Re Reduplication

Author(s): Alec Marantz

Source: Linguistic Inquiry, Vol. 13, No. 3 (Summer, 1982), pp. 435-482
Published by: The MIT Press
Stable URL: [Link]
Accessed: 22/10/2009 18:41

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
[Link] JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
[Link]

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@[Link].

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

[Link]
Alec Marantz Re Reduplication

In the recent literature,reduplicationhas been claimed to cause two serious problems

for theories of morphologyand phonology. First, as McCarthy(1979)points out, no one
has developed an observationallyadequateformalizationof reduplicationrules without
adopting a notation which allows morphologicalrules to be written that never occur
[Link] investigatorshave either implicitly(e.g. Wilbur(1973), Munro
and Benson (1973)) or explicitly (e.g. Carrier(1979)) employed a transformationalno-
tation for reduplicativeprocesses. Munro and Benson (1973, 18) characterizea redu-
plicative adjective formationrule of Luisefio as shown in (1), while Carrier(1979, 353)
formulatesa reduplicationrule of Tagalog as shown in (2) (where M is a morpheme).
(1) CIV1C2V2 + C1V1C2V2 + i + C
(2) [(M) C V X

1 2 3 4- 1, 2, 3 ,2, 3,4
[ - long]
Given the formal apparatusof (1) or (2), one can write many types of morphological
rules which are not instantiatedin [Link] example, althoughexpressible
in the transformationalnotations of (1) and (2), mirror-imagereduplicationrules such
as (3a,b) are not found in any language.
(3) a. C1V1C2V2+ V2C2V1CI
b. CVCV
1 2 3 4 -4, 3, 2, 1, 1, 2, 3, 4
An adequate theory of reduplicationwould formalizethe process while allowing one to
express only the sorts of reduplicationwhich occur cross-linguisticallyand not the sorts
of rules exemplified in (3).
Wilbur(1973)presentsthe most extensive discussionof the second apparentproblem
for an analysis of reduplication-its unusual interactionwith certain phonologicalpro-
Among the many people who have aided and abetted with the proverbialuseful discussions, I would
particularlylike to thankMorrisHalle and Shelly Lieber, who providedthe title. I have also benefitedgreatly
from the commentsof three anonymousLI reviewers. This work was supportedby an NSF GraduateStudy
Fellowship.
The researchsummarizedhere was conductedduringthe 1979-1980academicyear. The articleitself was
writtenin August, 1980,and has since been revised. In the time since the ideas presentedhere were developed,
they have been productivelyset to work by a numberof otherlinguists,e.g. in Halle and Vergnaud(1980a,b),
Pranka(1981), and Yip (1981).

Linguistic Inquiry, Volume 13, Number 3, Summer 1982

0024-3892/82/030435-48 $02.50/0
? 1982 by The Massachusetts Institute of Technology 435
436 ALEC MARANTZ

cesses. As we shall see in section 2, some phonologicalrules appearto "overapply" to

reduplicatedforms, that is, to apply to segmentsin both the base and the copied material
in a reduplicatedform although only one set of these segments occurs in the proper
environmentfor the applicationof the rule. This behaviorof reduplicatedforms has led
some linguists (e.g. Carrier(1979)) to order the overapplyingphonological processes
before reduplicationin the [Link] output of these processes may thus be copied
by reduplication, yielding the appearanceof [Link] this approach
accounts for the occurrenceof the outputof the phonologicalprocesses in both the base
and the copied material, it has the result of mixing morphologicaland phonological
processes. The conceptually simpler theory would place these processes in separate
components of the [Link], as Wilbur(1973) points out, the orderingap-
proach would not be able to account for the other puzzling interactionof phonological
rules with reduplication:some phonologicalrules appearto underapplyto reduplicated
forms; that is, they do not apply to segments in either the base or the copied material
in a reduplicatedform althoughone set of these segments occurs in the environmentof
the rule. An adequate theory of reduplicationwould explain these apparentlyaberrant
interactions of reduplicationand phonological processes and predictjust which pho-
nological processes will "over- and underapply"to reduplicatedforms.
The solution to these problemsassociated with reduplicationis simply to make the
minimalspecial assumptionsor statementsabout [Link] for the fact that
the material attached to the stem in reduplicationresembles the stem phonologically,
reduplicationrules look like normal affixation processes. To provide the best account
of reduplicationrules, we say they are normal affixation processes. The one unique
feature of reduplication,the feature which leads us to group together diverse morpho-
logical processes under the title reduplication,is the resemblanceof the added material
to the stem being [Link] demonstratedin section 1, we can devise a simple
procedure for lending to the reduplicatingaffix phonemic material from the stem to
which it attaches without adding unneeded power to the grammar.
The solution to problemsposed by reduplication,then, is to say that there is nothing
special about reduplicationother than the resemblancebetween the affix and the stem
to which it is attached. An extension of John McCarthy's recent account of Arabic
verbal paradigms(McCarthy(1979; 1981))provides a simple formalismfor reduplicating
processes which does not involve the full power of transformations(section 1). As for
the interactionof reduplicationwith phonologicaland other morphologicalrules (section
2), once we establish that reduplicationis simply affixation, recent improvements in
phonologicaland morphologicaltheory explainthis interactionby predictingwhich rules
will appear to over- and underapplyto reduplicatedforms. No special orderingof re-
duplicationrules or special conditionson phonologicalrules prove necessary to account
for the data.

1. The Formal Nature of Reduplication

In this section I exploit a proposal made by John McCarthy (1979; 1981) in an analysis
of the Arabic verbal system to provide a formal account of reduplicative processes.
RE REDUPLICATION 437

McCarthyclaims that words shouldbe represented(in part)as consonant-vowel skeleta

(hisprosodic templates)connected to phonemicmelodies on separatetiers in accordance
with the principlesof autosegmentalphonology (see Goldsmith(1976)on autosegmental
phonology and Halle and Vergnaud(1980a)on "tiered phonology").
(4) phonemic melody Pi P2 P3 P4 P5 P6 P7 . ..

consonant-vowel skeleton C V C C V C V.

syllabic skeleton >~~V V ..

morphemesymbol
pi phoneme
C consonant
V vowel
= syllable
I will review McCarthy'sdiscussion of Arabic verbs in section 1.2. I will support the
claim that most reduplicationprocesses are best analyzed as the affixationof a conso-
nant-vowel (C-V) skeleton, itself a morpheme,to a stem. The entire phonemic melody
of the stem is copied over the affixed C-V skeleton and linked to C and V "slots" in
the skeleton accordingto principlesto be made explicit below. Section 1.4 attempts to
unify the analysis of C-V skeletal reduplicationwith the remainingexamples of redu-
plication found in the literature.

1.1. What Reduplication Is Not and Preliminary Indications of What It Is

Ignoringdifficulties, I will tentatively identify reduplicationas a morphologicalprocess
relatinga base form of a morphemeor stem to a derived form that may be analyzed as
being constructed from the base form via the affixation (or infixation) of phonemic
materialwhich is necessarily identical in whole or in part to the phonemic content of
the base form. This workingdefinitionof reduplicationmatches, for the most part, the
term's use in the [Link] providinga formal analysis of reduplicationprocesses,
this article will give, in essence, a formaldefinitionof reduplicationto replace the rough
characterizationabove. To the extent that this formal definition excludes processes
which share crucial propertieswith the processes it includes, to that extent my analysis
of reduplicationis in error.'
1.1.1. Reduplication Is Not Constituent Copying. Many languages include morpholog-
ical processes which copy entire morphemes or words (Moravcsik (1978) cites many
examples). For instance, Warlpiriforms the plural of some nouns, primarilythose re-
ferringto humans, by total reduplication(Nash (1980, 130)).

' This definitiondoes exclude some of the Semitic "doubling"and "gemination"processes discussed
in McCarthy(1981). McCarthyexplains how these processes differ cruciallyfrom what I will call "redupli-
cation"; see footnote 7.
438 ALEC MARANTZ

(5) a. kurdu'child' kurdukurdu'children'2

b. kamina'girl, maiden' kaminakamina'girls, maidens'
c. mardukuja'woman, female' mardukujamardukuja 'women, females'
Any theory of morphologywill need some mechanismto effect the copying of an entire
morphemeas in (5).
Bankingon this fact, one might propose a straightforwardanalysis of reduplication
claiming that it always involves the copying of a constituent of a morphemeat some
level of analysis or at some tier in an autosegmentalrepresentationof the morpheme.
On this hypothesis (found, for example, in McCarthy(1979, chapter 4)), reduplication
could copy a phoneme, a syllable, a metricalfoot, an entire morpheme,or some other
constituent of a morpheme but could not copy pieces of constituents which do not
themselves make up a constituent.
However, well-attested reduplicationrules do copy sequences of consonants and
vowels from a morpheme which form no constituent of the morpheme. For example,
we find reduplicationrules which prefix a copy of the first CV of a stem to the stem
regardlessof whether the C and V constitute the entire first syllable of the word or only
its onset and syllabic nucleus.3 Quileute forms plurals with such a rule according to
Andrade (1933, 189) as reportedin Moravcsik(1978):4
(6) a. ci-ph6kwat' 'Negro' cici-phokwat''Negroes'
b. qa*x'bone' qaqa*x'bones'
Tagalog employs three distinct processes of reduplication,one which prefixes a copy
of the first CV of a stem to the stem makingthe copied V short, one which prefixes a
copy of the first CV of a stem to the stem makingthe copied V long, and one which
prefixes a copy of the first CV(C)CV(C)of a stem to the stem makingthe copy of the
second V long (Carrier(1979)).These reduplicationprocesses, sometimes in connection
with additionalaffixation,are used for a varietyof derivationaland inflectionalpurposes.
Although the inclusion of the last C in the third reduplicationprocess depends on the
stem being reduplicatedin a manner discussed in footnote 5, none of the Tagalog re-
duplicationrules respect the syllabificationor constituentstructureof the forms to which
2
For the most part I will employ the orthographyof my sources in cited examples, providingphonetic
interpretationonly when relevantto the discussion. I have occasionallymodifiedtranscriptionsto substitute
morewidely used symbolsfor moreobscureones, to ease typographicreproduction,or to standardizedifferent
sources on the same language.I have also left out stress and tone markingsin some examples where these
have no bearingon the issues at hand.
3 In the pages to follow I will write as if there were no problemin decidingwhethera given reduplication
process involves prefixing,suffixing,or [Link] fact, there are usuallystrongargumentsfor classifyinga
reduplicatingrule in one of these categories, some of which may be foundin the sources I cite. However, for
most of the discussion below, it will not be importantwhetherwe considerany particularreduplicationrule
as prefixing,suffixing,or infixingthe "copied" material.
4 Andrade(1933, 189):

Reduplicationconcerns regularlyonly the initial consonantor the first vowel of the word or both. . ..
This principleis strictlyadheredto even in cases in which a monosyllabicstem has a terminalconsonant,
or when we may infer from the generalphonetictendenciesthat the consonantfollowingthe first vowel
belongs to the initial syllable.
RE REDUPLICATION 439

they apply. That is, they copy a CV or CV(C)CV(C)whether or not these form a
constituent (syllable or metricalfoot).
(7) a. lakad 'walk' pag-lalakad'walking'
b. kandilah'candle' pag-kakandilah'candle vendor'
c. linis 'clean' mag-lilinis'will clean'
d. um-takboh'run' um-tatakboh'will run'
(tumakbohafter (>tumatakboh)
infixation)
e. ma-talinoh'intelligent' ma-talftalfnoh'ratherintelligent'
f. baliktad balibaliktad'all topsy-turvy'
Examples (7b,d,f) clearly display reduplicationprocesses which do not copy a constit-
uent.
The CV reduplicationprocesses in QuileuteandTagalogcopy a C and a V regardless
of whetherthey make up a syllableor only partof a syllable. Thereare also reduplicating
processes which prefix a copy of the first CVC of a stem to the stem regardless of
whether the CVC constitutes the first syllable of the stem or the first syllable plus the
onset of the following syllable. For example, Agta forms varioussorts of pluralsby CVC
prefixingreduplication(examples from Healey (1960, 7)).
(8) a. bari 'body' barbari-kkid-in'my whole body'
b. mag-saddu'leak (verb)' mag-sadsaddu'leak in many places'
c. na-wakay'lost' na-wakwakay'many things lost'
d. takki 'leg' taktakki'legs'
In examples (8a,c), reduplicationcopies materialwhich does not make up a constituent
of the word being reduplicated.
As reported in Krause (1980), Chukchee also exhibits a CVC reduplicationrule
which does not respect syllabic structureand which, therefore, does not copy a pho-
nologicalconstituent. Copyingthe initialCVC of a nounto the rightof the nouflproduces
the absolutive singularin Chukchee.
(9) a. jil?e- 'gopher' jil?e-jil'abs. singular'
b. nute- 'earth, ground' nute-nut'abs. singular'
Exampleslike (9b)demonstratethe copyingof a CVC sequencewhich does not constitute
a syllable in Chukchee.

1.1.2. Reduplicationas Affixationof Skeletal Morphemes. We have seen that the con-
stituentcopying theory of reduplicationfails because reduplicationmay copy sequences
of Cs and Vs which do not form a constituent. However, every reduplicationprocess
may be characterizedby a "skeleton"'of some sort, either a C-V skeleton, a syllabic
skeleton, or a skeleton of morphemesymbols (see (4)). That is, the shape of the copied
materialin reduplicationis fixed for the reduplicationprocess; the shape is independent
440 ALEC MARANTZ

of the hierarchical structure of the morpheme being copied.5 After reviewing a large
sample of reduplication rules from the world's languages in connection with the Stanford
Project on Language Universals, Moravcsik (1978, 307) concluded that6

whereas the relevant string [i.e. the portion of a stem to be copied by reduplication]could
in principlebe defined by any phonetic property(segmentalor suprasegmental)or in terms
of absolute linear position, or in terms of simply the numberof adjacentsegments involved;
and it could also be left undefined(i.e. "reduplicateany one or more segments in the total
string"), reduplicatedphonetic strings I found invariablydefined in reference to conso-
nant-vowel sequences and absolute linear position.

My own research has identified only one exception to Moravcsik's claim (brought to
my attention by David Nash), the Yidiny reduplication rule to be discussed in section
1.4.
Moravcsik's generalization suggests that reduplication rules involve the affixation
of a C-V skeleton to a stem, the C-V skeleton borrowing phonemes from the phonemic
melody of the stem to which it attaches. After an introduction to C-V skeleta in the
form of a review of McCarthy's (1979; 1981) work on Arabic verbs, I will present a
theory of reduplicative processes which claims that most reduplication is just that-the
affixation of a C-V skeletal morpheme to a stem and the association of a copy of the
stem's phonemic melody with the affixed skeleton. As will be shown in section 1.4, this
theory readily extends to the syllabic reduplication of Yidiny and the full morpheme
reduplication found in many languages.

1.2. An Introduction to C-V Skeleta: McCarthy's Analysis of the Arabic Verbal System
With the preliminary observations of the workings of reduplicative processes made in
section 1.1, we are almost prepared to develop a complete formal account of redupli-
cation. First, however, we must examine McCarthy's (1979; 1981) use of C-V skeleta

5Aside fromthe Yidinysyllabicreduplicationandthe wholemorphemereduplicationsdiscussedin section

1.4, I know of two cases in which the C-V skeleton of the reduplicatingmorphemedoes dependon the stem
to which it attaches: CVCCV(C)reduplicationin Dyirbal (Dixon (1972))and Tagalog (Carrier(1979); see
examples (7e,f)). In both languages, a morpheme-finalC following the first CVC(C)Vof a stem is either
optionally(Dyirbal)or obligatorily(Tagalog)reduplicatedalongwith the CVC(C)[Link] that the dependence
of the C-V skeleton of the reduplicatingmorphemeon the stem in these cases has nothingto do with the
syllabificationor hierarchicalstructureof the stem. The rules in questiondo not copy an additionalsyllable-
final C unless it is also [Link],when the constitutionof the reduplicatingC-V skeleton
does seem to depend on the stem, i.e. when the extra C is copied, what the reduplicationrule reduplicates
is an entire stem. We have alreadynoted that total morphemeor stem reduplicationis quite commoncross-
[Link], what we shouldsay aboutthe reduplicationin Tagalogand Dyirbalwhich copies the extra
morpheme-finalC is not that the C-V skeleton of the reduplicatingmorphemeis dependenton the stem in
these cases but ratherthat there is no C-V reduplicatingskeleton involvedat all-the rule is copyinga stem.
That is, the reduplicatingaffix has two allomorphsin Tagalogand [Link], a morphemeskeletal affix
(see section 1.4), attaches either obligatorily (Tagalog) or optionally (Dyirbal) to stems of the form
(C)V(C)CVC#,copying the whole stem. The other allomorph,the C-V skeletalaffix CVCCV,attachesto all
other stems. See section [Link] for furtherdiscussion of the Tagalogcase.
6 McCarthy(1979, 367-368) providesan insightfuldiscussionof the significanceof Moravcsik'sfindings
for the "tiered phonology" model assumed in this article.
RE REDUPLICATION 441

in his analysis of Arabic verbs. This review serves two purposes: to explain the mech-
anismsof C-V skeleta, phonemicmelodies, andtheirassociationandto give independent
justification for the formal machineryrequiredfor an analysis of reduplication.
Consider table 1, an expanded piece of McCarthy'stable 1 (1981, 385). Each row
displays part of the inflectionalparadigmfor the Arabic root ktb 'write' in one binyan
(plural, binyanim)or conjugation.
The first binyan [not included in my table 1/AM]is a possible category for nearly all roots
that can appearas verbs. It is relativelyunmarkedphonologically,at least in the finite forms
and it has no special semanticproperties. . . But the others, the so-calledderivedbinyanim,
generallyinvolve some special modificationof the meaningof the root. So, for instance, the
thirdtriliteralbinyan is usually reciprocal,while the sixth is usuallythe reflexive or effective
of the [Link] is, in general,an idiosyncraticpropertyof any root whetherit can appear
in a particularbinyan. Nevertheless, neologisms abound,loanwordsare easily incorporated
into the system, and speakers of Modern StandardArabic report a reasonable facility in
extending a root to other binyanimand interpretingthe result. (McCarthy(1979, 239))

McCarthynotes that, for the most part, each binyan has a characteristicC-V shape,
shown in the last column of table 1. The same triliteralroot, ktb, appearsin all the forms
in table 1 and, with one exception (the imperfectiveactive), the same vocalic melodies
appearconsistently within each column (see the second to last row in table 1). The key
to a revealing analysis of the Arabic verbal system, McCarthyclaims, is to separatethe
root consonants and vocalic melodies from each form as morphemes in themselves.
These morphemes attach to the various C-V skeletal binyanimof the second to last
column in table 1 according to the principlesof autosegmentalphonology with a more
or less predictablesemantic effect.
The binyanim,considered as C-V skeletal morphemes,operate in a mannersimilar
to derivationalaffixes in other languages. Although, as McCarthynotes, they do not
always impartthe same meaning to the roots with which they associate, they possess
a usual semantics which allows their extension, for example, to borrowed roots. For
what follows, we shall assume that the binyanim,the consonantalroots, and the inflec-
tional vocalic melodies are all morphemes-they all have lexical entries containing
informationabout their possible combinationwith other morphemesand about the cat-
egorial and semantic results of such combinations.
Constituentsof the autosegmentalmorphemesroot, binyan, and inflectionalvowel
melody link together as dictated by the constraints of autosegmentalphonology (cf.
Goldsmith(1976)). The overridingprincipleof autosegmentalphonology states that
(10) Linking lines never cross.
In the Arabic verbal system, anotherprinciplerequiresthat
(11) Each slot in the skeleton is linked to at least one segment in the phonemic
melody.
On McCarthy's analysis, each morpheme is arrangedon a separate "tier" or level.
Table 1. Paradigm for the Arabic root ktb 'write'

Perfective Imperfective Participle

Binyan Active Passive Active Passive Active Passive C-V Skele

11 kattab kuttib ukattib ukattab mukattib mukattab CVCCVC
III kaatab kuutib ukaatib ukaatab mukaatib mukaatab CVVCVC
IV ?aktab ?uktib u?aktib uOaktab mu?aktib mu?aktab CVCCVC

VI takaatab tukuutib atakaatab utakaatab mutakaatib mutakaatab CVCVVC

I
t
VIl nkatab nkutib ankatib unkatab munkatib munkatab CCVCVC
I
n
Vill ktatab ktutib aktatib uktatab muktatib muktatab CCVCVC

t
X staktab stuktib astaktib ustaktab mustaktib mustaktab CCVCCV
II
s t
Inflec- a ui (fu+?a Ti (u+)a (mu+)ai (mu+)a
tional LaJJ4
Vocalic
Melody

Note: "The forms in the table are all stems, so they do not contain mood, agreement,or case, gender, or numberm
abstract away from certain generally accepted phonological processes. . ." (McCarthy (1979, 242-243))
RE REDUPLICATION 443

Principle (10) prevents the crossing of lines that connect the same two tiers, while
principle (11) ensures that each of the C and V slots in a binyan will be connected to
some phoneme.
The perfective active of ktb in binyanIII (see table 1) serves as a simple illustration
of the association of autosegmentalmorphemes.
(12) root morpheme k t b
I I |= kaatab
"derivational"skeletal CVVCVC
morpheme(binyan)

inflectionalmorpheme a
Principle(10) rules out the linkingof morphemesshown in (13); principle(11) rules out
the incomplete linking in (14).
(13) k t b
I = *kaabat
CVVCVC

a
(14) k t b
I I = *katab
CVVCVC

a
Binyan II illustrates an attractivefeature of the autosegmentalapproachto Arabic
verbs. Consider ktb in the perfective active of binyan II (see table 1).
(15) k t b
I AI = kattab
CVCCVC
\Va
Principle(11) insists that one of the root consonantsattachto two C slots in the skeleton,
while McCarthyjustifies the rules and principleswhich yield kattab instead of *kaktab
or *katbab. McCarthythus provides a simple account of the geminationof the middle
consonant of a triliteralroot in the second binyan.
Finally, consider ktb in binyan VIII perfective active.
(16) k t b
I I ktatab
CCVCVC

a
444 ALEC MARANTZ

This form illustratesthe fact that phonemesmay be preattachedto slots in a C-V skeletal
morpheme. Not only is the skeleton CCVCVC characteristicof binyan VIII, but also
the second C is invariablyt (cf. ktasab < root ksb 'earn'). This latter generalizationis
capturedby linking a t to the second C slot of the CCVCVCskeletal morphemewithin
the lexical entry of the skeletal morphemeitself. We shall see that the preattachment
of phonemes or features to skeletal morphemesis a widespreadfeatureof reduplication.
McCarthyclaims that the phonemes (i.e. feature matrices) in a phonemic melody
are unspecified for the feature [? syllabic]. This feature is acquired by the phonemes
throughtheir attachmentto the C-V skeleton, where C is equivalentto [- syllabic] and
V to [ + syllabic]. However, arguingfrom the Arabic data, Halle and Vergnaud(1980b,
7) conclude "that in the melody tier the feature [ ? syllabic] is specified and that the
linking [of melody to C-V skeleton] is subject to the furthercondition (Sb)" (= (17)).
(17) Unless overridden by a special proviso, feature complexes containing the
feature [- syllabic] can be linked only to C slots in the skeleton, and feature
complexes containingthe feature [ + syllabic] can be linked only to V slots in
the skeleton.
One argumentfor condition (17) comes from the perfective passive binyan II form
of ktb, kuttib (see table 1). If a phoneme of the vocalic inflectionalmorphemeui were
unspecified for [ ? syllabic], it could link either to a C slot, acquiringthe feature L- syl-
labic] and therefore designatinga glide, or to a V slot, acquiringthe feature [ + syllabic]
and therefore designatinga vowel. Thus, if McCarthywere correct, we would have no
general way of ruling out associations of morphemes such as (18a) for the binyan II
perfective passive of ktb, in place of the correct (18b).

(18) a. k t b b. k t b
I I | *kuwtib I /\ I = kuttib
CVCCVC CVCCVC

On the other hand, if phonemes in a phonemic melody are specified for the feature
[ ? syllabic], the linking of the u, a [ +syllabic] element, to the C slot in (18a) would be
prohibitedby Halle and Vergnaud'scondition (17), which proves essential for the anal-
ysis of reduplication.
Because the consonant and vocalic melodies in Arabic form distinct morphemes
and therefore may be placed on distinct "tiers", McCarthy(1979; 1981)is able to avoid
assumingprinciple(17) and specifyingthe feature [ ? syllabic] in his phonemicmelodies.
For each sort of morphemeconsisting of a phonemic melody, McCarthyspecifies the
melody-bearingelements, Cs or Vs, to which elements of that sort of morphememay
attach. McCarthyrules out the linking illustratedin (18a), for example, by stipulating
that the melodic elements of inflectionalmorphemesmust attach to V slots, not C slots.
For the Arabic data at least, McCarthy's method of ruling out linkings like (18a) is
RE REDUPLICATION 445

equivalent to assigning all of the melodic elements in a morpheme one value of the
feature [?syllabic] and adopting principle (17). As will become apparentbelow, Mc-
Carthy's method is unavailablewhen [ + syllabic] and [ - syllabic] melodic elements are
intermixedon a single tier. In the analysis of reduplication,we must adopt (17) over the
procedure of specifying for each melodic tier to which melody-bearingelements the
items on that tier may link.

1.3. Reduplication and C-V Skeleta

In section 1.1, I claimed, following Moravcsik (1978), that each reduplicatingprocess
can be characterized by a skeleton. The essence of this article's formal analysis of
reduplicationis the claim that the skeleton which is characteristicof a reduplicationrule
is actually a reduplicatingmorpheme, i.e. a skeletal affix comparableto the binyanim
of the Arabic verbal system. Reduplicationis simply the affixation of a skeleton to a
stem, as in (19), where Agta initial CVC reduplicationis exemplified (see the forms in
(8)).
(19) t akk i
11111
CVC + [CVCCV]
This section discusses reduplicationcharacterizedby C-V skeleton morphemes; the
analysis is extended to syllable and full morphemereduplicationin section 1.4.
On the analysis of C-V reduplicationas the affixationof a C-V skeleton morpheme,
we are left with the problem of how to link up some of the phonemes from the stem to
the added C-V skeleton. However we accomplish this, we will be expressing that sole
characteristicof reduplicatingmorphologicalprocesses which distinguishesthem from
other affixing processes: namely, that the phonemic melody of a reduplicatingaffix is
dependent on the phonemic melody of the stem to which it attaches.
I propose that the only mechanism available to morphologicaltheory to lend the
stem's phonemic melody to a reduplicatingaffix is the copying "over" the reduplicating
affix of the entire phonemic melody of the stem, i.e. the copying of the stem's phonemic
melody on the same tier as the melody and on the same side of the stem melody to which
the affix is attached.7This wholesale copying of a phonemic melody, as shown in (20),
along with some specific constraintson the autosegmentalassociation of the phonemes
of the copied melody with the Cs and Vs of reduplicatingmorphemes, constitute the
7 Adaptingthe proposals of an earlier version of this article, McCarthy(1981, 412-413) suggests that
reduplicatingskeletal affixes should carrya feature [ +reduplication],which "has the effect of causing auto-
matic copying of all the melodic elements in some morpheme-formally, all the daughtersof some > in a
particulartier." This copying process may appearto require "transformationalpower", but, as McCarthy
pointsout, "The copyinginducedby the presenceof the feature[ + reduplication]is partof universalgrammar,
not part of some language-particular reduplicationtransformation,and consequently it is irrelevantto the
whole problemof restrictiveness."In connectionwith his discussionof the [ + reduplication]feature,McCarthy
(1981, 414) explains the formal distinctionbetween the sorts of reduplicationdiscussed in this article, all of
which involve the [ + reduplication]feature,and the sorts of "reduplication"he himselfanalyzesfrom Semitic
languages,which do not involve this feature.
446 ALEC MARANTZ

increased power in the morphological component necessary to handle reduplication

cross-linguistically.
(20) t akk i t akki t akk i
111 I ,* I|II I I I II = taktakki
CVC + CVCCV CVC + CVCCV
It seems unlikelythat any less machinerycould be sufficientto account for reduplication
phenomena.
Note that, once we assume reduplicationto be the affixation of a C-V skeletal
morpheme,phonemic melody copying as in (20) is requiredto yield the correct output.
If we simply attached the phonemes in the phonemic melody of the stem to the added
C-V skeleton, association lines would cross in violation of the basic condition on
autosegmentalphonology given in (10).8
(21) t akk i

CVC - CVCCV
If we copy the entire phonemic melody of the stem over the C-V skeletal redupli-
cating affix, how do we ensure that the correct phonemes are asoqciatedwith the ap-
propriate slots in the skeleton? Four general conditions on the linking of phonemic
melodies with C-V skeleta, two of which we have alreadyencounteredin our discussion
of the Arabic verbal system, predict the correct association for mq$treduplicativepro-
cesses.
ConditionA: Unless overriddenby a special proviso, feature complexes containing
the feature [- syllabic] can be linked only to C slots in the skeleton, and feature
complexes containing the feature [ +syllabic] can be linked only to V slots in the
skeleton.
ConditionB: After as many phonemes as possible are linked to C-V slots one-to-
one in accordance with other conditions and principles, extra phonemes and C-V
slots are discarded. There is no multiple attachmentof phonemes to C-V slots or
of C-V slots to phonemes.

* Pi P2 * p

{1; {} {A;}
ConditionC: The slots in a C-V skeleton may be preattachedto distinctivefeatures.

8
As an LI reviewer points out, even if we allowed association lines to cross in reduplication,special
stipulationswould be necessary to demandjust the sort of "nested" crossingillustratedin (21) and to rule out
other conceivable association line crossings (in particular,crossings which would allow the mirror-image
reduplicationrules schematizedin (3)).
RE REDUPLICATION 447

These features take precedence over the featuresof any phonemesfrom a phonemic
melody which may link to these slots.
ConditionD: (i) Linkingof the phonemicmelody to the reduplicatingskeleton either
begins with the leftmost phoneme of the melody linking to the leftmost C-V slot
in the skeleton eligible under ConditionA and proceeds from left to rightor begins
with the rightmostphoneme of the melody linkingto the rightmostC-V slot of the
skeleton and proceedsfromrightto left. In the unmarkedcase, reduplicatingprefixes
associate with their melodies from left to right, reduplicatingsuffixes from right to
left. (ii) The association of phonemic melodies and C-V reduplicatingaffixes is
"phoneme-driven"in the sense that, for each phoneme encountered linkingfrom
left to rightor from rightto left, the association procedurescans along the skeleton
to find a C-V slot eligible for association with the phoneme under ConditionA.
The first condition constrainingthe linking of borrowedphonemic melodies in re-
duplicationto the reduplicatingC-V skeleta, ConditionA, prevents the association of
a [ - syllabic] phoneme to a V slot or a [ + syllabic] phonemeto a C slot. This condition
is just (17) above, which was motivated in connection with the use of C-V skeleta in
the analysis of Arabic verbs.
Agta provides the clearest example of the need for Condition A in reduplication.
Recall that Agta forms plurals by initial CVC reduplication(see the examples in (9)).
However, when the stem begins with a vowel, only the initial VC is copied.
(22) a. takki 'leg'
takki takki
III 1 11 = taktakki'legs'
CVC + CVCCV
b. uffu 'thigh'
uffu uffu uffu uffu
[[I|
| | | = ufuffu 'thighs' *| | *uffuffu
CVC + VCCV CVC + VCCV
u ffu uf fu
*N\\ t|III = *wufuffu
CVC + VCCV
c. ulu 'head'
ulu ulu ulu ulu
III = ululu 'heads' *| u | = *u1uu1u
CVC + VCV CVC + VCV
ulu ulu
*h\ IIII = *wu1u1u
CVC + VCV
ConditionA prevents the attachmentof the u to the C slot in (22b),for example, yielding
perhaps *uffufu or *wufuffu.
448 ALEC MARANTZ

Dakota provides further evidence that the linking of phonemes with the C-V re-
duplicating skeleton respects their ? syllabicity (data from Shaw (1976); see also the
discussion of Dakota reduplicationin section 2). Dakota suffixes a CCVC reduplicating
skeleton to verbs to form their plurals, but only the final (C)CV of a V-final stem is
copied.
(23) a. sic 'be bad'
s 1c s' 1 c
I II III = 'siksic'be bad, pl.'
CVC + CCVC (c > k by phonologicalrule)
b. h,aska'be tall'
haska haska
i tI t a /// = h,askaska
CVCCV + CCVC 'be tall, pI.'
haska haska
* | j | | | j/// = *haskaaska
CVCCV + CCVC
haska haska
= *haskaskaa
CVCCV + CCVC
Given that a phonemic melody associates with the slots in a suffixal skeleton from right
to left (see ConditionD), ConditionA accounts for the phonemic shape of the redupli-
cated portions in (23), rulingout the starredassociations in (23b).
Unlike Condition A, Condition B on the one-to-one linking of phonemes to C-V
slots is not operative in the association of phonemes with C-V skeletal morphemesin
the Arabic verbal system. Although it does not appear to be a general constraint on
autosegmentallinking, Condition B does find ample motivation in the analysis of par-
ticular reduplicationprocesses. It is ConditionB that ensures the properassociation of
the Dakota CCVCreduplicatingsuffix with the melody S'icas shown in (23a), prohibiting
the association displayed in (24).9
(24) 's sic
*CVC1A|+ CCVC
I = *sikssic
or *sissic (via the general rule of Consonant
Cluster Simplification;see Shaw
(1976, 331))
or *sisic (via Consonant Cluster Simplification
and Degemination;see Shaw (1976, 339))
9 ConditionB may also allow us to capturethe fact that Sanskritinitialreduplicationcopies only one
consonantof an initial cluster.
(i) dada - da caskand - skand
bibhi - bhi susru - sru
paprach - prach qiqlis - clis
RE REDUPLICATION 449

ReduplicatingC-V skeleta, like the binyanimskeleta of the Arabic verbal system,

may have features or phonemes preassociated with C-V slots, a possibility formally
encoded in Condition C. For example, Akan copies an initial CV of verbs to form
"multipleactivity" or "'multiplestate" forms of the verbs (see Schachterand Fromkin
(1968)). The V of the copy is always a [?+high]version of the first stem vowel.'0
(25) s e2 se S 3? sc ?
II |I = sise? II = sus:?
CV + CVC CV + CVC

[+ high] [+ high]
se? 'say' s:9 'light'
The preattached[ + high]featurein (25) takes precedenceover any [ high] specification
of the phoneme associated with the V slot of the reduplicatingprefix.
Yoruba (Delano (1965)) forms nouns from verbs by prefixing a CV reduplication
skeleton whose V is fixed to i (these examples were providedby Doug Pulleyblank).
(26) lo lo diun d'un
|Ij |j = lilo l | = did-un
CV + CV CV + CVC
l.
1
l
i
lo 'to go' dun 'to be tasty, sweet'
lilo (nominalized) diduin(nominalized)
I have assumed here that the case of a preattachedphoneme in a reduplicatingskeleton
is simply a limitingcase of preattachedfeatures. Whilein Akan only the feature [ + high]
is preattachedto the V of a C-V skeleton, in Yoruba the complete set of distinctive
features necessary to yield the phoneme lil is so attached. Although a vowel from the
stem's phonemic melody links to the V slot in the reduplicatingprefix in Yoruba, all of
its features are overridden by preattachedfeatures. One might assume, on the other
hand, that when a full set of features is preattached to a slot, no phoneme from a

The reduplicatingprefix in these Sanskritexamples is CV. By prohibitingthe attachmentof more than one
consonantalphoneme to a single C slot, ConditionB insists that only one memberof an initial cluster be
copied.
(ii) a. sru s ru sru s ru
I I I II > susru *V I III
CV + CCV CV + CCV
b. skand skand skand s k a nd
I| 1111I I > caskand I I Ill
CV + CCVCC CV + CCVCC
Althoughit seems clear that the least sonorousmemberof an initialcluster has priorityfor linkingto the sole
C slot of the CV reduplicationprefix (see Kiparsky(1979)), how to state this conditionformallyremainsa
[Link],what is requiredfor Sanskritis a markedphonemeto C-V skeletonlinkingrule, in addition
to ConditionsA-D.
'? An LI reviewernotes that a similarsituationexists in Nupe.
450 ALEC MARANTZ

phonemic melody may associate with the slot. McCarthymakes such an assumption
in his analysis of Arabic verbs (see above). The two different approaches to the pre-
attachmentof phonemes make differentpredictionsfor certainsituations,none of which
I have been able to find in real-languagedata, [Link] example the
association of the phonemic melody /tasidu/ with the C-V skeleton CVCCV, to which
In! has been preattachedin the second C slot. If phonemes may link to slots to which
a full set of featureshas been preattached,the associationwill yield tani-(given Condition
D, to be discussed below), as in (27a); if phonemes may not attach to such slots, the
association will yield tansi-, as in (27b).
(27) a. t as i du t as i du
III \ 111111 = tani-tasidu
CVCCV + CVCVCV

b. tasidu tasidu
I I \\ 11111 I = tansi-tasidu
CVCCV + CVCVC V

n
A final constraint on the association of phonemic melodies with C-V skeletal re-
duplicatingaffixes, Condition D, determines which phonemes and C-V slots are dis-
cardedwhen there are not enough C-V slots to link to all of the phonemesor not enough
phonemes to link to all of the slots. It is ConditionD, workingin consort with Condition
B, that allows us to copy the entire phonemic melody of a stem over the reduplicating
C-V affix although only part of the stem may be reduplicated. We need no special
(presumably transformational)mechanism to copy just the correct phonemes of the
stem; it suffices to specify the C-V shape of the reduplicatingaffix, which, as we have
seen, is independentof the shape of the stem to which the affix attaches.
All of the reduplicationprocesses illustratedto this point (except the Sanskrit re-
duplicationrule discussed in footnote 9) have employed the linkingprocedureunmarked
under Condition Di. For an example of unmarkedlinking, consider the Dakota redu-
plicationprocess illustratedin (23). Since this is a suffixingreduplicationrule, association
begins from the rightmostphoneme in the copied phonemic melody and proceeds left-
ward. Associating in the opposite directionyields the incorrectresult displayedin (28b).
(28) a. haska haska
l l /7/ = haskaska
CVCCV + CCVC

b. haska haska
* | | | | | | \\ = *haskahas
CVCCV + CCVC
RE REDUPLICATION 451

I have found sporadicexamples of linkingmarkedunderConditionDi in my survey

of reduplicatingrules. Consider, for example, the Chukchee reduplicationillustrated
earlier in (9). The CVC reduplicatingsuffix in Chukchee links to its phonemic melody
from left to right, the markedlinkingaccordingto ConditionDi.
(29) a. nute- 'earth, ground'
nute nute nute nute
|H| | | ||
(H = nute-nut *1111 /7 = *nute-te
CVCV + CVC 'abs. sg.' CVCV + CVC
b. inu- 'part of reindeerleg'
mnu mnu mnu i nu
I| I \\ inu-un *1I /I = *inu-nu
VCV + CVC 'abs. sg.' VCV + CVC
(after
Vowel
Assimilation)
As shown in (29), unmarkedlinking of phonemes to skeletal slots produces the wrong
results in Chukchee." Madurese (Stevens (1968, 34)) provides an example of a redu-
plicating prefix which links to phonemic melodies from right to left. One use of this
prefix is to form plurals.
(30) buiwaiq-ain 'fruit'
bu'wa'q buwaqq-an
/ I I I I I = waq-bu'waqa'n
CVC + CVCVC- 'fruits'

bu'wa'q buwaq- an
IIII I = *bui(w)buwaiqain
CVC + CVCVC-
Again, unmarkedlinkingof skeleta to melodies yields the wrongresultfor this Madurese
reduplicationprocess. 2
Condition Dii proves necessary for reduplicationrules such as the CVCCV(C)
prefixing process in Tagalog discussed above (see examples (7e,f)). Phoneme-driven
association ensures that the linking(31a) results, ratherthan the linking (31b).'3
It appearsthat Tzeltal (Berlin (1963))also contains several reduplicatingsuffixes which link to their
phonemicmelodies from left to right, the markedlinkingaccordingto ConditionD.
12
Another example of markedlinking for reduplicatingprefixes is found in Til (Reichard(1959)). Til
prefixes a reduplicatingC to form a sort of pluralor [Link], the phoneme which links
to this skeletal C is the last, not the first, consonantin the stem.
'" The use of the feature[ ? long] in (31) is [Link] arguedagainstthe
existence of [ t long] as a distinctivefeature. Withinthe tiered phonologyframeworkassumedin this article,
long vowels have been analyzedas a single + syll phonemelinkedto two V slots at the C-V skeletal tier:
(i) p
I+ syll
A
V V
452 ALEC MARANTZ

(31) a. talinoh 'intelligent'

talinoh tallnoh
III \I I I = talitaltnoh 'rather
CVCCV + CVCVC C intelligent'

[ + long]
b. talinoh talfnoh
I I1/ I/ I I I IH
I = *taln6talinoh
CVCCV + CVCVCVC

+ long]
When the association procedurereaches the i in the phonemicmelody, talfnoh, it scans
until it finds a V slot in the C-V skeleton with which to link the i, skippingover a C in
the skeleton.
Nash (1980)shows that ConditionDii on the association of phonemicmelodies with
C-V reduplicatingaffixes accounts for an otherwise puzzling fact of Warlpiriredupli-
cation. Verbal reduplication,forming a sort of intensive of the verb, normally copies
the initial CVCCV of a stem or preverb but copies only the first CVV when the first
vowel of the stem or preverb is long. Assuming that long vowels are represented as
sequences of two Vs in Warlpiriat the C-V skeleton level, simply writingthe redupli-
cating prefix as CVCCV accounts for the difference in the copied material between
stems with initial long vowels and those with initial short vowels (the diagramsin (32)
are adapted from Nash (1980, 143)). (In the orthographyNash employs, rn and rl rep-
resent single phonemes.)
(32) a. pakar ni paka r ni
III \ I I /I = pakapakarni
CVCCV + CVCVCV
b. t i i rl t i i rl parnka ja
I1I1\ III I I1 I/1//
CVCCV + CVVC + CVCCVCV
= tii-tiirl-parnka-ja
'(ground)split
lengthwise (by tuberunderneath)'

Because the linkingof phonemesto C-V slots in reduplicationis "phoneme-driven",as specifiedin Condition
D, we cannot give the Tagalog reduplicatingprefix in (31a) the skeleton CVCCVV, avoiding the [-+-long]
feature. Accordingto the linkingprinciples,a CVCCVVskeleton would yield the form *taliotalTnoh in (31a).
One mightreplace the final V in the skeleton in (31a) with the followingnotation:
(ii) A
V V
This notationwould be read to indicatethat a single phonememust attachto both of the connected V slots,
as in (i). Since it is tangentialto the main points of this article, I will not pursuehere the analysis of vowel
length in Tagalog.
RE REDUPLICATION 453

According to ConditionDii, the association proceduremust scan the C-V skeleton for
a V slot for the second i in (32b), skippingover both C slots and yielding the correct
result, tiitiirlparnkaja.'4

1.4. Syllabic and Whole Morpheme Reduplication15

The theory presented so far accounts only for reduplicationprocesses which may be
characterizedby a C-V skeleton whose make-upis independentof the constituentbeing
reduplicated. However, other sorts of reduplicationprocesses do exist. In addition to
the reduplicationof entire morphemesor stems, which is a fairly common phenomenon
among the world's languages, at least one language,Yidiny,exhibits a syllable copying
reduplicationrule. The examples of Yidinyreduplicationpresentedin Dixon (1977)and
discussed in Nash (1979; 1980) clearly indicate that Yidiny reduplicates the first two
syllables of a stem regardlessof the C-V make-upof these syllables. Nominal redupli-
cation forms plurals (Dixon (1977, 156)).
(33) a. qimurU 'house' cimu4imurU'houses'
b. gindalba'lizardsp.' gindalgindalba'lizards'
Note that the r of cimurU, which belongs to its thirdsyllable, is not reduplicated,while
the 1of gindalba, which closes its second syllable, is copied. Thus, neitherof the skeletal
prefixes CVCCVCor CVCCV will account for Yidinyreduplication;whether or not the
final C in the first CV(C)VCof a word is reduplicateddepends on the syllabificationof
the word. Verbal reduplication,forming an intensive or repetitive of a verb, has the

14
Althoughdetails remainto be worked out, the analysis of reduplicationpresentedin this section can
[Link] less commonin my survey of reduplicativeprocesses than
initialand final reduplication,clear cases of internalreduplicationdo exist. For example, Samoanconstructs
pluralforms of some verbs by addinga copy of the CV of the penultimatesyllablejust before (or just after)
this CV (Marsack(1962)).
(i) singular plural
alofa alolofa 'love'
galue galulue 'work'
maliu maliliu 'die'
Just as I claimedthat initialand final reduplicationare the prefixationand suffixation,respectively,of a C-V
skeletal morpheme,so I would claim that internalreduplicationis the infixationof a C-V skeletalmorpheme.
As with any infixingprocess, the problemfor an analysisof internalreduplicationis how to specify where in
a stem the infix belongs. Once we place the C-V skeletalinfix in the correctlocationin our Samoanexamples,
the machineryand conditionsalreadyintroducedproducethe properresult.
(i) a Ilofa a lofa lofa
|'
I I I I I |- I II I I| = alolofa
V + CV + CVCV V + CV + CVCV
15 McCarthy(1979; 1981)provides a somewhatdifferentanalysis of syllable and full morphemeredupli-
cation from the one outlined in this section. I do not present an alternativeto McCarthy'sanalysis because
I believe there is evidence to decide between the approaches;rather,the analysis of this section represents
an attemptto unify syllable and full morphemereduplicationwith the C-V skeletal reduplicationprocesses
discussed above.
454 ALEC MARANTZ

same formal analysis as nominalreduplication-it also copies the first two syllables of
a stem regardless of their C-V make-up(Dixon (1977, 233-236)).I6
(34) a. 4a4ama-n jump' 4acaacjadama-n'jumpa lot'
b. c4ugarba-n'have an cugarqugarba-n'have an unsettled
unsettled mind' mindfor a long period'
An extension of the analysis of reduplicationpresented in the above paragraphs
allows for syllabic reduplicationwithout greatly increasingthe power of the morpho-
logical component. It also permits us to connect C-V skeletal reduplicationprocesses
with full morpheme [Link] us suppose, following McCarthy(1979) among
others, that the structure of a morpheme is hierarchical,includingat least the levels
shown in (35).
(35) phonemic melody p p p p p p p

C-V skeleton
I I II I I I
C V C V C C V

syllabic skeleton u a

morphemesymbol 11
C-V skeletal reduplicationaffixes a C-V skeleton to a stem with a structurelike (35)
and borrows the phonemic melody of the stem.'7
(36) ba ri bari ba ri

CVC + CV CV CVC + CV CV

(r (f ar (F (r (r

\ VIL Ij V
We may analyze Yidinysyllabicreduplicationas the prefixationof a syllabic skeleton

16 One apparentexception to the principlethat Yidinyreduplicatesthe first two syllables of a stem is

successfully explained in Nash (1979). A nasal apparentlyclosing the second syllable of a stem which is
homorganicwith a following stop is not copied.
(i) galambata'marchfly' galagalambara'marchflies'
mac!inda-an'walk up' ma4ima4inda-n'keep walkingup'
Nash arguesconvincinglythat such homorganicnasalsbelongto the followingsyllableandthat, if reduplication
copies the first two syllables of a stem, they are thus not expected to reduplicate.
17 Note that I have connectedthe Cs and the V of the CVCreduplicating prefixin (36)to a syllablesymbol
u. SouthernPaiute provides some evidence that the C-V slots of skeletal morphemesshould be organized
into syllablesin this manner.(For the datathatfollow, I rely on Pranka(1981).However,the analysissuggested
here is due to a conversationbetween MoiraYip and the author.)Settingaside certaincomplications,we may
say that SouthernPaiute copies the first (C)V of a stem plus the next C only if it is nasal. If we postulatea
CVC skeletal prefix for SouthernPaiutewith the Cs and the V attachedto a single r, we predictexactly this
behavior. Since, accordingto Sapir(1930, 37), only nasal Cs may close a syllable in SouthernPaiute, only a
nasal second consonantfromthe stem's phonemicmelodymay attachto the second C slot in the reduplicating
RE REDUPLICATION 455

to a stem, as shown in (37). The syllabic skeleton, lacking a phonemic melody and a
C-V skeleton, borrows both from the stem to which it attaches.
(37) a. d,i mu rU 4i mu rU 4i mu rU

CV I CV C V CyYV CV CV

a a + c r C (Y Y + cr cr

\V \V \/

b. gi-n dat ba gi ndal ba gi ndal ba

I1I 111I 11 11 I1111 11 I1 111 11
CVCCVC CV CV CCVC CV CV CCVC CV

or or + cr
~\V
c (F
\/ V4VVW
cr (x + cr ar c

We must assume that the Cs and Vs of the stem in syllabic reduplicationare copied
clustered in the syllabic units that they form in the stem. If the copied Cs and Vs were
not joined together in these syllabic units, the reduplicatedform of cimurU would be
*4imurqimurU,with the onset of the root's second syllable copied as the coda of the
second syllable of the prefix:
(38) d,i mu rU cUi mu rU 4i mu rU

CVCVCV CV CV CV CV CV
- VV V V
(J Uf + (x ff a (x cs + Cy cr 0

di mur U cqi mu rU
I II VI CV
II CVC II
I I CV
II CV
C7 *(qimur4imurU

+
(ra
. +yy cr a (

prefix. Since a nasal C beginningthe second syllableof a stem will be copied, it is not the first syllableof the
root which is copied in [Link],SouthernPaiutecopies a C anda V plus another
C, if this may form a syllablewith the first C and the V. Withinthe theorypresentedhere, we cannotaccount
for the Southern Paiute facts by preattachingthe feature [+nasal] to the second C slot in a CVC prefix.
Accordingto our linkingconditions, such a prefix would effect the copying of the first C and the first V of
a stem plus a nasalizedversion of the second stem consonant.
456 ALEC MARANTZ

If I am correct in unifying C-V and syllabic reduplication,we must see the linking of
syllabic clusters to the syllable slots labeled u as parallelto the linkingof phonemes to
C and V slots.
Following the above line of analysis, full morpheme reduplicationbecomes the
additionof a morphemicskeleton to a stem. The morphemicskeleton, lackinga syllabic
skeleton, a C-V skeleton, and a phonemic melody, borrows all three from the stem to
which it attaches.
(39) kur du kur du kur du
111 11
CVC CV>CVCCV
111 11 CVC 11
111CV
cr
9i (f ar k'YA'(J ff f

F + L . +

Note how normalaffixationfits naturallyonto this continuumfrom total morpheme

reduplicationthrough C-V skeletal [Link] affixationis the addition to
a stem of a morphemewhich has a C-V skeleton and a phonemic melody of its own
(but perhaps not a syllabic skeleton if the affix cannot be syllabified in isolation); the
morphemeborrows nothing from the stem to which it attaches.

(40) efn j oy + me n t en joy +menT

lIi
11I1I1 CVCC
VC CVC
11111 II'I'1
II CVCC
VC CVC

(r a (r a~~~c(T cr

Although the above analysis of a syllabic reduplicationprocess leads to an elegant

unification of C-V skeletal reduplicationwith full stem reduplicationand of both of
these with normal affixation, it nonetheless leaves us with a mystery. Why, of all the
reduplicationprocesses studied by Moravcsik, myself, and others, is there only one
clear example of syllabic reduplication(namely, Yidiny)?

1.5. Conclusion
In this section I have supportedan analysis of reduplicationas the affixation(or infix-
ation) of a skeletal morpheme. I have shown that the analysis is able to formalize
reduplicationprocesses of the world's languagesand to account for many of their prop-
erties. Most of the formal apparatusemployed in the analysis finds independentjusti-
fication in the analysis of the Arabicverbalsystem (forfurtherexamples of the operation
of "tiered phonology", see Halle and Vergnaud(1980a,b)).The one mechanismadded
to the grammarspecifically for reduplicativeprocesses is the wholesale copying over
RE REDUPLICATION 457

the reduplicatingmorphemeof an autosegmentaltier or tiers-e.g. the phonemic mel-

ody-from the stem to which the reduplicatingmorpheme affixes. By avoiding the
exploitation of a full transformationalnotation, the present analysis explains why re-
duplicativeprocesses like those illustratedin (3) are not found in any [Link] mimic
the processes in (3) employingthe grammaticalapparatusdescribedabove would require
crossing association lines in violation of the fundamentalconstrainton autosegmental
representations,(10).
(41) pP 4 1 P2 P3 P4

VC+C V C V

2. The Interactionof Reduplicationand PhonologicalRules

In the previous section, I introducedand extensively supportedan analysis of redupli-
cation according to which reduplicatingprocesses are simply affixation. This account
runs up against the problems for analyses of reduplicationin generative grammardis-
cussed at length in Wilbur(1973).Reduplicationis generallya derivationalor inflectional
process used to encode, for example, the intensive of verbs or the pluralof nouns. If
we consider reduplicationto be a normalword formationprocess, as in section 1, we
expect it to precede all phonologicalrules in the derivationof a reduplicatedform (cf.
Aronoff (1976)),and we expect reduplicatedforms to be subjectto all phonologicalrules
just as any derived or inflected word would be. In fact, as Wilburdocuments, certain
phonologicalrules in a variety of languagesappearto over- and underapplyto redupli-
cated forms. Reduplicatedforms are not entirely aberrant;most generally applicable
phonologicalrules apply to them exactly as they apply to any morphologicallycomplex
word. However, there seems to be a class of phonologicalrules which treat reduplicated
forms specially.
In Dakota, for example, morpheme-initialvelars are palatalizedfollowing a prefix
ending in i (Dakota data from Boas and Deloria (1941)and Shaw (1976)).
(42) k'a 'to mean' nic"a'he means thee'
kNi 'to give' nic&ui
'he gives to thee'
In reduplicatedforms, the initialvelars of both the reduplicatedportionand the original
root palatalize, whether or not the second velar is preceded by i.
(43) -nape kicoscoza 'he waved his hand to him' (from koza 'to wave')
Palatalizationseems to have overappliedto the second velar in (43), doing its work even
though its structuraldescriptionis not met.
The overapplication of a phonological rule to reduplicatedforms can always be
handled by ordering reduplicationafter the phonologicalrule in question. The output
of the rule then reduplicateswith the rest of the morpheme,with either the originalor
the copy sometimes ending up in an environmentwhich would not have triggeredthe
458 ALEC MARANTZ

rule. One problem with this solution is that reduplicationhas all of the properties of a
regularword formationrule and regularword formationrules can be orderedto precede
all phonological rules. Orderingreduplicationafter certain phonological rules implies
that one can place a derivationalor inflectional affixing rule somewhere in the middle
of the phonology, an option that is, apparently,not otherwise needed.'8
In addition to loosening constraints on the organizationof grammar,the ordering
solution to the overapplicationof phonological rules to reduplicatedforms cannot be
extended to explain the underapplicationof certainrules to these forms. To take another
example from Dakota: certain word-finalas in Dakota change to e before a numberof
morphemesincludingthe phrase-endingmorpheme,/?.
(44) a. haska'9 'to be tall'
b. cha-kiiyuihaha'ske-?'all the trees are tall'
However, the final a of at least a certain class of reduplicatedverbs does not change
to e before these same morphemeseven when the final a of their unreduplicatedroots
does change. Thus, despite the fact that the final a of ha'skachanges to e before _?, as
shown in (44b), the final a of its reduplicatedform, ha'ska-ska,does not change to e
before -?, as shown in (45).
(45) cha-kihaska-ska-?'the trees are tall'
If reduplication were ordered after the rule changing a to e, we would expect
*haske-ske-?in (45); if before, we would expect *haska-ske-2 Since rule orderingpro-
vides no explanation for the underapplicationof rules to reduplicatedforms, it is a
questionable solution to their [Link] would expect the same analysis to
cover both cases of irregularrule interaction.
Wilburherself offers a reason for the fact that rules over- and underapplyto redu-
plicated forms. She attributesthis special behaviorof reduplicatedforms to the Identity
Constraint(Wilbur(1973, 58)):
(46) The Identity Constraint
There is a tendency to preserve the identity of R, [what is copied in redupli-
cation] and Rr [the copy] in reduplicatedforms.
Wilbur suggests that the Identity Constraintmay be realized as a global condition on
the rules which over- and underapplyto reduplicatedforms. A rule which overapplies
would be written to apply both to a segment in Ro and to the correspondingsegment in
Rr if the rule's environmentis met for the segment in Ro. A rule which underapplies
would be written to apply to a segment in Rr only if the correspondingsegment in Ro

18
An anonymousLI reviewerhas pointedout that, even settingaside reduplication,some morphological
rules have been arguedto "follow" phonologicalrules in some sense. I know of no convincingargumentsto
this effect consistent with the restrictivetheories of phonologyand morphologyI am assuminghere (but see
Anderson(1975)).
19
A vowel with a superimposedcomma, e.g. q, representsa nasalizedvowel in this orthography.
RE REDUPLICATION 459

also occurs in the [Link] rule would thus fail to apply when a
segment X in Rr but not the correspondingsegment X' in R, appears in the right en-
vironment.
The difficulty is that the Identity Constraintexplains nothing. Wilburobserves that
rules appear to over- and underapplyto reduplicatedforms and invents a constraint
which merely encodes this fact. Many rules apply "normally" to reduplicatedforms;
that is, they apply wherever-and only where-their environments are met. For ex-
ample, a rule of Devoicing in Dakota, which devoices fricatives before boundaries, is
responsible for the s in example (43), repeated here as (47).
(47) kicoscoza (from koza 'to wave')
Although they destroy the identity of R, and Rr, rules like Dakota Devoicing (given as
(62) below) do not violate the IdentityConstraint,which does not insist that phonological
rules preserve the identity of R0 and Rr but merelyallows that they might.A real solution
to the applicationproblemwill explain why only certain rules and not others over- and
underapplyin reduplicatedforms. If we can make the behaviorof rules with respect to
reduplicatedforms follow from an independentlymotivatedtheory of phonologywithout
keyingon the duplicativenatureof reduplication,we will have explainedthe phenomenon
instead of merely remarkingupon it.
The Identity Constraint as formalized by Wilbur makes an empirical prediction
regardingthe possible applicationof phonologicalrules to reduplicatedforms which data
from Karok (Bright (1957)) actually disconfirm. Therefore, Wilbur's solution to the
applicationproblemcan be rejectedon empiricalas well as explanatorygrounds.20Karok
forms a derived intensive verb indicatingthe repetitionof a short action by suffixing a
CVC reduplicatingskeletal morpheme.
(48) parak'to separatewith a wedge' parak-rak'to split logs with wedges'
tasif 'to brush' tasin-sif 'to brush(repeatedly)'
(In (48), f is a morphophonemewhich nasalizes to n before a consonant.) The morpho-
phoneme that Brightwrites as v deletes between an a or o and a consonant-initialsuffix.
In the reduplicatedforms of {a}v-final stems, however, neither the v before the re-
duplicating suffix nor the final v of the suffix deletes, even when a consonant-initial
suffix is added to the reduplicatedform.
(49) 2u*mxavxam(< ?u*mxav)'to pull up by the roots'; -tih 'durative'
?u-mxavxa'ivtih'to be pullingup by the roots'
When v (but not v) comes to stand between a(-) or o and a consonant, in that order,
it is replacedby lengtheningof the precedingvowel (if not alreadylong) . . . The only
exception occurs in reduplicatedforms, where v is always retained.(Bright(1957, 34))
Thus, v Deletion appears to underapply in reduplicated forms. However, Wilbur's Iden-
20
Tagalog presents another empirical problem for Wilbur's proposals; see section [Link].
460 ALEC MARANTZ

tity Constraint cannot account for the underapplicationof v Deletion in Karok. The
Identity Constraintblocks the applicationof a rule to a segmentor segments in the copy
portion of a reduplicated form when the correspondingsegment or segments in the
originalare not in the properenvironmentfor the rule. In 2u mxavxa vtih, both the stem
and copy v stand between an a and a consonantand thus shoulddelete underthe Identity
Constraint to yield *Pu-mxa xa tih.2`
I will demonstrate that the behavior of phonological processes with respect to
reduplicationis predicted by currenttheories about the organizationof phonology and
the lexicon. To account for the data, all we need to assume is our conclusion from
section 1-that reduplicationis simply derivationalor inflectionalaffixation. To switch
perspective, given that my data are correct, the behavior of reduplicatedforms will
provideconsiderablesupportfor the currenttheories of phonologyand the lexicon which
predict this behavior. The strategy of this section, then, is to show that the apparent
problems and paradoxes associated with the interactionof reduplicationwith phono-
logical processes are in fact pseudoproblemsand [Link] the grammars
of reduplicatinglanguages are examined with care, difficulties surroundingthe inter-
action of reduplicationrules with phonologicalprocesses disappear.
The cases of apparentover- and underapplicationof phonologicalrules to redupli-
cated forms fall into two classes. In the first, we find phonologicalprocesses that do not
apply within the reduplicatingaffix (the "copy") althoughtheir environmentsare met
there. It will be shown (section 2.1) that in Luisenlo,the rule in question is a cyclic rule,
which, accordingto currentinterpretationsof the cycle (see e.g. Mascaro(1976), Halle
(1979)), should not apply within morphemesin [Link],we
should not expect it to apply within the reduplicatingmorpheme.
The second class of cases exhausts the remainingtypes of examples of over- and
underapplicationof phonologicalprocesses to reduplicatedforms. For Dakota and Ta-
galog, I will suggest that these processes meet the criteriato be consideredmorpholexical
rules in the sense of Lieber (1980). That is, they are not phonologicalrules at all, but
rather rules which express the relationshipsbetween allomorphsof morphemes, both
of which are listed in the lexicon. If the phonologicalprocesses in question are mor-
pholexical rules and reduplicationis consideredan affixingword formationrule like any
other affixing rule, then we can explain the interactionof these processes with redu-
plication without any extra [Link] fact, Lieber's theory predicts the appearance
of under- and overapplicationof morpholexicalrules in reduplicatedforms; it prohibits
what would look like the "normal"applicationof these rules to such forms. Since both
the "input" and "output" of a morpholexicalrule are listed in the lexicon, reduplication,
an affixing process, must build on one of these listed allomorphs. If it builds on the

21
In an earlierversion of this article, I includedan analysisof Karokreduplicationin section 2.2. There
I providedevidence that v Deletion between {a, o} and a consonantis "morpholexical"in the sense to be
defined below and that, therefore,apparentunderapplicationof v Deletion to reduplicatedforms is expected
(see section 2.2 for a discussion of the relationshipbetween the morpholexicalcharacterof a rule and its
apparentover- or underapplicationto reduplicatedforms).
RE REDUPLICATION 461

pseudoinputto the morpholexicalrule, the illusionof underapplicationresults; if it builds

on the pseudooutput,the result is the illusionof [Link] take up this second
class of cases in section 2.2.22

2.1. Reduplication and Cyclic Rules: Luiseho23

Munro and Benson (1973, 16) note that "the surface phones c and s of Luiseiio are in
complementarydistributionsuch that [(50) holds of them]":

(50) c - s/ {[cont]}

The grammarof Luisenlo includes a regular derivation process, schematized in (51),

which forms moderative adjectives from verbs by reduplication.
(51) CIVIC2V2 CIV1C2V2 + C1C2V2 + S
kava'be red' ?ava'?vas'pink'
maha 'to stop' mahamhas'slow'
Under the analysis of reduplicationpresented in section 1, we would say that Luisenlo
forms moderative adjectives from verbs by suffixingthe C-V skeletal morphemeCCV
(or CCVC). The apparentproblemconcerningthe interactionof phonologicalrules and
I
c
reduplicationin Luiseinois this: when the initial consonant of a verb is c",it does not
become s in the adjective-formingreduplicatingsuffix even though it immediatelypre-
cedes a consonant. That is, c s appears to underapplyin reduplicatedforms.
(52) 'cSra'to tear' cara'cras'torn'
coka 'to limp' cukdc'kas'limping'
What prevents c s from applying in the reduplicatedforms of (52)? Wilbur (1973)
claims that c s- > is subject to the Identity Constraint,(46). This constraintblocks the
22
1 will not discuss here all of Wilbur's(1973)examplesof the over- and underapplication of phonological
processes to reduplicatedforms, but will insteadrestrictattentionto those languagesfor which sufficientdata
exist to answercrucialquestionsaboutthe phonologicalprocessesinvolved. As will become clear in the pages
to follow, an explanationof the apparentover- and underapplicationof phonologicalprocesses depends on
determiningpropertiesof these processes which often cannotbe extractedfromeven a fairlycarefulgrammar
of a [Link],for the most part,simplyaccepts the analysesof her sources, writingthata phonological
process is a rule, for example, if her sources claimthatit is. To understandmoreclearlywhy not all of Wilbur's
(1973)examples are treatedhere, considerthe case of the underapplication of Palatalizationin Akan redupli-
catedforms. In an earlierversionof this article,I claimedthatPalatalizationunderappliedwithina reduplicating
prefix in Akan because it was a cyclic rule. Reviewing Wilbur'ssource on Akan (Schachterand Fromkin
(1968))more carefullyin preparationfor rewritingthe article, I realizedthat there was no evidence for the rule
of Palatalizationat all; it was simplya device used to reducethe underlyinginventoryof [Link] state
of the art was such in 1968that one freely exploitedrule orderingand lexical exceptionsto replacephonemes
with rules.
23 Aronoff (1976) and McCarthy(1979), among others, have arguedfrom the Luisenio
data to different
conclusions. However, they were not aware of the crucialdata exemplifiedin (53) and (54), or at least they
do not mention such data.
462 ALEC MARANTZ

rule in the copy (Rr) of the reduplicatedform because the c in the original(R,) does not
meet the environmentof the rule, that is, it does not immediatelyprecede a consonant.
The failure of c s- > to apply in the reduplicatedforms thus preserves the identity of Rr
and R0.
On the analysis of reduplicationgiven in section 1, we need not say anythingabout
the interaction of c --> . and reduplication in Luisenio. Since, as we shall see, the c -s
rule is "cyclic", in a technical sense, and since the c-C combinationsin the reduplicated
forms in (52) are entirely internal to the suffixal adjective-formingmorpheme, CCV,
currenttheories of the applicationof cyclic rules (see e.g. Mascar6(1976), Halle (1979))
predict that c -s ? will not apply within the copy portions of these reduplicatedforms.
Contraryto the claims of MunroandBenson (1973),c ands are not in complementary
distributionin Luisefio. In fact,,c -s is a neutralizingrule with a restricted range of
application. Davis (1976) indicates that there is one Luisefio word in which s precedes
a vowel, the exclamatorysox 'oh!' (cf. cara 'to tear', coka 'to limp'), but discounts this
as a counterexample to complementarydistributionbecause "exclamations in many
languagesfrequentlydisplay exceptional phonologicalcharacteristics"(p. 197).Kroeber
and Grace (1960) include an example of s before a [ + cont] consonant, x, but their
transcriptionof masxai 'isn't it?' might be a result of their failure to distinguishs from
X, a retroflex (see Davis (1976)).
(53) quawicxal 'Bloomeria aurea' masxai 'isn't it?'
pacxam- 'to launder'
If s'ox and mas?xaiconstitutedthe only examples of the breakdownin complementary
distributionbetween c and s outside of reduplication,it would be difficultto argue that
c s'is a neutralizingrule, that is, that s is an underlyingphonemein Luiseflo. Although
sox and mas"xaido suggest that s and c are not in complementarydistribution,what is
important for the present analysis is that c occurs in underived morpheme-internal
environments before the very consonants in front of which c -' > applies when the
context, [ - cont], is derived,
(54) a. po-xecla 'its point, of an arrow'24
but pu'sla 'eye, nom.', pucil 'eye, obj.'
mos-la-t 'belt' < moci 'to weave'
b. cacmis 'a stone tool'
but pa--awismi 'them of the water', pa--gawici 'him of the water'
nes-ma-l 'old woman' < ne-cu- 'to become an old woman'
Althoughpo-xecla appearsamong my sources only in Kroeberand Grace (1960), Bright
(1968) confirms their cacmis.
24
Kroeberand Grace(1960,22) suggestthatpo-xeela 'its point, of an arrow'is derivedfromxeei- 'strike'
and -la 'place of. However, this is undoubtedlyan historicalderivation,not a synchronicanalysis. First, note
that the meaningof po-xecla is not a predictablecombinationof the meaningof its constituentparts;rather,
it has developed a specialized usage. Moreover, as Kroeberand Grace point out, -la is not a "truncating
suffix";thatis, the finalvowel of xeei- wouldnot deletebefore-la if the proposedderivationwere synchronically
valid.
RE REDUPLICATION 463

Drawing on the work of Kiparsky (1973), Mascaro (1976) and Halle (1979) identify
as "cyclic" the class of rules which appearto apply only in derived [Link]
' fails to apply before certain consonants (e.g. before m in cacmis) when c appears
before these consonants in the underlyingform of a morphemebut applies before the
same consonants when the environmentis derived indicates that c -s S is "cyclic" in
the technical sense. Halle (1979, 18) formulates the condition on cyclic rules which
prevents them from applying in nonderived environments in a manner which blocks
applicationof a cyclic rule when the environmentfor the applicationof the rule is entirely
contained within the underlyingrepresentationof a morpheme,be it root or affix. Since
the c-[ - cont] combinations in the reduplicatedforms in (52) are entirely contained
within the reduplicatingsuffix, CCV, Halle's formulationof the conditionon cyclic rule
applicationcorrectly predicts that c > s will fail to apply within such forms.
One might argue that the environmentinternalto the reduplicationaffix in Luisefio
is a "derived" environment since the copying and linking of a phonemic melody in
reduplication"derives" the phonemicshapeof the affix. However, as long as the copying
and linkingprocesses in reduplicationare not phonologicalrules, the technicaldefinition
of the cycle in Halle (1979, 18) yields the correct results for Luisefio. The juxtaposition
of the c and the [ - cont] within the reduplicatingaffixes in (52) is not the result of some
phonologicalrule, nor would the applicationof c s- > withinthose affixes make "specific
use of information" outside the reduplicationaffix. Therefore, according to Halle's
definition of the cycle, c -* s should not apply in (52), if it is a cyclic rule.
To summarize,c > K fails to apply withinthe Luisenioreduplicatedforms discussed
above because c > s is a cyclic rule, the interior of the reduplicatingprefix is not a
derived environment, and cyclic rules apply only in derived [Link] that,
for our purposes, showing that c -s> applies regularlyin derived environmentsbut fails
to apply in nonderived environments was sufficient to demonstrate that the rule is
"cyclic". That c > s is "cyclic" in the sense of applying once in each phonological
cycle is of no importancehere. We do not even need a theory of the phonologicalcycle
to predict that c -s ' will not apply within the derived adjectives of (52). The forms in
(54) demonstratethat c -s> does not apply within nonderivedenvironments, and the
interiorof the reduplicatingsuffixes in the adjectivesof (52)is not a derivedenvironment.
Therefore, c s should not apply in (52). The theory of the cycle found in Halle (1979)
and the other sources cited above simply leads us to expect to find rules like the Luisenio
c- s, i.e. rules which apply only in derived environments;such rules need not be
considered problematic.
Knowledge of more of the phonology of Luisefio than I have presented here might
lead one to raise two objections against the analysis of Luisenlooutlined above. First,
Luisefio contains a rule of Vowel Syncope which might be used to derive the surface
form of reduplicatedadjectives shown in (51) from the underlyingstructureschematized
in (55), as suggested by Munroand Benson (1973).
(55) C1V1C2V2+ CIVIC2V2 + s
If a rule of Syncope produces (51) from (55), we would expect the cyclic rule of c s-
464 ALEC MARANTZ

to apply to any C1 = c, since the preconsonantal environment for the rule would be
derived.
However, there is no reason to suppose that Syncope is involved in the derivation
of reduplicatedforms. First, Syncope is not a regularphonologicalrule. Davis (1976,
212) lists the three main environmentsfor the applicationof Syncope; all of them are
at least partiallymorphologicaland none of them resemble (55). Second, Syncope does
not apply uniformly in constructions almost identical to (55). There is an intensive-
formingverb reduplicationprocess in Luisenlowhich creates a structurealmost exactly
like (55), but Syncope fails to apply within the reduplicatedportion of this structure,
illustratedin (56) (Munroand Benson (1973, 18)).
(56) CIV1C2V2 + C1V1C2V2 tikf-8iki(tiki 'light, set fire')
As Syncope is not regularin the necessary environment,if we wished to have Syncope
delete the second occurrence of VI in (55), we would have to add an additionalmor-
phological environmentto the rule, stating explicitly that it applies to the first vowel in
the adjective-formingreduplicatingsuffix. Clearly, nothingargues against simply giving
the adjective-formingreduplicatingsuffix the form CCV in place of CVCV; in fact,
simplicity considerationsdemand this move.
The second objection to attributingthe failureof c s- > in reduplicatedforms to the
underivednatureof the c-C combinationsin these forms is that c -s ? sometimes blocks
in derived environmentsas well. Kroeberand Grace (1960)point out, and Davis (1976)
confirms, that there are certainconsonant-initialsuffixes in Luisenlowhich do not trigger
-c s in a stem-final c. These iniclude the agentive forming -kawut and -ku t, which
delete the final vowel of stems to which they attach, and the Tense/Aspect markers
-q(a), -quat, and -qu. (examples from Davis (1976, 202)).
(57) a. mic-ku*t'strangler' mici 'to choke someone'
nec-kawut 'one who pays' neci 'to pay'
b. wac-qa 'are a few (of things)'
wac-quat 'were a few (yesterday)'
wac-qug 'used to be a few'
In addition, when c is the final consonant in a CVC reduplicatingprefix (which adds a
protractedreadingto verbs), it fails to undergoc -s> (Davis (1976, 201)).
(58) nec-nici-q 'pays in dribs and drabs' < neci- 'to pay'
nuc-nuci-q 'keeps going (and) squashingthings' < nuci- 'to squash'
The failureof c -s S to applyin (57)-(58) seems to indicatethatthe rule's environment
is morphological;therefore, its blockage in the reduplicatedforms might show only that
the reduplicatingsuffix is not on the list of c'-- s [Link], in every case where
c s fails to apply in a derived environmentoutside of reduplicatedforms, a boundary
intervenes between the c and the [ - cont] segment which serves as the environmentfor
the rule. As noted by Davis (1976), the rule operates without exception when its envi-
RE REDUPLICATION 465

ronmentis morpheme-internaland derived via Vowel Syncope. It seems that the failure
of c -s* &in (57)-(58) can be explained as a boundaryphenomenonand that, therefore,
'
c is a regularcyclic rule. Thus, the fact that c does not become in the reduplicated
forms in (52) supports the claim that the reduplicatingaffix is in fact a morphemeand
is thus a nonderivedenvironmentin the technical sense.

2.2. Reduplication and Morpholexical Rules

In section 2.1 we saw that a rule which appearedto underapplyinternallyin reduplicated
forms was a cyclic rule, which we do not expect to apply morpheme-internallyin non-
derived [Link] by identifyingthe natureof the rule in question we could
predict its behavior with respect to reduplicatedforms. Similarly,I shall show that the
remainingclass of phonologicalprocesses which either over- or underapplyto redupli-
cated forms share a propertywhich predictstheir over- and [Link],
these processes do not apply to all morphemesmeetingtheir [Link]
each morpheme, one must learn separately whether or not the process will apply (al-
though there may be some generalizationsthat can be expressed regardingthe classes
of morphemeswhich undergoor do not undergothe process). The processes also have
morphologicalratherthan purely phonologicalenvironments.
Aronoff (1976) has called processes meeting the above criteria allomorphyrules.
In Aronoff's theory these rules apply in consort with word formationrules basically to
adjuststems accordingto which affixes have been addedto them. Carrier(1979)suggests
that all allomorphyrules precede all phonologicalrules, with reduplication,considered
as a form of allomorphy, forming the border between allomorphy and phonology.
Carrier's theory predicts the "overapplication" of allomorphy rules to reduplicated
forms-the outputof an allomorphyrule will be copied throughreduplicationif it appears
in the appropriateportion of a stem. However, it should be obvious that the ordering
of allomorphy before reduplicationhas nothing to say about the underapplicationof
allomorphyin [Link] shall returnto Carrier'sproposal below.
Lieber (1980)has given evidence that the processes described above should not be
consideredrules at all in the usual sense. Rather,both the putative"input"and "output"
of these processes are to be listed in the lexicon. The relationbetween the "input" and
"output" allomorphsis describedby a morpholexicalrule, but morpholexicalrules play
no generative role in the phonology.

Morpholexicalrules are predicates which define sets of orderedpairs of lexical items [the
"input" and "output" of the morpholexicalrule], both of which are listed in the permanent
lexicon . . . It is purely arbitrarywhether or not any given lexical item conforms to the
specificationsof a lexical class as defined by its morpholexicalrules. (Lieber (1980, 74-75))

Listing both allomorphsrelated by a morpholexicalrule, Lieber argues, explains their

availabilityto various derivationaland compoundingprocesses.
Processes which by Lieber's independentlyjustified criteriaare identified as mor-
466 ALEC MARANTZ

pholexical rules appear to over- and underapplyto reduplicatedforms simply because

reduplicationis a normal affixing process, which must build on one of the listed allo-
morphs of a morpheme. If reduplicationapplies to the allomorphwhich is the output
of the morpholexicalrule wrongly considered as a generativeprocess, the process will
appearto have overappliedto the reduplicatedform (if the structuralchange is located
within the reduplicatedportion). If reduplicationapplies only to the pseudoinputof the
morpholexicalrule consideredas a phonologicalprocess and the reduplicatedform ends
up in an environmentmeeting the structuraldescriptionof the process, the process will
appearto have underappliedto the reduplicatedform. Thus, assumingthat reduplication
is merely an affixingword formationrule, Lieber's theory demandsthat processes which
independent criteria identify as morpholexical rules either over- or under"apply" in
reduplicatedforms.
2.2.1. Dakota. Reduplicationof verb roots in Dakota either signals the pluralityof an
inanimatesubject or forms the iterative, distributative,or intensive of the verb. Dakota
verb roots fall into two classes: C-finaland V-final stems (see Shaw (1976)).The C-final
stems appear with a stem-extendinga before a word or enclitic boundary. The redu-
plicatingmorphemeis a CCVC skeletal suffix. Accordingto the principlesof association
of a C-V skeletal morphemewith a phonemic melody given in section 1, this suffix has
the effect of copying the last (C)CVCof a C-final stem and the last (C)CV of a V-final
stem. The reduplicatedforms of C-final stems behave like their C-final basic forms in
exhibitinga stem-extendinga before a word or enclitic boundary.
(59) a. V-final stems
ksa ksa-ksa' 'to cut'
haska haska-ska 'be tall'
h h h,
p e p e-p e 'sharp'
b. C-final stems
xap-a xap-xap-a 'to rustle'
nup-a nup-nup-a 'two'
Many phonologicalrules operate normallywithin reduplicatedforms, often making
the reduplicatedroot phonologicallydistinctfrom the reduplicatingsuffix. Wilbur(1973)
would say that these rules are not subject to her IdentityConstraint.I list some of these
rules and examples of their operation as they appearin Shaw (1976, 332ff).
(60) C--' / C C25
ksap-a ksap-ksap-a ksa-ksap-a

cont(
(61) [-[-son J [-glottal1 /
J [cons]
[asp
'cik'-a c'ik'-c'ik'-a ->c'ik-c'ik'-a

25
The Cs in (60) are merely cover symbols for a complex of distinctivefeatures;they do not refer to the
Cs of C-V skeleta. All of the phonologicalrules presentedhere applyto segmentalphonemes,i.e. to elements
of the phonemicmelodylinkedto C-V slots. UnlinkedphonemesandC-V slots left over fromthe reduplication
process should be considereddeleted when the phonologicalrules apply.
RE REDUPLICATION 467

(62) L + -o ] -voice] / [-seg]

puz-a puz-puz-a -> pus-puz-a

1 F- cor1
(63) L +cor
- cont I- ant /
-sn
+ [+cor]

zut-a zut-zuit-a-- zuk-zuit-a

o-zul-a o-zul-zul-a o-zuk-zul-a

(64) [co?t1 ->+ voice]! sy_

+
[- cor son]
lil-a lil-lil-a ( lik-lil-a .!?)2 lig-lil-a
(65) Ci -*/ +Cj
(62) (65)
xux-a xuy-xux-a 2> xux-xux-a 5> xu-xuy-a
F-son 1 +son1
(66) -cont [-nasal] [-seg]
L+ corJ
khat-ya -> khal-ya thec-ya -> thel-ya skic -* skil
khat-a khat-khat-a-- khal-khat-a
?ot-a 2ot-2ot-a -> ?ol-?ol-a
That rules (62) and (66) operate in reduplicatedforms at the boundarybetween the stem
and reduplicatingsuffix completes the argumentthat reduplicationmust be considered
an affixation process. Recall that the cyclic rule of c -s ' in Luiseinodoes not apply
within the "copy" portion of reduplicatedforms althoughthe environmentof the rule
is met within the copy. The interiorof the copy is thus not a "derived environment"
in the technical sense; it is treated as the interiorof a [Link] applicationin the
above examples of rules (62) and (66), rules that apply only at boundaries,clearly shows
the presence of a boundarybetween the copied materialand the stem in reduplicated
forms. When we claim that reduplicationis affixation,we are claimingthat reduplication
involves the additionof a morphemeto a stem, creatingthe derived structure[redupli-
cating morpheme [stem]] or [[stem] reduplicatingmorpheme](as in the Dakota case),
quod est demonstratum.
Although the rules described above apply regularly to reduplicatedforms, two
phonological processes in Dakota appearto treat reduplicatedforms irregularly:a rule
of Velar Palatalizationand a rule of [ + nasal] a to e Ablaut. Among the rules of Dakota,
these two alone operate so as to preserve the identity of copy and stem in reduplicated
forms. Why is it thatjust these two processes and not those in (60)-(66) obey Wilbur's
Identity Constraint?
[Link]. Velar Palatalization. Dakota Velar Palatalization changes k, kh, and k' to c,
c , and C"after a nonlow front vowel.
468 ALEC MARANTZ

(67) khtute'shoot' ni-chute'he shoots at you'

ka'hita 'sweep' i-c'ahite 'broom'
k'a 'dig' mni-c"a-pi'well'
khiyata'side' le-chiyata,'on this side'
However, some initial k, kh, and k' never change to c, Ch, and c.
(68) ku 'to covet' niku 'he covets thee'
ma-kha'ta'I am warm' ni-khata'you are warm'
kho 'to prophesy' t?enikho 'he prophesies your death'

Although one can make certain generalizationsabout which morphemes with initial
velars will undergo Palatalization("neutral verbs", for example, never change initial
velars (Boas and Deloria (1941, 14))), within morpheme classes which can undergo
Palatalization,which morphemeswill have changeableand which unchangeablevelars
is an arbitrarymatter (Boas and Deloria (1941, 14 and elsewhere)).
Moreover, one cannot predicton independentgroundswhich morphemeswith final
i or e will [Link] (1976)claims that all i-finalmorphemesmay trigger
Velar Palatalizationbut that the rule is sensitive to the sort of boundary intervening
between the i and the velar. However, althoughcertain boundariesmay block Palatal-
ization, the evidence indicates that boundaryinformationalone is not sufficient to de-
termine whether or not the rule will apply. Thus, Boas and Deloria (1941, 14) note:

Verbs with initial k, kh, and k' except those with fixed initial ki change in their possessive
forms, but retain k, kh, k? in dative forms.
ki-chfwa 'he pursues his own'
ki-khiwa 'he pursues for him' [fromkhuwa 'pursue']

The two forms given differ only in the palatalizationof the second velar. If we separate
velar from prefix in either form by a boundarywhich would, on Shaw's account, block
Palatalization,main stress would fall on either the initial or the last syllable according
to the rules in Shaw (1976), not on the second syllable. The stress pattern shown is
expected only if the prefix and velar are separated by a boundarywhich does allow
Palatalizationto cross it.
The data show, then, both that the morphemeswhich undergoVelar Palatalization
form an arbitraryclass and that the triggeringenvironmentfor the rule must be mor-
phologically specified. By the criteriagiven above, Velar Palatalizationtherefore qual-
ifies as a morpholexicalrule in Lieber's (1980) sense; we must list in the lexicon both
the palatalized and unpalatalizedallomorphsof each morphemeexhibitingthe alterna-
tion.
Velar Palatalizationappearsto overapply in reduplicatedforms. Althoughonly the
initial velar of the root is in the correct environmentfor Palatalization,the initial velars
of both root and copy show up as palatals when a reduplicatedform of a root with a
changeableinitial velar follows a Palatalizationtrigger.
RE REDUPLICATION 469

(69) a. kaya 'to make'

wic`d-ki-caxcax-?iyeya'he made it for them quickly'
them-DAT-makemake-quickly
b. koza 'to wave'
nape ki-coscoza26'he waved his hand to him'
hand DAT-wavewave
If the palatalized and unpalatalizedallomorphsof a root with a changeable initial
velar are both listed in the lexicon, then the "overapplication"of Palatalizationin
reduplicatedforms is exactly what we expect. Reduplicationsuffixes a CCVC skeleton
to a stem. Now morphemesadded to the right of a morphemewith a changeableinitial
velar do not choose either the palatal-initialor velar-initialallomorphs;they may be
attached to both. For example, the derivationalprefix ka - ca, which adds the instru-
mental reading 'by strikingor sudden force' to verbs, has an initial changeable velar.
The verbs to which it attaches do not choose between the allomorphsof the prefix.
Rather, since a prefix added to the combinationof instrumentalprefix plus verb may
choose either the palatal-initialor velar-initialversion of this derivedverb, the verb root
must be able to attach to both allomorphsof the instrumentalprefix.
(70) ka-khoka'to make a noise by striking'
i-ca-khoka'to make a noise by strikingagainst' (i = locative prefix)
The locative prefix i, a Palatalizationtrigger, chooses the palatal-initialform of the
instrumentalprefix plus verb root combination.
Just as a verb root attachingto the rightof the instrumentalprefix attaches to either
the palatal-or nonpalatal-initialailomorphof this prefix, so the reduplicatingmorpheme,
which attaches to the rightof the verb root, may be affixed to either allomorphof a root
with a changeable initial velar, as shown in (71).
(71) a. koz > koz-koz-a
b. coz -coz-coz-a
The dative prefix ki chooses the palatal-initialallomorphof any stem-simple, redupli-
cated, or prefixed-which has both palatal- and velar-initialalternativeforms related
by the morpholexicalrule of Velar [Link], it chooses the reduplicatedform
in (7lb) to yield (69b),producingthe illusionof the overapplicationof VelarPalatalization
in reduplicatedforms.
Palatalizationcannot apply normallyto reduplicatedforms (i.e. just to the initial
velar) because it does not apply at all. If the reduplicatingmorphemeattaches to the
palatal-initialallomorphof a verb, apparentoverapplicationof Velar Palatalizationre-
sults (see (71b)). If the reduplicatingmorphemeattaches to the velar-initialform, the

26
Note that the palatalizingdative in these examplesis the "second dative". The nonpalatalizingdative
crucial to the argumentthat palatalizingtriggersform an unpredictableclass is the "first dative" (see Boas
and Deloria (1941, 87) for a discussion of the distinctionbetween the datives).
470 ALEC MARANTZ

result is (71a). Since reduplicationin Dakota is a suffixing process and only prefixes
choose palatal-or nonpalatal-initialallomorphs,reduplicationappliesto both allomorphs,
yielding the range of data displayed above.
[Link]. Excursus on MorpholexicalRules. Before continuing,I should make explicit
an elaborationon Lieber's morpholexicalrule theory which I assumed in the discussion
of Dakota Velar Palatalization.A morpholexicalrule expresses the relationbetween two
sets of allomorphs, a markedset and an unmarkedset. (Lieber (1980)calls the marked
set stems and the unmarkedset roots and provides an algorithmfor pickingout the root
in a lexical entry.) The marked set contains the "output" of the morpholexicalrule
wrongly considered as a phonologicalrule; the unmarkedset contains the input. In the
case of Dakota Velar Palatalizationdiscussed above, the velar-initialallomorphs are
unmarked, the palatal-initialallomorphs marked. Each morpholexical alternation is
either righthanded or lefthanded. A "righthanded"morpholexicalalternationis one
whose markedforms are chosen by materialthat subcategorizesto the rightfor the stem
exhibiting the [Link] the paradigmaticcase, the markedform of a righthanded
morpholexical alternation is chosen by a prefix. Since palatal-initialallomorphs are
chosen by prefixes, the Velar Palatalizationmorpholexicalruleof Dakotais righthanded.
A "lefthanded" morpholexicalalternationis one whose markedforms are chosen by
materialthat subcategorizes to the left, e.g. suffixes. Every morphemewhich attaches
to a stem exhibiting a righthandedmorpholexicalalternationand subcategorizesto the
right for this stem must choose one of the allomorphsof the stem. A set of such mor-
phemes is specially marked to choose the markedallomorph;the rest choose the un-
markedallomorphby default. A subset of i-finalprefixes in Dakota choose palatal-initial
allomorphs; the remainderchoose velar-initialforms. Similarly, every morpheme at-
taching to a stem exhibitinga lefthandedmorpholexicalalternationand subcategorizing
to the left for this stem must choose one of the allomorphsof the stem. Again, some of
these morphemesare markedto choose the markedallomorph,while the rest automat-
ically choose the unmarkedallomorph. A morphemeattachingto a stem exhibiting a
morpholexicalalternationbut not subcategorizingto the right for a stem with a right-
handedalternationor to the left for a stem with a lefthandedalternationdoes not choose
either allomorphand thereforeattaches to both. Thus, in the paradigmaticcase, a suffix
will not choose either allomorphof a stem with a righthandedmorpholexicalalternation
and a prefix will not choose either allomorphof a stem with a lefthandedalternation.
The combinationof such a morphemeand stem will exhibit the same alternationas the
stem alone.
In the Dakota case examined above, since reduplicationis a suffixing process in
Dakota and Velar Palatalizationis a righthandedmorpholexicalrule, the reduplicating
morpheme does not choose either the velar-initialor the palatal-initialallomorphof a
stem exhibiting this alternation. Therefore, the reduplicatedform itself exhibits the
palatal- vs. velar-initialalternation, and, when a Velar Palatalizationtriggeringprefix
chooses the palatal-initialallomorph of the reduplicatedstem, the result is apparent
overapplicationof Velar Palatalization.
RE REDUPLICATION 471

[Link]. Ablaut. Dakota includes a rule of {a, a} -* e Ablaut, which changes a mor-
pheme-final a or a (nasalized a) to e before a list of morphemes, mostly enclitics,
including:

(72) ya adverbialending
sni 'not'
sI?e 'as though'
ka, ca 'a kind of, rather'
lakha 'evidently, for'
? phrase-finalenclitic
(73) haska 'to be tall' haske-ya adverb
apha 'to strike' aphe-sni 'he didn't strike it'
aphe-s e 'as thoughhe had hit him'

As Shaw (1976, 134-135) points out, many enclitics, including those in (74), do not
triggerAblaut.

(74) sna habitual

khes 'but always; whenever'
he interrogative
{ha, he} continuative
(75) ya 'to go'
omani-ye-?'he goes about to travel'
omani-ya-he-?'he was going about to travel'

It should be clear from the lists of Ablaut triggersand nontriggersthat the rule's envi-
ronment is morphological(Shaw (1976) takes great pains to prove that there is no way
to predict which morphemeswill triggerAblauton independentgrounds).We must also
simply list the morphemesundergoingthe rule:

Not all stems change terminala, but no generallyvalid rules can be given that would allow
a classificationof stems accordingto definite principles.(Boas and Deloria (1941, 30))

Agreeing with this conclusion, Shaw (1976) divides verb stems according to their be-
havior with respect to Ablaut into six classes whose membershipcannot be predicted
on any independentgrounds.

(76) Class i: V-final stems ending in unchanginga

Class ii: V-final stems ending in unchanginga
Class iii: V-final stems ending in changeablea
Class iv: V-final stems ending in changeable a
Class v: C-final stems whose stem extension a does not change
Class vi: C-final stems whose stem extension a changes
472 ALEC MARANTZ

(77) Class i: niya''to breathe'

niya'-sni'he is not breathing'
Class iii: eya' 'to say'
eye-sni 'he didn't say'
Class v: ca'i-a 'to freeze'
cay-a-s?e 'as if frozen'
Class vi: yuiz-a'to take hold of'
yuz-e-sni 'he didn't catch it'
Since both the morphemes undergoingthe rule and the morphemestriggeringit must
be listed, Dakota Ablaut meets the criteriafor morpholexicalstatus. Consequently,both
the a- or a- and e-final allomorphsof stems with changeablea or a must appear in the
lexicon.
Reduplicatedforms of Class iii and iv roots do not change final a, a althoughtheir
unreduplicatedcounterpartsdo.

(78) haska 'to be tall'

cha-ki iyuihaha'ske-?'all the trees are tall'
tree-DET all tall-enclitic
chaki ha'ska-ska-?'the trees are tall'
tree-DET tall-redupl-enclitic
apha 'to strike'
aphe-sni 'he didn't strike it'
ap a-p a-sni 'he didn't strike it repeatedly'

If Ablaut were a phonological rule, we would have to explain why it fails to apply in
reduplicatedforms. However, the independentlymotivated identificationof Ablaut as
a morpholexicalrule predicts its under"application"to reduplicatedverbs. Consider a
Class iii or Class iv root. Both the a- or a- and e-final allomorphsof the root are listed
in the lexicon and are potentiallyavailablefor affixationof the reduplicatingmorpheme.
However, the data above show that Ablaut is a lefthandedmorpholexicalalternation;
that is, the marked-e-final-allomorph is chosen by suffixed [Link] pre-
fixes attach to both allomorphsof a stem, a morphemeattachingto the rightof the stem
must choose either the a-, a-final or e-final allomorph.
(79) a. ya 'to go'
[o[mani[ye]]]-? omaniye-? 'he goes about to travel'
[about[walkl[go]]]-enclitic
[[o[mani[ya]]]he]-? omaniya-he-?'he was about to travel'
[[about[walk[go]]]continuative]-enclitic
b. apha 'to strike'
aphe-sni 'he didn't strike it'
apha-pi-sni'they did not strike it'
RE REDUPLICATION 473

In (79a), the prefixes o and mani attachto both allomorphsof ya - ye. Thus, the enclitic
2 may choose the e-final version of the derived verb, while the enclitic ha - he chooses
the a-final form. In (79b), on the other hand, the enclitic pi must choose one of the
allomorphs of apha - aphe, since pi is a right-attachingmorphemeaffixing to a stem
with a [Link] it is not on the list of morphemes
takingthe marked(e-final)allomorph,it chooses the unmarked(a-final)form. The enclitic
sni, which chooses the marked(e-final) allomorphof a stem exhibitingthe ablaut alter-
nation, must attach to the form of the combinationof aplac - aphe and pi with verb
root-finala in (79b) because there is no other form.
Since Ablaut is a lefthanded morpholexicalalternationand the reduplicatingaffix
is a right-attaching,left-subcategorizingmorpheme in Dakota, the reduplicatingmor-
pheme must choose one or the other of the allomorphsof a stem with changeablea or
a. The reduplicatingsuffix does not appear on the short list of Ablaut triggers(which
does not contain any derivationalmorphemesresemblingthe reduplicatingmorpheme);
therefore, it must choose the unmarkedor a-, a-final [Link], althougha root
may have two allomorphs, its reduplicatedform will have only one-the a-, a-final
version-simply because reduplicationis a suffixing process. Since the reduplicated
form has only the a-, a-final alternate, an Ablaut triggeringenclitic cannot choose the
e-final allomorphand attaches to a reduplicatedverb as if it were a Class i or Class ii
verb, giving the appearancethat Ablaut blocks in reduplicatedforms.
In contrast to Class iii and iv stems, Class vi verbs do change final a in their
reduplicatedforms.

(80) sic 'be bad'

sic-a-pi 'they are bad'
sic-e-ki 'the bad one'
siksic-e-ki 'the bad ones'

This behaviorof Class vi roots confirmsthe hypothesis, amplysupportedin Shaw (1976),

that the verbal extension a is indeed not partof what we have been callingC-finalstems.
As should be clear from the phonological shape of the reduplicatedforms of Class vi
verbs, the reduplicatingsuffix attaches directly to the root without the stem extension,
suffixing CCVC to a form with the shape . .. (C)VC to yield . .. (C)VC-(C)(C)VC. We
must assume that C-final stems subcategorizeeither for the stem extension exhibiting
the a -e morpholexicalalternation(Class vi stems) or for the stem extension not subject
to this alternation(Class v stems). This selectional informationis inheritedby the re-
duplicated form from the stem to which the reduplicatingsuffix is attached. When an
Ablaut trigger chooses the e-final form of a Class vi verb in either its reduplicatedor
unreduplicatedversion, it is choosing not an allomorphof the stem but ratheran allo-
morph of the stem extension.
2.2.2. Tagalog. Two phonological processes-Nasal Substitutionand Syncope-ap-
pear to overapply to reduplicatedforms in Tagalog. Arguingthat these processes are
474 ALEC MARANTZ

"allomorphy rules" in the sense of Aronoff (1976), Carrier(1979) suggests that all
allomorphyrules precede all phonologicalrules in the grammarand that reduplication,
as a special sort of allomorphy,appearsin a special componentwhich marksthe border
between allomorphyand phonology. Allomorphyprecedes reduplication;reduplication
precedes phonology. The criteriaCarrieruses to determinewhether a process qualifies
as allomorphyarejust those that we have claimedmarka process as morpholexical:the
set of morphemes which undergo the process cannot be identified on independent
groundsand the process must be [Link] have arguedabove that
Carrier'sanalysis is basically correct, as she has identifiedthe class of rules which must
necessarily "precede" reduplicationin some sense. However, because she treats allo-
morphyrules as rules which apply in the same manneras phonologicalrules, she cannot
explain the underapplicationto reduplicatedforms of rules which meet her criteriafor
allomorphy status (e.g. Ablaut in Dakota). If allomorphyprecedes reduplication,we
always expect the illusion of overapplicationof allomorphyrules to reduplicatedforms.
Also, although Tagalog reduplicationresembles allomorphyin importantrespects (see
Carrier(1979) and Lieber (1980) on this point), in most languages reduplicationmore
closely parallelsinflectionaland derivationalaffixation,as the examples in the preceding
pages should make clear. Thus, it is difficult to motivate cross-linguisticallyCarrier's
isolation of reduplicationin a special component following word formation rules and
allomorphy. On the other hand, the analysis given above for the apparentover- and
underapplicationof morpholexicalrules in Dakota accounts for the apparentoverap-
plicationof Nasal Substitutionand Syncope in Tagalogwithoutany special assumptions
about reduplication. Reduplicationis simply affixation and as such precedes all pho-
nological rules.
[Link]. Nasal Substitution. Tagalog exhibits three different sorts of reduplication,
which Carrierlabels RA, RI, and R2. RA prefixes a CV with the V specified [+long];
RI prefixes a CV with the V specified [-long]; and R2 prefixes CVCCV to a stem
with the second V specified [+long] unless the stem begins CVC(C)VCfollowed by a
morpheme boundary, in which case the entire stem is copied (see footnote 5 for a
discussion of this aspect of R2 reduplication;see (7) above for furtherexamples of all
three sorts of reduplication).27
(81) RA: C V + mag-linis'ST (subjecttopic marker)-clean'
mag-li-linis'ST-willclean'
+ long]
Rl: C V+ lakad 'walk'
pag-la-lakad'walking(gerund)'
[-long]
R2: CVCC V + ma-talfnoh'intelligent'
I ma-tali-talfnoh'ratherintelligent'
[+ long]
27 I
will sometimes write as if the various reduplicating prefixes RA, RI, and R2 are each single, uniform
morphemes. Actually, as each prefix has a variety of uses (see Carrier (1979) for numerous examples of the
RE REDUPLICATION 475

A rule of Nasal Substitutionreplaces a stem-initialobstruentwith the homorganicnasal

after a prefix-finaly, deleting the u.
(82) Nasal Substitution
U - [-cont]
1 2
2
[ + nasal]
/mag-bilih/ mamilah 'ST-shop'
/maj-dikit/ manikit 'ST-get thoroughly stuck'
Carrier(1979)demonstratesthat if Nasal Substitutionwere a rule, it would have to occur
before reduplication.
(83) mag-ka?ilanan mai3-ka?ilanan
Nasal Subst ma-Da?ilanan RA Redup mag-kaka?ilanan
RA Redup ma-tjdija2ilanan Nasal Subst *ma-qjkailanan
'ST-will need'
mag-pulah maj-pulah
Nasal Subst ma-mulah R2 Redup mag-pulahpulah
R2 Redup ma-mulahmulah Nasal Subst ma-mulahpulah
Vowel Shortening ma-mulahmulah Vowel Shortening ma-mulahpulah
h Deletion ma-mulamulah h Deletion *ma-mulapulah
'ST-turna
little red'
If reduplication,as a normalword formationprocess, were ordered before Nasal Sub-
stitution, considered a normal phonological rule, Nasal Substitutionwould appear to
overapply to reduplicatedforms, applyingto obstruentsin both the copy (reduplicating
prefix)and the stem althoughonly the obstruentin the copy would meet the environment
of the rule.
(84) ma-pulah'red'
pa-mu-mulan'turningred' < pag-C V -pulah

[-long]
ma-mui-mulah'will turn red' < maij-C V -pulah

[+ long]
ma-mula-mulah'turn a little red' < mag-R2-pulah

reduplicatingprefixes at work), each must be understoodas the morphologicalformof a set of homophonous

[Link] extensive use of a small set of affixes in [Link] additionto the
reduplicatingprefixes, such affixes as ka- and -an are multiplyambiguous.
See footnote 13 on the use of the feature [ +long] on the reduplicatingprefixes in (81).
476 ALEC MARANTZ

Carriershows that Nasal Substitution"is restrictedto apply to a certain morpho-

logically specified set of stems in the environmentof a morphologicallyspecified set of
prefixes" (1979, 64). Some obstruent-initialstems never undergoNasal Substitution.

(85) /mag-basah/ > mambasah 'ST-read'

/maij-dukut/ > mandukut 'ST-pick pockets'
/mag-guloh/ magguloh 'ST-create disorder'
-

Comparethe examples in (85) to those in (82) (the /g/ of the ST prefix assimilates to a
following obstruent if it does not delete throughNasal Substitution).Also, some y-final
prefixes never triggerNasal [Link] such prefix is the accidental/resultprefix
mag-kay-.

(86) /mar-dikit/-> ma-nikit'ST-get thoroughlystuck to'

but /mag-kaij-dikit/-* mag-kan-dikit'get stuck to accidentallyas a result of'

Since which obstruent-initialstems will undergo Nasal Substitutionand which y-final

prefixes will trigger it is generally arbitrary,Nasal Substitutionqualifies as a morpho-
lexical rule, and both nasal-initialand nonnasal-initialallomorphs of each morpheme
"undergoing"the rule must be listed in the lexicon (as must both the y-final and vowel-
final allomorphsof the prefixes which triggerNasal Substitution).We may account for
the apparentoverapplicationof Nasal Substitutionin reduplicatedforms by allowingthe
reduplicatingprefix to attach to either allomorph of stems exhibiting the alternation
capturedby the morpholexicalrule of Nasal Substitution.

(87) fpulah pupulah

mulah mumulahJ
A Nasal Substitutiontriggerwould choose the nasal-initialallomorphof a reduplicated
form, yielding the appearanceof overapplicationof Nasal Substitution.28

28 This account of the interactionof Nasal Substitutionand reduplicationis problematicin one way,
however. Since prefixes choose the nasal-initialallomorphsof stems subjectto Nasal Substitution,the rule
is a righthandedmorpholexicalprocess. Thus, the reduplicatingmorphemes,being prefixes, should choose
one or the other of the allomorphsof a stem subjectto Nasal Substitution,presumablythe nonnasal-initialor
unmarked(root) [Link] seems then that morpholexicaltheory predictsthe underapplicationof Nasal
Substitutionto reduplicatedforms ratherthan its overapplication.A reduplicatedform should exhibit only a
nonnasal-initialallomorph.
We could circumventthis problemby simply allowingreduplicatingprefixes not to choose between the
allomorphsof a stem subjectto Nasal [Link], this move would involve weakeningthe theory
of morpholexicalrules to permitprefixes not to choose allomorphsof righthandedmorpholexicalprocesses,
a weakeningthat is not independentlymotivated. In an earlierversion of this article, I arguedthat Tagalog
reduplicationdisplays a peculiarpropertywhich 1 ads us to predictthat reduplicatingprefixeswill not choose
allomorphsof righthandedmorpholexicalrules in"thislanguage:Tagalogreduplicatingprefixes do not sub-
categorizefor the stems to which they attach. I referthe readerto Lieber(1980)for some discussionof Tagalog
reduplicationalong these lines.
RE REDUPLICATION 477

[Link]. Syncope. In additionto Nasal Substitution,a rule of Syncope in Tagalogalso

appearsto interactunusuallywith [Link] deletes the last vowel in stems
before the verbal suffixes -an and -in. As illustratedin (88), in some verbs Syncope is
obligatory, in some it is optional, and in some it is [Link] independentcriteria
determine to which morphemes Syncope will apply: "among those roots which have
light penultimate syllables, it is not possible to distinguishin phonological terms the
ones which undergo syncope from the ones which do not" (Carrier(1979, 87)).

(88) t-um-iijin *tiUin-an

'ST-watch' tihn-an
'watch-OT(object topic marker)'
d-um-umih dumih-an
'ST-makedirty' dumh-an
'make dirty-OT'
mag-wakas wakas-an
'ST-end' *waks-an
'end-OT'

Only the verbal suffixes -an and -in trigger Syncope. The nominalizingsuffix -an, for
example, which is homophonous with the verbal suffix -an, does not trigger Syncope
even in verbs which undergo the rule when the verbal suffix -an attaches.

(89) mag-bigay'ST-give'
bigy-an 'give-OT' bigay-an 'a giving to one another'
s-um-akay'ST-board'
saky-an 'board-IOT' sakay-an 'a boardingby many'

Since both its environmentand its target must be morphologicallyspecified, Syncope

qualifies as a morpholexicalrule.
If both processes are considered rules, R2 reduplicationmust follow Syncope.

(90) tijin-an tigin-an

Syncope tign-an R2 Redup tiOin-tijin-an
R2 Redup tihnan-tignan Syncope tiUin-tijn-an
Vowel Shortening tijnan-tirnan Vowel Shortening *tijin-tiIn-an
/tiiin-an/ -* tiinan 'look at-OT'
R2-tijin-an -- tijnantijnan 'look at a bit-OT' *tijintilnan

The problem here is not actually one of overapplicationof a phonological process in

reduplicated forms. If we reduplicated tiqin-an to derive tiyin-an-tijin-an, Syncope could
apply normally to both yin-an sequences, yielding the correct tiynantiynan. However,
since R2 reduplicationnever copies more than two syllables, we cannot derive the input
478 ALEC MARANTZ

to Syncope, ti4in-an-tiyin-an, necessary for this analysis. Noting that the "overappli-
cation" of Syncope in Wilbur's (1973) sense cannot derive forms like tiynantiynan,
Carrier(1979) rightly points out that the reduplicationof syncopated forms in Tagalog
is an example of an unusual interaction of a phonological process and reduplication
about which Wilbur'sIdentity Constrainthas nothingto say. We must explain here why
a phonological process, Syncope, appears to apply before a word formationprocess,
reduplication.
As was the case with Nasal Substitution,the interactionof Syncope with redupli-
cation follows from the identificationof Syncope as a morpholexicalrule and of redu-
plicationas a normalprefixationprocess. Since syncopatedformsare chosen by suffixes,
Syncope is a lefthandedmorpholexicalrule. Therefore,the R2 reduplicatingprefix will
attach to both allomorphs of a stem exhibiting the Syncope alternation, yielding two
alloforms of the combinationR2 plus root.
(91) a. f R2 - CVCVC
b. tR2 - CVCC J

Syncope triggerschoose the syncopated form (91b).

This analysis of the interactionof reduplicationand Syncope raises an interesting
problem for the theory of reduplicationpresented in section 1. Let us suppose, with
Lieber (1980), that morphemesare inserted into unlabeledbinary branchingtree struc-
tures. The subcategorizationframes of morphemes ensure the proper deployment of
morphemes within these structures, while a system of feature percolation labels the
nodes and combines features of the insertedmorphemesto producefeature matricesfor
the entire word and subunits of the word. Since R2 prefixes to V roots, not V' forms
(see Carrier (1979)), a word like tiUnantiynan must have the tree structure shown in (92).

(92) VI

V -an

R2 V

tiijn-

Recall that R2 copies the first CV(C)CV of a stem unless the first CV(C)CVCis
followed by a morphemeboundary,in which case it copies the entire stem (see footnote
5). We might say that R2 has two allomorphs,the C-V skeleton CVCCV and a mor-
phemic skeleton p.(see section 1.4). The morphemicskeletalallomorphattachesto stems
beginningCV(C)CVC[- segment],while the C-V skeletalallomorphattacheselsewhere.
Forms like timnantiUnandisplay the shape we would expect if R2 attached to the com-
RE REDUPLICATION 479

bination of a V stem and verbal suffix, as in (93), ratherthan to the V stem alone, as
in (92); the stem tignan but not tizn begins CV(C)CVC[ - segment].

(93) 1

R2 2

V -an

tiun-
Since V' in (93) begins CVCCVC[- segment], the morphemicskeletal allomorphof R2
appears, yielding the correct result. The paradoxhere is this: the selectionalrestrictions
associated with the R2 reduplicatingprefix demandthe structurein (92), while only the
structurein (93) produces the correct reduplicatedform.
Pesetsky (1979)provides examples from Russian and English in which certain con-
siderations(e.g. semantic interpretation)requirea prefix to attachto a root and a suffix
to attach to the derived stem, as in (92), while other considerations(e.g. the phonological
cycle) demand that the prefix attach to the combinationof the root and the suffix, as
in (93). For example, the Englishcomparativesuffix, -er, attachesto many monosyllabic
adjectives, to some disyllabic adjectives with light final syllables, but to no adjectives
with three or more syllables.
(94) red redder dark darker
little littler happy happier
putrid *putrider turquoise *turquoiser
palatable *palatabler excellent *excellenter
The selectional restrictions on -er thus demand the structure in (95a) for the word
unhappier;if the structurewere (95b), -er would have to attachto a trisyllabicadjective.
However, the semantic interpretationof unhappierrequiresthe bracketingin (95b); a
man who is unhappierthan Elmer is more unhappythan he is, not not more happy.
(95) a. A b. A
A ~/A\
un- A A er

A
A -er un- A

happy happy
480 ALEC MARANTZ

It is temptingto speculate that whatever the solution to the dual bracketingproblemin

Russian and English turns out to be, it will carry over to the Tagalog example as well.
At the moment, how to derive the R2 reduplicatedform of syncopated stems remains
a problemfor future research.

3. Summaryand Conclusion
In section 1, I claimed that reduplicationis best analyzed as the affixationof a skeletal
morphemeto a stem. In section 2, I tried to show that the analysis of reduplicationas
affixationexplains the otherwise puzzling interactionof reduplicationwith certain pho-
nologicalprocesses. The fact that reduplicationprocesses can generallybe characterized
by a fixed consonant-vowel shape, a fact capturedin the identificationof reduplicating
morphemes as C-V skeleta, provides considerable support for McCarthy's autoseg-
mental representationof words on different tiers includinga phonemic melody and a
C-V skeleton. Given that reduplicationis simply affixation, the interactionof redupli-
cation and phonologicalprocesses, which seemed so mysteriousto earlierinvestigators,
is predictedby Halle's (1979)interpretationof the phonologicalcycle and Lieber's (1980)
morpholexicaltheory. This article thu-sprovides strikingconfirmationof both Halle's
and Lieber's proposals.
At first glance, reduplicationseems a rare bird among morphologicalprocesses, an
exotic curiosity. Struckby the apparentnovelty of reduplication,previous investigators
have treated reduplicationas a special case-and were confrontedwith problems as a
result. Beginning with the assumption that reduplicationis minimally different from
processes we are accustomed to, I have not only constructed an elegant account of
reduplicationand its properties but have also been able to illuminateimportantissues
in phonology and [Link] we recognize that the same theory must cover the
duckbill platypus as well as the house cat, we gain a deeper understandingof both the
platypus and the cat.

References
Anderson, S. R. (1975) "On the Interactionof PhonologicalRules of Various Types," Journal
of Linguistics 11, 39-62.
Andrade,M. J. (1933) Quileute, ColumbiaUniversity Press, New York.
Aronoff, M. (1976) WordFormation in GenerativeGrammar,LinguisticInquiryMonograph1,
MIT Press, Cambridge,Massachusetts.
Berlin, B. (1963) "Some Semantic Features of Reduplicationin Tzeltal," IJAL 29, 211-218.
Boas, F. and E. Deloria (1941)Dakota Grammar,Memoirsof the National Academy of Science
23, Part 2, United States GovernmentPrintingOffice, Washington,D.C.
Bright, W. (1957) The KarokLanguage, University of CaliforniaPress, Berkeley, California.
Bright, W. (1968)A LuisefioDictionary, UCPL, no. 51, Universityof CaliforniaPress, Berkeley,
California.
Carrier,J. L. (1979)TheInteractionof MorphologicalandPhonologicalRules in Tagalog,Doctoral
dissertation,MIT, Cambridge,Massachusetts.
Davis, J. F. (1976) "Some Notes on Luisefio Phonology," IJAL 42, 192-216.
RE REDUPLICATION 481

Delano, 1. 0. (1965)A Modern YorubaGrammar,Thomas Nelson and Sons, London.

Dixon, R. M. W. (1972) The Dyirbal Language of North Queensland, Cambridge University Press,
Cambridge.
Dixon, R. M. W. (1977)A Grammarof Yidiny,CambridgeStudiesin LinguisticsNo. 19, Cambridge
University Press, Cambridge.
Goldsmith,J. (1976) "An Overview of AutosegmentalPhonology," LinguisticAnalysis 2, 23-68.
Halle, M. (1979)"Formalvs. FunctionalConsiderationsin Phonology,"distributedby the Indiana
University Linguistics Club, Bloomington,[Link] in B. Brogyanyi,ed. (1979)Am-
sterdam Studies in the Theory and History of Linguistic Science IV, Current Issues in
LinguisticTheory Vol. 11, John BenjaminsB. V., Amsterdam.
Halle, M. and J.-R. Vergnaud(1980a) "Three DimensionalPhonology," Journal of Linguistic
Research I, 83-105.
Halle, M. and J.-R. Vergnaud(1980b) "Tiered Phonology," unpublishedms., MIT, Cambridge,
Massachusetts.
Healey, P. M. (1960) An Agta Grammar,Summer Institute of Linguistics, PhilippineBranch,
Manila.
Kiparsky, P. (1973) "PhonologicalRepresentations,"in 0. Fujimura,ed., ThreeDimensions of
Linguistic Theory, TEC, Tokyo.
Kiparsky,P. (1979) "MetricalStructureAssignmentis Cyclic," LinguisticInquiry10, 421-441.
Krause, S. R. (1980) Topics in Chukchee Phonology and Morphology, Doctoral dissertation,
University of Illinois, Urbana-Champaign,
Illinois.
Kroeber, A. L. and G. W. Grace (1960) The SparkmanGrammarof Luiseho, University of
CaliforniaPress, Berkeley, California.
Lieber, R. (1980) On the Organizationof the Lexicon, Doctoral dissertation,MIT, Cambridge,
[Link] availablefrom the IndianaUniversityLinguisticsClub, Bloomington,
Indiana.
McCarthy, J. (1979) Formal Problems in Semitic Phonology and Morphology, Doctoral disser-
tation, MIT, Cambridge,Massachusetts.
McCarthy,J. (1981) "A Prosodic Theory of NonconcatenativeMorphology,"LinguisticInquiry
12, 373-418.
Marsack,C. C. (1962) Teach YourselfSamoan, The English University Press, London.
Mascar6, J. (1976) Catalan Phonology and the Phonological Cycle, Doctoral dissertation, MIT,
Cambridge,[Link] the IndianaUniversityLinguisticsClub, Bloom-
ington, Indiana.
Moravcsik, E. A. (1978) "ReduplicativeConstructions,"in J. H. Greenberg,ed., Universalsof
Human Language, Vol. 3: Word Structure, Stanford University Press, Stanford, California.
Munro, P. and P. J. Benson (1973) "Reduplicationand Rule Orderingin Luisenlo,"IJAL 39,
15-21.
Nash, D. (1979) "YidinyStress: A MetricalAccount," in E. Battistella, ed., CUNY Forum 7/8
(Proceedingsof NELS IX), City University of New York, New York.
Nash, D. (1980) Topics in WarlpiriGrammar,Doctoral dissertation,MIT, Cambridge,Massa-
chusetts.
Pesetsky, D. (1979) "Russian Morphologyand Lexical Theory," unpublishedms., MIT, Cam-
bridge, Massachusetts.
Pranka,P. (1981) "Reduplicationin SouthernPaiute," unpublishedms., MIT, Cambridge,Mas-
sachusetts.
Reichard,G. A. (1959) "A Comparisonof Five Salish Languages,"IJAL 25, 239-253.
Sapir, E. (1930) Southern Paiute, A Shoshonean Language, Proceedings of the American Academy
of Arts and Sciences vol. 65, no. 1.
482 ALEC MARANTZ

Schachter, P. and V. Fromkin(1968)A Phonology of Akan:Akupem,Asante & Fante, Working

Papers in Phonetics No. 9, University of California,Los Angeles, California.
Shaw, P. A. (1976) Dakota Phonology and Morphology, Doctoral dissertation, University of
Toronto. Publishedin 1980by GarlandPublishing,New York.
Stevens, A. M. (1968)MaduresePhonology and Morphology,AmericanOrientalSeries vol. 52,
AmericanOrientalSociety, New Haven, Connecticut.
Wilbur,R. B. (1973)ThePhonology of Reduplication,Doctoraldissertation,Universityof Illinois.
Distributedby the IndianaUniversity LinguisticsClub, Bloomington,Indiana.
Yip, M. (1981) "Reduplicationand C-V Skeleta in Chinese Secret Languages,"paperpresented
at NELS XII, MIT, Cambridge,Massachusetts,November 1981.

Harvard UniversitySociety of Fellows

78 MountAuburnStreet
Cambridge,Massachusetts 02138