dependent Clauses

Head of Head of Modifier in Modifier in

2.3.2 The cross-linguistic perspective

2.3.2.1 General introduction