1

1 [Post-print version – For personal use only]

3 Title: Building mental models from multiple texts: How readers construct coherence from inconsistent

4 sources

5 Running title: Building mental models from multiple texts

6 Authors’ names and affiliations:

7 Gaston Saux a

8 M. Anne Britt b

9 Nicolas Vibert c

10 Jean-François Rouet c

11

a
12 Pontifical Catholic University of Argentina -- National Scientific and Technical Research Council
13 (CONICET). Av. Alicia Moreau de Justo 1500, Edif. San José, 2do piso (1107) Buenos Aires,
14 Argentina
b
15 Northern Illinois University, office 363, 100 Normal Rd. DeKalb,IL 60115, USA
16 c
Centre de Recherches sur la Cognition et l’Apprentissage, CNRS, Université de Poitiers, Université
17 de Tours, MSHS - Bâtiment A5, 5, rue Théodore Lefebvre, TSA 21103, 86073 Poitiers cedex
18 9, France
19

20
2
21 Abstract

22 Text comprehension theories propose that readers try to achieve a coherent representation of the

23 situation depicted in a text. When reading multiple texts on the same topic, however, achieving a

24 coherent representation of a situation poses additional challenges. Different sources may offer

25 inconsistent descriptions or interpretations of the situation, or even contradict one another. Thus,

26 additional assumptions are needed to explain how readers build a coherent representation of a situation

27 when reading multiple individual texts on the same subject. This paper reviews research contributions

28 from the psychology of text comprehension on how readers integrate inconsistent information from

29 multiple sources. We concentrate on two key processes: the construction of an interconnected

30 representation of sources and text contents, and the evaluation of the acquired information. We begin

31 by briefly discussing the standard approach of single-text comprehension theories to representational

32 coherence. Then, we examine the Documents Model Framework and other, associated proposals that

33 claim that readers can achieve coherence of divergent text contents by constructing a representation of

34 the texts which integrates information about the sources of the contents. We also consider the role of

35 source evaluation as a reader strategy for constructing a single, coherent solution. Finally, we discuss

36 theoretical and practical implications and we make suggestions for further readings.

37

38 Keywords: Multiple text, Coherence, Information Source, Mental Models, Inconsistencies

39

40 1. Introduction

41 Reading is a pervasive, but also complex and highly versatile activity. Readers read diverse types

42 of texts, from simple stories to online news to abstract scientific explanations. They read for all sorts of

43 purposes and with all levels of engagement, from a casual glance to systematic and deep processing of

44 the text information. A common assumption of text comprehension theories is that regardless of genre,

45 context and goals, readers try to attain a coherent mental representation of the situation presented in the

46 text (McNamara & Magliano, 2009). Coherence, as it applies to text comprehension, broadly refers to
3
47 the integration, or connectedness of one's cognitive representation of the text content (McNamara &

48 Magliano, 2009; van den Broek & Gustafson, 1999). Connections, in turn, refer to the continuity and

49 logical organization of ideas along situational dimensions such as time, space, or causation (Zwaan &

50 Radvansky, 1998). Readers may achieve a coherent representation by using preexisting schemata (e.g.,

51 a story schema), or by organizing individual ideas en route into an ad hoc frame of reference. An author’s

52 use of linguistic markings for referential continuity and situational dimensions can facilitate readers’

53 construction of a coherent text representation. However, coherence also requires the information to be

54 mostly consistent with the readers’ knowledge and beliefs about the world (Black et al., 1986) and

55 inferable from that knowledge, as connections between individual propositions are not always explicit

56 (e.g. "Jack bumped into the desk. Coffee splashed onto the laptop."). Thus, coherence is both an inherent

57 feature of well-written texts and the outcome of readers’ active comprehension based on their relevant

58 background knowledge.

59 Defining comprehension as the construction of a coherent representation has served as a useful

60 organizing framework for a great deal of research on discourse processes (Gernsbacher, 1990; Goldman

61 & Varma, 1995; Graesser et al., 1994; Kintsch, 1998; McKoon & Ratcliff, 1992; Myers & O’Brien,

62 1998; O’Brien & Cook, 2016; van Dijk & Kintsch, 1983; van den Broek et al., 1996; see McNamara &

63 Magliano, 2009, for a review). However, over the past few decades new and emerging reading contexts

64 have brought a number of different and intriguing challenges for theories of reading comprehension.

65 One such challenge has been the democratization of publishing brought about by the internet and World

66 Wide Web.

67 Reading in the digital era comes with a paradox. The Internet has supported the dissemination of

68 an incommensurable wealth of information about any potential topic of interest. Access to written texts

69 was never easier or faster. In fact, citizens in post-industrial societies spend a large and possibly

70 increasing amount of time interacting with written information (Britt et al., 2018). Yet - or perhaps

71 because of this, forming a coherent representation of situations seems to pose greater challenges to

72 readers of online texts. In fact, in 2020 the World Health Organization coined the term 'infodemic" to

73 describe the unconstrained spread of information that followed the Covid-19 pandemic (WHO, 2020).
4
74 Whereas the over-abundance of information is a problem by itself, there is also the added difficulty of

75 finding trustworthy sources and reliable advice. Now more than ever it falls on readers to navigate

76 through multiple sources, to gauge, to prune, to compare and contrast information in order to build an

77 integrated representation that makes sense out of what they read.

78 Such a dramatic evolution in the production and dissemination of texts warrants a critical

79 examination of the theories that account for the processes at work when reading multiple texts. We refer

80 to multiple text comprehension as making sense of a situation presented in two or more texts, where

81 each text originates from a different source (e.g., author, journal, media outlet; Perfetti et al., 1999;

82 Rouet et al., 2019). The basic cognitive processes underlying multiple text comprehension are arguably

83 similar to those involved in single text comprehension. For example, when trying to connect the

84 information found in one text with other information found in a previously read text, readers need to

85 activate previously acquired information and to integrate it with their current reading focus (e.g., Beker

86 et al., 2016). Likewise, comprehending multiple texts requires processes such as shifting attention in-

87 and-out of texts to resolve a reference, skimming a text by disregarding already known information to

88 locate new information, and reasoning about the validity of what’s being comprehended, among other

89 possible tasks.

90 However, multiple texts dealing with the same situation do not always neatly combine into a

91 coherent picture of the situation. This creates new challenges for readers and a need to explain how

92 readers can make sense of the texts as a set. First, differences in time, space, and modality of presentation

93 can reduce the chances that the reader will connect them (Britt et al., 1999). Second, unlike a single text,

94 multiple texts present differences in structure, genre and language style that can hinder the identification

95 of common points (e.g., Rouet & Britt, 2014). Third, the connection across several texts is hard to signal

96 with figures, graphs, and videos, or linguistic cohesion markers, such as connectors, anaphors, and other

97 resources, typically employed to prop up intra-textual coherence (Salmerón et al., 2018; Stadtler et al.,

98 2018). Finally, different texts may refer to the same content from dissimilar perspectives, yielding

99 differences, discrepancies, or even flagrant contradictions across accounts (e.g., Bråten & Braasch,

100 2018; Perfetti et al., 1999). For instance, a reader faced with two texts claiming respectively that climate
5
101 change is caused by human activity and unrelated to human activity cannot include these idea units into

102 a common situation model. How then do readers build a unique and stable representation that includes

103 two (or more) incompatible versions of the same account, the same time-space sequence, the same causal

104 structure, etc.? A core undertaking of research into the comprehension of multiple texts has been to

105 uncover the additional processes that are required for readers to construct coherent representations of

106 multiple text contents, including discrepant or conflicting ones.

107 The following sections review research contributions over the last 30 years from the psychology

108 of text comprehension on the problem of forming a coherent representation out of inconsistent or

109 conflicting accounts. Among the multiple processes involved in dealing with complex information

110 (Rouet, 2006), we focus on two key processes: the construction of an integrated representation of

111 contents and sources and the evaluation of the acquired information. Although present in single text

112 comprehension, these processes differ markedly when comprehending multiple texts.

113 This article is written with an interested but non-specialist audience in mind. Therefore, we have

114 intentionally adopted a simplified, introductory approach that focuses on empirical contributions

115 associated with theoretical models. For a more technical review, we suggest consulting the Further

116 Reading section, at the end of the paper.

117 2. Text comprehension and the problem of inconsistencies

118 What happens in our minds as we read and comprehend the meaning of a text? Over the past four

119 decades, cognitive psychologists have forged a view of text comprehension as the reader's gradual

120 construction of a cognitive representation that captures the meaning of the text. The representation may

121 be modeled as a network of concepts and semantic propositions, although it may also include other

122 constituents such as mental images or the reader's emotional responses (Kintsch, 1998).

123 Texts are expected to be internally coherent, and a core goal for the reader is to achieve a

124 representation that corresponds to the organization of ideas, facts or events as depicted in the text

125 (Goldman et al., 2007). The Construction-Integration model (Kintsch, 1988, 1998) proposed that

126 comprehension begins with a construction phase, (with both coherent and incoherent meanings

127 potentially included in the representation) before transitioning to an integration phase, in which the
6
128 representational network settles into a coherent state. This is achieved by prioritizing the information

129 that best fits the discourse context and the readers' prior knowledge, to the detriment of less central or

130 less consistent information. For instance, upon reading the word “spade” in a text, the reader may

131 initially represent both the symbol on a playing card and the gardening tool. If the text is a story about

132 a group of friends playing poker, however, the skilled reader's representation will quickly converge

133 toward the former meaning, discarding the second, less relevant one (Gernsbacher, 1990). Thus,

134 semantic nodes with a weak relation to the rest of the network, and those that may contradict or

135 destabilize other parts of the network are excluded from the final representation.

136 A core finding supporting the coherence-based approach to comprehension comes from the study

137 of readers' reaction to statements that contradict information previously stated in the same text. For

138 example, participants slow down when reading a sentence about a character ordering a cheeseburger if

139 the character was previously introduced as being a strict vegetarian (Albrecht & O’Brien, 1993). This

140 provides evidence for a phenomenon called comprehension monitoring (Baker, 1979), and attributed in

141 part to the "resonance" of previously read information in readers' memory (Myers & O’Brien, 1998).

142 Other studies have examined the influence of new, inconsistent information on readers updating

143 of their initial understanding of the situation (for a review, see van Oostendorp, 2014). For example,

144 readers were asked to read a report stating that a fire in a warehouse was caused by the negligent storage

145 of flammable materials. Then, they read a subsequent report that retracts this information by stating that

146 there were no such materials when the fire started (Wilkes & Leatherbarrow, 1988). When required to

147 respond to inferential questions, such as what caused the fire in the warehouse, readers still refer to the

148 flammable materials, even though they recall that this information had been discarded as the reports

149 unfolded. These findings are consistent with readers’ reactivating previously read information, but they

150 suggest that readers will tend to “stick” to their initial understanding of the situation and discard new

151 information that does not fit their existing representation (e.g., van Oostendorp & Bonebakker, 1999),

152 particularly when the correction would require a deep reorganization of the representation (van

153 Oostendorp, 1996).

7
154 To sum up, the manipulation of within-text inconsistencies has provided significant knowledge

155 of how memory works as comprehension unfolds, but not necessarily of how readers make sense of

156 inconsistent information. The inconsistencies themselves were not part of the theory, but rather a

157 technical way to highlight the importance of coherence in normal reading. In today's world, though,

158 inconsistencies and even outright contradictions are more often part and parcel of the reader's daily

159 experience. The way people used to acquire written information 30 years ago is very different from how

160 they acquire it today, in part because we can now easily access multiple, divergent texts (think again

161 about the "infodemic" associated to the 2020 Coronavirus pandemic). Following a traditional, cognitive

162 approach to text comprehension (e.g., Gernsbacher, 1990; Kintsch, 1998), constructing a coherent

163 representation would require the rejection of some of the perspectives because they would prevent one

164 from achieving a coherent representation. Indeed, ignoring or bypassing the inconsistencies are possible

165 strategies when reading multiple texts (e.g., Richter & Maier, 2017; Stadtler & Bromme, 2014).

166 However, there is also extensive evidence that readers can form a coherent mental representation out of

167 a set of multiple discrepant texts without excluding the contradicting information (e.g., Rouet et al.,

168 2016).

169 3. Comprehending multiple texts: representing contents and sources of information

170 Research interest in the cognitive processes involved in multiple text comprehension emerged in

171 the early 1990’s, prompted both by some early works, such as Wineburg's (1991) seminal study on the

172 comprehension of historical documents, and the growing interest in the then-emerging digital reading

173 paradigm (Rouet et al., 1996). History was a fertile area for multiple text research because learning

174 within this domain typically requires the consultation of several documents. Wineburg (1991) found

175 that history experts, unlike high school students, used various heuristics when studying a set of

176 documents about an episode in the American Revolution: they paid attention to the source of each

177 document (sourcing), they compared and contrasted the information across documents (corroboration)

178 and they reflected on the information based on their knowledge of the period, the context and the actual

179 situation (contextualisation). In a series of experiments, Rouet et al. (1996, 1997) asked college students

180 to read several documents on the history of the Panama Canal. The documents presented different
8
181 perspectives and evidence, although they also maintained a basic core of common information. After

182 collecting several measures (e.g., summary production, comprehension questions, reasoning probes,

183 etc.), they concluded that readers of multiple texts represent more than the situation described in the

184 text(s). Their representation also includes information about who says what and about how multiple

185 accounts connect to each other. This approach came to be known as the Documents model framework

186 (Britt et al., 1999; Perfetti, 1999).

187 3.1. Overview of the documents model framework

188 The Documents model framework (thereafter DMF; Britt et al., 1999; Britt & Rouet, 2012;

189 Perfetti et al., 1999) argued that, in addition to comprehending the situation, readers must comprehend

190 how the documents represent (and sometimes distort) the situation. A documents model is a cognitive

191 representation that integrates the different versions of the situation referred to in the texts along with

192 information on the sources that produced those texts. A documents model consists of two components:

193 the integrated mental model and the intertext model (Figure 1).

194

195 [INSERT FIGURE 1 HERE]

196

197 The integrated mental model (to the right of Fig. 1) is a representation of the different versions

198 of the common subject matter depicted in the texts1. Most often, people read multiple documents just to

199 increase their knowledge of a situation. Additional documents may provide new details on aspects of

200 the situation or phenomenon, or they may elaborate on circumstances, antecedents, causes, or

201 consequences of the phenomenon. In those cases, documents complement each other and the

202 information is presumably merged into a single mental model. However, multiple documents sometimes

203 provide divergent and even conflicting information (Britt et al., 2013). As an example, let us reframe

204 the fire story presented in the previous section (Wilkes & Leatherbarrow, 1988). Let’s suppose now that

205 the fire did not take place in a warehouse, but that it hit one of the largest extensions of primitive forest

1
Originally called “Situations Model” (Perfetti et al., 1999, p. 102), the name was changed to “Integrated Mental Model”
in the last decade (Britt & Rouet, 2012, p. 285) to extend the approach to texts depicting abstract phenomena.
9
206 on the planet. A person browsing the news to find out more about the fire is likely to find discrepant

207 explanations. A newspaper article could attribute the cause of the fire to an extraordinary increase of

208 human exploitation of the natural resources, while a government statement could instead attribute it to

209 a normal, natural process (for a similar case, see United Nations News, 2019). If, after reading the texts,

210 someone explained that “the causes of the fire are not clear” s/he would be assumed to have connected

211 the situation models derived from each text. It could be argued, though, that this connection emerges

212 from simply accumulating information as novel texts are read. The problem with this last option is that

213 it is impossible to represent a state of affairs where the cause of the fire is extraordinary and ordinary at

214 the same time. Instead, the DMF proposes that two situation models have been generated and related

215 together into an integrated mental model (Perfetti et al., 1999).

216 The second component of the documents model is the intertext model (in the center of Fig. 1).

217 This is a representation of the sources of information. The DMF conceives of sources as a set of

218 parameters that identify the origin of a text’s content, including information about the author (name,

219 characteristics and motives), the context of content production (editor, volume, date and cultural context

220 of the publication, etc.), the type of document (blog, textbook, scientific article, etc.) and possibly the

221 communicative intentions of the message (inform, persuade, disinform, etc.). The intertext model is

222 structured into document nodes and intertext links (represented respectively by circles and lines in Fig.

223 1). The nodes represent discrete knowledge about the source, whereas the links connect the source nodes

224 to text content (or Source-Content links; e.g., "A asserts X", "B asserts Y") and the source nodes between

225 them (or Source-Source links; e.g., "A agrees with B", "A supports B", A contradicts B ", etc.). The

226 Source-Content links form indexes that separate the specific contributions of each text, whereas the

227 Source-Source links relate the texts with respect to who produced the information they contain.

228 Returning to the jungle fire example, the reader who connects each version of the cause (i.e., human

229 abnormal exploitation vs. normal events) to its source (i.e., the newspaper article and the government

230 statement, respectively) and relates both sources through a rhetorical link (e.g., the newspaper article

231 contradicts the government statement), would have built an intertext model.
10
232 One can think of the degree of integration in a documents model representation as occupying a

233 position along a continuum which ranges from completely separate representations of each text to a

234 mushed fusion of all information (Britt et al, 1999). The construction of a documents model requires

235 integration but also enough representational separation to accommodate any discrepancies. This is

236 achieved by creating indexes that organize knowledge based on its origin or source. Furthermore, the

237 intertext model is constructed from information that does not necessarily belong to the contents of the

238 texts themselves. This implies an understanding of documents as artifacts created by people for social

239 purposes and not only as situation models (Britt et al., 2013; Wineburg, 1991).

240 3.2. How documents models establish coherence when contents do not cohere

241 Inconsistencies across texts prevent readers from just merging the texts’ contents into a single

242 mental model, because this would result in an incoherent representation per the traditional definition of

243 coherence. The DMF offers a solution to the coherence problem by proposing that a reader will construct

244 more than one mental model, and that he or she will connect the mental models by making an

245 instrumental use of the information sources. Thus, the framework proposes that coherence is achieved

246 when multiple pieces of discourse can be integrated into a single model of who says what. This amounts

247 to broadening the range of dimensions whereby discourse elements can be integrated, to include so-

248 called rhetorical relationships, such as “supports”, “corroborates”, “opposes”, “contradicts” and so forth

249 (Perfetti et al., 1999).

250 Empirical data accumulated over the past decades have largely supported this view. In an

251 influential study, Braasch et al. (2012) asked their participants to read news stories in one of two possible

252 versions. The “discrepant” version included two contradictory statements about the same event, each

253 attributed to a different source (e.g., a character stating that the audience loved the show vs. another

254 asserting that the audience hated the show). The "consistent" version was identical, except that the

255 statements were compatible. Compared to the consistent version, participants reading the discrepant

256 version looked more often at the sources during reading. They also included more references to the

257 sources when writing summaries of the stories and better remembered who said what in a memory task.

258 To explain these results, the researchers postulated a discrepancy-induced source comprehension
11
259 hypothesis (or D-ISC), framed within the more general DMF proposal. According to D-ISC,

260 discrepancies induce the reader to create source-content links as a strategy to maximize coherence (note,

261 however, that readers could also resort to other strategies to deal with the discrepancy, as we will discuss

262 later).

263 The D-ISC hypothesis has received additional support in several studies conducted in the last

264 decade (see Braasch & Scharrer, 2020 for a recent review). These follow-up studies have extended the

265 original effect to different age ranges and educational levels (De Pereyra et al., 2014a; Salmerón et al,

266 2016); different components of the intertext model (e.g., Saux et al., 2017), sources embedded in the

267 same text or across multiple texts (De Pereyra et al, 2014b), different sourcing aspects such as attention,

268 memory, evaluation, and use of the sources (e.g., Kammerer et al., 2016; Strømsø et al., 2013; Barzilai

269 & Eshet-Alkalai, 2015; Rouet et al., 2016; Stadtler et al., 2014), and situations in which the discrepancy

270 involves the reader’s prior knowledge (De Peryera et al., 2014b) and prior beliefs (Bråten et al., 2016).

271 These extensions have also involved testing the effect in different languages, including English,

272 German, Hebrew, Norwegian, and Spanish (see Braasch & Bråten, 2017). The original D-ISC studies

273 were conducted in French). Of particular note, some research has investigated the relationship between

274 multiple conflicting perspectives outside the original D-ISC scope (e.g., Anmarkrud et al., 2014;

275 Strømsø et al., 2010; von der Mühlen et al., 2016; Wiley et al., 2009). Typically, research within the D-

276 ISC tradition has dealt with factual, flagrant contradictions between two sources and has used the term

277 discrepancy to refer to the situational incoherence. Related research has tended to examine larger sets

278 of sources which present partially inconsistent information, extending not only to facts but also to

279 perspectives and attitudes, and has preferred the term “conflict” to refer to the lack of coherence among

280 the pieces of information. Still, the distinction between discrepancies and conflicts should not be taken

281 categorically. Overall, the empirical data concurs with the idea that readers can use document-level

282 representational elements to create coherence when they perceive a lack of it.

283 4. The role of evaluation when representing multiple texts

284 Creating an interconnected representation of multiple sources and contents implies more than

285 merely mentally coding or symbolizing this information—sources also require evaluation. Many
12
286 features related to source evaluation have been extensively studied by psychologists in context of

287 assessing the persuasiveness of a message (e.g., Wallace, Wegener, & Petty, 2020), our focus here,

288 however, is limited to factors affecting the representation of texts and sources rather than the impact of

289 these representations on changing pre-existing attitudes. Let’s consider the role of evaluation that can

290 occur from a text processing perspective by revisiting the jungle fire example. Imagine now that the

291 same administration that claims the fire is not related to human overexploitation provides a new

292 statement declaring that protected lands are an obstacle to economic growth and commercial

293 exploitation. Would the reader still consider this source trustworthy? Would this alter the organization

294 of the documents model?

295 4.1. First vs. second-hand evaluation

296 Multiple text evaluation refers to making judgments or assessing the different documents in a

297 related set based on the reliability of the information (e.g., Stadtler et al., 2018), defined as the reader’s

298 perception of whether the information is accurate, up-to-date and trustworthy. Bromme, Kienhues, and

299 Porsch (2010) proposed two ways of evaluating the reliability of multiple perspectives: based on the

300 reader’s own understanding of the subject matter (also called first-hand evaluation), or based on the

301 information about the sources (also called second-hand evaluation). The evaluation parameters vary

302 between the two cases. In the first, they are likely to be content-related (e.g., the consistency perceived

303 in the arguments; Scharrer et al., 2017), while in the second they are likely to be source-related. In this

304 later case, trustworthiness is usually described as formed by two basic dimensions: the source’s

305 intentions (e.g., the integrity in adhering to communicational standards and the benevolence toward

306 others) and the source’s ability (e.g., their competence and pertinence with respect to the topic).

307 Importantly, the construction and evaluation of a documents model should not be considered

308 separately. In fact, the relationship between reliability assessment processes, such as the evaluation of

309 the trustworthiness of a source, and other comprehension processes, such as the construction of a

310 coherent representation of discourse, may be more related than traditionally assumed (e.g., Richter &

311 Maier, 2017; O’Brien & Cook, 2016). This is because multiple text evaluation should be seen as an

312 extension of a broader and more basic set of processes, known as text validation. In its basic form, text
13
313 validation implies the tendency to establish consistency within the mental representation and to remedy

314 possible contradictions between different text parts as well as between the text and world knowledge

315 (Singer, 2019). Multiple text evaluation relies on additional skills (e.g., Rouet & Potocki, 2018), but it

316 should be conceived as part of a continuum with more basic comprehension processes, rather than as a

317 completely different activity.

318 4.2. Source evaluation as a resolution strategy: The Content-Source Integration model

319 The impact of text discrepancies on evaluation has been linked to an effortful attempt by the reader to

320 restore coherence, as well as to develop a personal stance toward the conflicting issue. The Content-

321 Source Integration Model (or CSI model, Stadtler & Bromme, 2014) provides a taxonomy of readers’

322 strategic responses to conflicts in texts and links some of these reactions to sourcing strategies. By

323 “sourcing” we mean a reader’s attempts to focus on, retrieve, mention, and/or evaluate sources when

324 performing tasks based on either current or previous reading (Britt et al., 2013; Britt & Rouet, 2012).

325 According to CSI, readers must first detect a coherence problem and classify it as a conflict between

326 text ideas. Next, they attempt to regulate the conflict by organizing the discrepant information. Conflict

327 regulation can be achieved through diverse strategies, such as ignoring the conflict (for instance by

328 picking a side and ignoring the other), reconciling it (by inferentially generating an explanation for the

329 conflicting information), or acknowledging it as due to different points of view (by constructing a

330 documents model). The original findings supporting the D-ISC effect (e.g., Braasch et al., 2012) can be

331 considered an example of the latter type of regulatory reaction. Finally, readers might attempt to resolve

332 the conflict by evaluating the validity of the competing assertions and developing a personal stance.

333 Here, readers can either rely on their own understanding of the subject to determine what to believe, or

334 they can turn to sources they trust to determine whom to believe. Whereas regulation serves mainly to

335 integrate sources with contents, the attempts to resolve the conflict by scrutinizing the trustworthiness

336 of the sources serves to validate the information.

337 Indeed, several studies have reported a difference in how readers inspect and use multiple sources

338 after they evaluate them. Interestingly, however, because the development of a personal stance towards
14
339 a topic leans both on strategic as well as automatic processes, the result of evaluating sources will not

340 always take the same direction. When a reader’s purpose is to actively elaborate a balanced

341 representation of the topic, source evaluation favors increased attention and use of trustworthy contents

342 (e.g., Wiley et al., 2009; Gottschling et al., 2019; Kammerer et al, 2016; Rouet et al., 2020; Salmerón et

343 al., 2016; Saux et al., 2018; Stadtler et al., 2014; von der Mühlen et al., 2016). However, if the evaluation

344 occurs in an automatic and routine fashion, it may lead to a one-sided, biased model that will exclude

345 information from reliable documents if they do not match the reader’s prior knowledge and beliefs (see

346 Richter & Maier, 2017, for a review).

347 5. Constructing coherence from multiple texts: beyond discrepancies

348 Unlike single texts, coherence gaps are to be expected when reading multiple texts about a topic, which

349 in a way defines the modern world. The Documents Model Framework explains how readers can deal

350 with multiple, incoherent texts by signaling the importance of integrating source information into the

351 representation of the texts. This is illustrated by the D-ISC hypothesis in its original formulation and in

352 follow-up research. Building a documents model, however, is not a perfect solution, because most of

353 the time readers also want to know what they should believe in the end. Proposals like the CSI model

354 emphasize the role of evaluation. Readers evaluate the contents and the sources to come up with a single,

355 coherent representation. Thus, a documents model is one way to achieve coherence at the discourse level

356 when confronted with multiple factual inconsistencies, but for many purposes resolving the discrepancy

357 is also desirable.

358 Studying how a documents model can help readers deal with the coherence problem, and how

359 this in turn is constrained by the situation the reader is immersed in, is relevant both for theoretical and

360 practical reasons. At a theoretical level, document models emphasize the situated, purposeful nature of

361 reading in two ways: First, by assuming that the semantic content of the text is integrated with features

362 of the text as an artifact (e.g., who wrote it; Perfetti et al., 1999; Britt et al., 1999). Second, by

363 recognizing that the nature of the representation is contingent upon the reader context and goals.

364 Constructing and evaluating a documents model is one strategy among other, perhaps more

365 parsimonious strategies that enable a form of coherence when reading multiple texts. These include
15
366 constructing a one-sided representation consistent with prior beliefs (Maier & Richter, 2017), drawing

367 unwarranted inferences to reconcile the detected inconsistency (Stadtler & Bromme, 2014), or hedging

368 the content of the text to minimize or self-explain the lack of coherence (Rouet et al., 2016). Why, then,

369 would a reader lean towards such a more demanding strategy? Research has identified several

370 document-related triggers that foster documents model construction, such as the presence of explicit

371 references to sources in the text (Strømsø et al, 2013), the degree of conceptual overlap between texts

372 (Kurby et al., 2005), the perceived difficulty of science documents (e.g., Scharrer et al., 2014), or the

373 position in which source information is presented (e.g., before or after the text contents, Merkt & Huff,

374 2020), among others.

375 Importantly, the effortful process of interconnecting sources with contents is more likely to take

376 place when the task context makes it highly relevant. When readers receive instructions that explicitly

377 recommend verifying the intentions and competencies of the sources, they improve intertext integration

378 and evaluation both during and after reading (see Wiley et al., 2018 for a recent review). Likewise,

379 certain reading goals or situations may require such a representation, as when reading legal briefs that

380 describe the positions of opposing sides of a dispute. Thus, the construction of a documents model is

381 based on knowing that sourcing can be relevant to comprehension, as much as on understanding when

382 and how to apply this knowledge. Recently, Britt and her colleagues (Britt et al., 2018; Rouet et al.,

383 2018) introduced the RESOLV model which construes reading as a problem-solving activity, in which

384 strategic decisions are made on the basis of the context and purpose of the reader. According to

385 RESOLV, the construction of a documents model will be more likely when it becomes instrumental to

386 the task and context.

387 The basic scheme proposed by the DMF has been deepened and complemented by a growing

388 body of research that entails a large range of mental activity, from basic memory processes to high-level

389 decision making, task monitoring and psychosocial processes. Much of the work has followed a

390 cognitive approach to reading comprehension. Given this base, it is perhaps time to begin to better

391 understand the role of motivation or emotion during reading multiple documents. Related proposals

392 have emerged in recent years to emphasize these different aspects, such as the role of prior knowledge
16
393 and beliefs (Richter & Maier, 2017) or the influence of affect and motivation (List & Alexander, 2017).

394 Research examining factors effecting readers’ benefit-cost analysis, proposed by the RESOLV model

395 (Britt et al., 2018), may help to elucidate the important role of motivation. Empirical research has also

396 advanced in describing individual differences and elaborating prescriptions for interventions that foster

397 the use of sources when learning or comprehending from texts (Braasch et al., 2018).

398 Although basic assumptions have been articulated, there is still much to learn about the

399 relationship between performing functional tasks, such as evaluating the reliability of sources, and the

400 organization of the intertext model in memory. A question of interest is whether readers use specific

401 features of a source when reasoning about the validity of the document. Does the name, occupation,

402 motivation of an author play an equivalent role? If some features are more frequently used, does this

403 imply a hierarchy or some organization within the intertext model? Also, more research is needed to

404 better understand the conditions that lead readers to build and use a documents model.

405 Much of the work reviewed here involves issues with creating coherence when the documents

406 include inconsistencies or contradictions. However, there are many situations in which readers must

407 construct an integrated model when there is no inconsistency. For example, when reading about the

408 causes of climate change, the reader may have to use the content from a text about the carbon cycle that

409 never mentions temperature but is relevant to the phenomenon of global temperature change (e.g.,

410 Griffin et al., 2012). In this line, several educational studies have shown that readers struggle to learn

411 from multiple documents even when they do not contain discrepancies (e.g., Cerdán & Vidal-Abarca,

412 2008; Goldman et al., 2019; Griffin et al., 2012; Wiley & Voss, 1999). More research is needed to

413 discover whether prompting readers to encode who says what would facilitate the production of

414 inferences that enable the connection of information across texts. So far, research on multiple document

415 comprehension has focused on contradictory contents because they challenge the very notion of

416 coherence-based integration. However, coherence gaps may also originate in the readers’ difficulty in

417 recognizing that several texts address the same issue, or in reconciling how different perspectives relate

418 to each other. Indeed, the DMF originally highlighted two conditions (in addition to the presence of

419 contradictions) that might prompt readers to construct a documents model: The first condition is when
17
420 one document provides direct evidence for a claim made in another document (argumentative support).

421 The second condition is when information within two separate documents both support a state of affairs

422 initially deemed uncertain (corroboration; Perfetti et al., 1999; Rouet, 2006). In both of these cases, it is

423 likely that source-to-source and source-to-content links would be included in the reader’s final

424 representation of the situation. A comprehensive approach to the problem of coherence in multiple text

425 reading should account for a broad range of intertextual relationships.

426 6. The relevance of multiple text comprehension in an information-rich worldAs research into

427 multiple text comprehension has rapidly expanded in the past 20 years, so has the way in which people

428 access, read and use written information. Rapid technological advances, such as mobile computing,

429 social media, and the profusion of online videos challenge researchers to keep pace with them in the

430 discussion of how people comprehend and learn from multiple documents (Kammerer et al., 2018). For

431 example, the growing Internet access through cell phones brings about a change in operational skills

432 (i.e., skills on how to use digital media tools, Van Deursen & Van Dijk, 2016). The extent to which

433 these differences affect documents model construction, as theorized in this paper, is still an open

434 question. Also, an emerging trend in online reading is to pick information from both written and

435 audiovisual sources. This is fostered by mobile technology which gravitates increasingly towards

436 multimedia communication (photos, videos and audio in messaging and social networks). Multimedia

437 presentations (i.e., combining words and pictures, video or audio; Mayer, 2001) do not imply per se the

438 notion of multiple sources as we have discussed it here. However, audiovisual documents may prompt

439 different evaluation strategies. For instance, Salmerón et al. (2020) reported that, as compared to written

440 format (i.e., a webpage), video format favored source trust and attitude change in primary school

441 students when they learned about a controversial topic (i.e., pros and cons of bottled water use), perhaps

442 because they perceived the video sources as more human and more capable than in the written format,

443 thus prompting a positive evaluation. There have been recent efforts to develop a joint explanation of

444 multiple resource learning (i.e., learning from both multiple representations and multiple perspectives;

445 van Meter et al., 2020), However, to date, research into multimedia and multi-source comprehension is

446 still scarce and deserves further efforts.

18
447 Children's increased exposure to multiple and sometimes dubious sources of information also

448 warrants an increased effort to teach multiple text comprehension skills. Multiple text comprehension

449 is supported by cognitively demanding processes and complex personal epistemic constructs, which are

450 usually considered a dimension of adult literacy. However, five-year old children already show some

451 notion “that knowledge is unequally distributed and that their own qualification as a source of

452 knowledge varies depending on the situation” (Stadtler et al., 2018, p. 49, referring to research by

453 Danovitch & Keil, 2004). In addition, teachers make ample use of documents of all sorts when teaching

454 subject matters such as history or science. Therefore, it appears critical to prepare young students to

455 tackle the challenges of multiple text comprehension. Up to now, however, intervention studies aiming

456 to promote primary school students' attention and use of sources as part of reading are still scarce. In

457 2018, Brand-Gruwel and van Strien published a systematic review and identified only four articles in

458 English, dated after 2000, aimed at fostering primary/elementary students’ sourcing skills (i.e., de Vries

459 et al., 2008; Kroustallaki et al., 2015; Kuiper et al., 2008; Macedo-Rouet et al., 2013). Whereas the four

460 studies presented variations, they all included training and assessing of the children’ skills to search,

461 evaluate, and select credible sources. All studies showed some improvement after the intervention,

462 although the effect tended to be qualified by various variables. Research on the domain of history

463 teaching in elementary school have yielded similar conclusions (e.g., VanSledright, 2002). Overall,

464 these studies suggest that some basic features of multiple text comprehension can fruitfully be taught to

465 developing readers, although multiple document literacy is also an important goal for secondary and

466 higher education. A conception of reading as a problem-solving activity (i.e., an activity that requires

467 gathering spread and diverse information and making decisions based on specific reading purposes) has

468 become essential to the definition of literacy and even of responsible citizenship in the twenty-first

469 century (Britt et al., 2018). Therefore, an empirical understanding of how multiple text comprehension

470 skills develop and how education can support them has become increasingly relevant, both for current

471 research and practice.

472 From a practical angle, the problem of coherence in multiple text reading is inextricably linked

473 to the massification of the internet and the general public's increased access to a vast and heterogeneous
19
474 number of texts. An important implication from the work presented in this paper is that being aware of

475 the versatility of reading and of the fact that texts are produced and spread with various communicative

476 intents, and knowing how to use this knowledge purposefully, has become a distinctive skill of proficient

477 readers. Educating someone to integrate and evaluate multiple texts through sourcing in the light of a

478 specific task and context may seem a massive undertaking. Indeed, despite the conditions that promote

479 document model formation, such as detecting intertextual contradictions, readers struggle to construct

480 an integrated intertext model. For example, List et al. (2019) reported that only a quarter of a sample of

481 undergraduates produced a documents model after a multiple text reading task, with the rest producing

482 mostly separate representations of the texts or mush models that do not differentiate where the

483 information came from. This deficit has stimulated an increasing number of intervention approaches that

484 try to direct readers’ attention to the construction of coherent intertext models. So far, the data support

485 the claim that short, theory-driven and targeted interventions can actually foster sourcing and multiple-

486 text integration successfully (Barzilai et al, 2018; Brante & Strømsø., 2017; McGrew et al., 2019, Pérez

487 et al., 2018). The effectiveness of these short-term focused interventions may lie in the fact that most

488 readers are aware that there is no text without a source, that multiple texts can diverge, or that their

489 standards for processing information vary as a function of context and stakes. They emphasize, however,

490 that in today's world, the teaching of reading comprehension extends to connecting heterogeneous pieces

491 of discourses, connecting contents to sources and adjusting coherence standards to one's goals and

492 purposes.

493 7. Further readings

494 Several models of multiple text comprehension subsequent to the DMF framework were

495 compiled in a special issue of Educational Psychologist, edited by List and Alexander (2017). For an

496 extensive review of the factors affecting source representation and use we recommend consulting the

497 Handbook of Multiple Source Use, edited by Braasch, Bråten, and McCrudden (2018) and the Handbook

498 of Learning from Multiple Representations and Perspectives, edited by Van Meter, List, Lombardi, &
20
499 Kendeou (2020). For a more comprehensive presentation of the RESOLV model highlighting the

500 importance of context and decision making in reading, please refer to Britt, Rouet, & Durik (2018).

501

502 8. Acknowledgements

503 Support for this work was provided in part by a grant from the Fulbright foundation to the last author

504 and by the Center for the Interdisciplinary Study of Language and Literacy (CISLL) at Northern Illinois

505 University.

506

507 9. References

508 Albrecht, J. E., & O'Brien, E. J. (1993). Updating a mental model: Maintaining both local and global

509 coherence. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19(5), 1061–

510 1070. [Link]

511 Anmarkrud, Ø., Bråten, I., & Strømsø, H. I. (2014). Multiple-documents literacy: Strategic processing,

512 source awareness, and argumentation when reading multiple conflicting documents. Learning and

513 Individual Differences, 30, 64–76. [Link]

514 Baker, L. (1979). Comprehension monitoring: Identifying and coping with text confusions. Journal of

515 reading behavior, 11(4), 365-374. doi:10.1080/10862967909547342

516 Barzilai, S., & Eshet-Alkalai, Y. (2015). The role of epistemic perspectives in comprehension of

517 multiple author viewpoints. Learning and Instruction, 36, 86–

518 103. [Link]

519 Barzilai, S., Zohar, A. R. & Mor-Hagani, S. (2018). Promoting integration of multiple texts: a review of

520 instructional approaches and practices. Educational Psychology Review, 30, 973–999.

521 [Link]

522 Beker, K., Jolles, D., Lorch, R. F., & Broek, P. (2016). Learning from texts: Activation of information

523 from previous texts during reading. Reading and Writing: An Interdisciplinary Journal, 29(6),

524 1161–1178. [Link]

21
525 Black, A., Freeman, P., & Johnson-Laird, P. N. (1986). Plausibility and the comprehension of

526 text. British Journal of Psychology, 77(1), 51–62. [Link]

527 8295.1986.tb01980.x

528 Braasch, J. L. G., & Bråten, I. (2017). The discrepancy-induced source comprehension (D-ISC) model:

529 Basic assumptions and preliminary evidence. Educational Psychologist, 52(3), 167–

530 181. [Link]

531 Braasch, J. L. G., Bråten, I., McCrudden, M. T. (2018). Introduction to research on multiple source use.

532 In J. L. G. Braasch, I. Bråten, & M. T. McCrudden, (Eds.), Handbook of Multiple Source Use (pp.

533 17-46). Routledge. [Link]

534 Braasch, J. L. G., Rouet, J.-F., Vibert, N., & Britt, M. A. (2012). Readers’ use of source information in

535 text comprehension. Memory and Cognition, 40(3), 450-65. [Link]

536 0160-6

537 Braasch, J.L.G. & Scharrer, L. (2020). The role of cognitive conflict in understanding and learning from

538 multiple texts. In P. van Meter, A. List, D. Lombardi, Kendeou, P, (Eds.), Handbook of learning

539 from multiple representations and perspectives. Routledge.

540 Bråten, I., & Braasch, J. L. G. (2018). The role of conflict in multiple source use. In J. L. G. Braasch, I.

541 Bråten, & M. T. McCrudden (Eds.) Handbook of multiple source use (pp. 184-201). Routledge.

542 Bråten, I., Salmerón, L., & Strømsø, H. I. (2016). Who said that? Investigating the Plausibility-Induced

543 Source Focusing assumption with Norwegian undergraduate readers. Contemporary Educational

544 Psychology, 46, 253-262. [Link]

545 Brand - Gruwel, S. & van Strien, J. L. H. (2018). Instruction to promote information-problem solving

546 on the Internet in primary and secondary education students: a Systematic Literature Review. In

547 J. L. G. Braasch, I. Bråten, & M. T. McCrudden (Eds.), Handbook of Multiple Source Use (pp.

548 401-422). Routledge. [Link]

549 Brante, E. W., Strømsø, H. I.(2018). Sourcing in text comprehension: a review of interventions targeting

550 sourcing skills. Educational Psychology Review, 30, 773–799. [Link]

551 017-9421-7
22
552 Britt, M. A., & Rouet, J.-F. (2012). Learning with multiple documents: Component skills and their

553 acquisition. In M. J. Lawson & J. R. Kirby (Eds.), The quality of learning: Dispositions,

554 instruction, and mental structures. Cambridge University Press.

555 Britt, M. A., Perfetti, C. A., Sandak, R., & Rouet, J.-F. (1999). Content integration and source separation

556 in learning from multiple texts. In S. R. Goldman, A. C. Graesser, & P. van den Broek

557 (Eds.), Narrative comprehension, causality, and coherence: Essays in honor of Tom Trabasso (p.

558 209–233). Lawrence Erlbaum Associates Publishers.

559 Britt, M. A., Rouet, J.-F., & Braasch, J. L. G. (2013). Documents experienced as entities: Extending the

560 situation model theory of comprehension. In M. A. Britt, S. R. Goldman, & J.-F. Rouet (Eds.),

561 Reading: from words to multiple texts (pp. 160–179). Routledge.

562 Britt, M. A., Rouet, J.-F., & Durik, A. (2018). Literacy beyond text comprehension: a theory of

563 purposeful reading. Routledge. [Link]

564 Bromme, R., Kienhues, D., & Porsch, T. (2010). Who knows what and who can we believe?

565 Epistemological beliefs are beliefs about knowledge (mostly) to be attained from others. In L. D.

566 Bendixen & F. C. Feucht (Eds.), Personal epistemology in the classroom: Theory, research, and

567 implications for practice (pp. 163 – 193). Cambridge University Press.

568 Cerdán, R., & Vidal-Abarca, E. (2008). The effects of tasks on integrating information from multiple

569 documents. Journal of Educational Psychology, 100(1), 209–222. [Link]

570 0663.100.1.209

571 Danovitch, J. H., & Keil, F. C. (2004). Should You Ask a Fisherman or a Biologist?: Developmental

572 Shifts in Ways of Clustering Knowledge. Child Development, 75(3), 918–

573 931. [Link]

574 De Pereyra, G., Belkadi, S., Marbach, L., & Rouet, J.-F. (2014a, August). Do teenage readers use source

575 information when faced with discrepant information? Paper presented at the 24th annual meeting

576 of the Society for Text and Discourse, Chicago, IL.

23
577 De Pereyra, G., Britt, M. A., Braasch, J. L. G, & Rouet, J.-F. (2014b). Reader's memory for information

578 sources in simple news stories: Effects of text and task features. Journal of Cognitive Psychology,

579 24(2), 187-204.

580 de Vries, B., Van der Meij, H., & Lazonder, A. W. (2008). Supporting reflective Web searching in

581 elementary schools. Computers in Human Behavior, 24, 649–666.

582 [Link]

583 Gernsbacher, M. A. (1990). Language comprehension as structure building. Lawrence Erlbaum

584 Associates, Inc.

585 Goldman, S. R., Golden, R., & van den Broek, P. (2007). Why are computational models of text

586 comprehension useful? In F. Schmalhofer & C. Perfetti (Eds.), Higher level language processes

587 in the brain. Lawrence Erlbaum Associates, Inc.

588 Goldman, S. R., Greenleaf, C., Yukhymenko-Lescroart, M., Brown, W., Ko, M-L, Emig, J. M.,

589 George, M. A., Wallace, P., Blaum, D., Britt, M. A., & Project READI (2019). Explanatory

590 modeling in science through text-based investigation: Testing the efficacy of the Project READI

591 intervention approach. American Educational Research Journal, 56, 1148–1216.

592 [Link]

593 Goldman, S. R., & Varma, S. (1995). CAPing the construction-integration model of discourse

594 comprehension. In C. A. Weaver III, S. Mannes, & C. R. Fletcher (Eds.), Discourse

595 comprehension: Essays in honor of Walter Kintsch (p. 337–358). Lawrence Erlbaum Associates,

596 Inc.

597 Gottschling, S., Kammerer, Y., & Gerjets, P. (2019). Readers’ processing and use of source information

598 as a function of its usefulness to explain conflicting scientific claims. Discourse Processes, 56(5-

599 6), 429-446. [Link]

600 Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative text

601 comprehension. Psychological Review, 101(3), 371–395. [Link]

602 295X.101.3.371
24
603 Griffin, T. D., Wiley, J., Britt, M. A., & Salas, C. R. (2012). The role of CLEAR thinking in learning

604 science from multiple-document inquiry tasks. International Electronic Journal of Elementary

605 Education, 5 (1) 63-78.

606 Kammerer, Y., Brand-Gruwel, S., & Jarodzka, H. (2018). The future of learning by searching the Web:

607 mobile, social, and multimodal. Frontline Learning Research, 6, 81-

608 91. [Link]

609 Kammerer, Y., Kalbfell, E., & Gerjets, P. (2016). Is this information source commercially biased? How

610 contradictions between web pages stimulate the consideration of source information. Discourse

611 Processes, 53(5-6), 430–456. [Link]

612 Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration

613 model. Psychological Review, 95(2), 163–182. [Link]

614 Kintsch, W. (1998). Comprehension: A paradigm for cognition. Cambridge University Press.

615 Kroustallaki, D., Kokkinaki, T., Sideridis, G. D., & Simos, P. G. (2015). Exploring students’ affect and

616 achievement goals in the context of an intervention to improve web searching skills. Computers

617 in Human Behavior, 49, 156–170. [Link]

618 Kuiper, E., Volman, M., & Terwel, J. (2008). Integrating critical Web skills and content knowledge:

619 Development and evaluation of a 5th grade educational program. Computers in Human Behavior,

620 24(3), 666-692. [Link]

621 Kurby, C. A., Britt, M. A., & Magliano, J. P. (2005). The role of top-down and bottom-up processes in

622 between-text integration. Reading Psychology, 26(4-5), 335–

623 362. [Link]

624 List, A. & Alexander, P.A. (2017). Cognitive Affective Engagement Model of Multiple Source Use.

625 Educational Psychologist, 52(3), 182-199. [Link]

626 List, A., Du, H., Wang, Y., & Lee, H. Y. (2019). Toward a typology of integration: Examining the

627 documents model framework. Contemporary Educational Psychology, 58, 228–

628 242. [Link]

25
629 Macedo-Rouet, M., Braasch, J. L. G., Britt, M. A., & Rouet, J.-F. (2013) Teaching Fourth and Fifth

630 Graders to Evaluate Information Sources During Text Comprehension. Cognition and

631 Instruction, 31(2), 204-226. [Link]

632 Mayer, R. E. (2001). Multimedia learning. Cambridge University Press.

633 [Link]

634 McGrew, S., Smith, M., Breakstone, J., Ortega, T., & Wineburg, S. (2019). Improving university

635 students’ web savvy: An intervention study. British Journal of Educational Psychology, 89(3),

636 485-500. [Link]

637 McKoon, G., & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99(3), 440–

638 466. [Link]

639 McNamara, D. S., & Magliano, J. (2009). Toward a comprehensive model of comprehension. In B. H.

640 Ross (Ed.), The psychology of learning and motivation: Vol. 51. The psychology of learning and

641 motivation (p. 297–384). Elsevier Academic Press. [Link]

642 7421(09)51009-2

643 Merkt, M., & Huff, M. (2020). Does the position of source information for multiple documents matter?

644 Insights from two experiments. Contemporary Educational Psychology, 62,

645 101900. [Link]

646 Myers, J. L., & O'Brien, E. J. (1998). Accessing the discourse representation during reading. Discourse

647 Processes, 26(2-3), 131–157. [Link]

648 O'Brien, E. J., & Cook, A. E. (2016). Coherence threshold and the continuity of processing: The RI-Val

649 model of comprehension. Discourse Processes, 53(5-6), 326–

650 338. [Link]

651 Pérez, A., Potocki, A., Stadtler, M., Macedo-Rouet, M., Paul, J., Salmerón, L., & Rouet, J. F. (2018).

652 Fostering teenagers' assessment of information reliability: Effects of a classroom intervention

653 focused on critical source dimensions. Learning and Instruction, 58, 53-64.

654 [Link]
26
655 Perfetti, C. A., Rouet, J.-F., & Britt, M. A. (1999). Toward a theory of documents representation. In H.

656 van Oostendorp & S. R. Goldman (Eds.), The construction of mental representations during

657 reading (p. 99–122). Lawrence Erlbaum Associates Publishers.

658 Richter, T. & Maier, J. (2017). Comprehension of multiple documents with conflicting information: a

659 two-step model of validation. Educational Psychologist, 52(3), 148-166.

660 [Link]

661 Rouet, J.-F. (2006). The skills of document use: From text comprehension to web-based

662 learning. Lawrence Erlbaum Associates Publishers.

663 Rouet, J.-F., & Britt, M. A. (2014). Multimedia learning from multiple documents. In R. E. Mayer

664 (Ed.), Cambridge handbooks in psychology. The Cambridge handbook of multimedia learning (p.

665 813–841). Cambridge University Press. [Link]

666 Rouet, J.-F., Britt, M.A., & Potocki, A. (2019). Multiple-text comprehension. In J. Dunlosky & K.

667 Rawson (Eds.), The Cambridge Handbook of Cognition and Education (pp. 356-380). Cambridge

668 University Press. [Link]

669 Rouet, J.-F. & Potocki, A. (2018). From reading comprehension to document literacy: learning to search

670 for, evaluate and integrate information across texts. Journal of the Study of Education and

671 Development, 41(3), 415-446. [Link]

672 Rouet, J.-F., Le Bigot, L., de Pereyra, G., & Britt, M. A. (2016). Whose story is this? Discrepancy

673 triggers readers’ attention to source information in short narratives. Reading and Writing: An

674 Interdisciplinary Journal, 29(8), 1549–1570. [Link]

675 Rouet, J.-F., Levonen, J. J., Dillon, A., & Spiro, R. J. (Eds.). (1996). Hypertext and cognition. Lawrence

676 Erlbaum Associates, Inc.

677 Rouet, J.-F., Saux, G., Ros, C., Stadtler, M., Vibert, N., & Britt, M. A. (2020). Inside document models:

678 The role of source attributes in integrating multiple text contents. Discourse Processes.

679 [Link]
27
680 Salmerón, L., Macedo-Rouet, M., & Rouet, J.-F. (2016). Multiple viewpoints increase students' attention

681 to source features in social question and answer forum messages. Journal of the Association for

682 Information Science and Technology, 67(10), 2404-2419. [Link]

683 Salmerón, L., Sampietro, A., & Delgado, P. (2020). Using internet videos to learn about controversies:

684 Evaluation and integration of multiple and multimodal documents by primary school

685 students. Computers & Education, 148, Article

686 103796. [Link]

687 Salmerón, L., Strømsø, H. I., Kammerer, Y., Stadtler, M., & van den Broek, P. (2018). Comprehension

688 processes in digital reading. In M. Barzillai, J. Thomson, S. Schroeder, & P. van den Broek (Eds.),

689 Learning to read in a digital world (pp 91-120). John Benjamins.

690 Saux, G., Britt, M. A., Le Bigot, L., Vibert, N., Burin, D., & Rouet, J.-F. (2017). Conflicting but close:

691 Readers’ integration of information sources as a function of their disagreement. Memory &

692 Cognition, 45(1), 151–167. [Link]

693 Saux, G., Ros, C., Britt, M. A., Stadtler, M., Burin, D. I. & Rouet J.-F. (2018). Readers’ selective recall

694 of source features as a function of claim discrepancy and task demands. Discourse Processes.

695 [Link]

696 Scharrer, L., Rupieper, Y., Stadtler, M., & Bromme, R. (2017). When science becomes too easy: Science

697 popularization inclines laypeople to underrate their dependence on experts. Public Understanding

698 of Science, 26, 1003-1018. [Link]

699 Scharrer, L., Stadtler, M., & Bromme, R. (2014). You'd better ask an expert: Mitigating the

700 comprehensibility effect on laypeople's decisions about science‐based knowledge claims. Applied

701 Cognitive Psychology, 28(4), 465–471. [Link]

702 Singer, M. (2019). Challenges in processes of validation and comprehension. Discourse Processes,

703 56(5-6), 465–483. [Link]

704 Stadtler, M., & Bromme, R. (2014). The content-source integration model: A taxonomic description of

705 how readers comprehend conflicting scientific information. In D. N. Rapp & J. L. G. Braasch
28
706 (Eds.), Processing inaccurate information: Theoretical and applied perspectives from cognitive

707 science and the educational sciences (p. 379–402). MIT Press.

708 Stadtler, M., Bromme, R., & Rouet, J.-F. (2018). Learning from multiple documents: How can we foster

709 multiple document literacy skills in a sustainable way? In E. Manalo, Y. Uesaka, & C. A. Chinn

710 (Eds.), Routledge research in achievement and gifted education. Promoting spontaneous use of

711 learning and reasoning strategies: Theory, research, and practice for effective transfer (pp. 46–

712 61). Routledge.

713 Stadtler, M., Scharrer, L., Skodzik, T., & Bromme, R. (2014). Comprehending multiple documents on

714 scientific controversies: Effects of reading goals and signaling rhetorical relationships. Discourse

715 Processes, 51(1-2), 93–116. [Link]

716 Strømsø, H. I., Bråten, I., & Britt, M. A. (2010). Reading multiple texts about climate change: The

717 relationship between memory for sources and text comprehension. Learning and Instruction,

718 20(3), 192–204. [Link]

719 Strømsø, H. I., Bråten, I., Britt, M. A., & Ferguson, L. E. (2013). Spontaneous sourcing among students

720 reading multiple documents. Cognition and Instruction, 31(2), 176–

721 203. [Link]

722 United Nations (2019, September 24). Brazilian President speaks out against ‘media lies’ surrounding

723 Amazon fires. UN News. Retrieved from [Link]

724 van den Broek, P., & Gustafson, M. (1999). Comprehension and memory for texts: Three generations

725 of reading research. In S. R. Goldman, A. C. Graesser, & P. van den Broek (Eds.), Narrative

726 comprehension, causality, and coherence: Essays in honor of Tom Trabasso (p. 15–34). Lawrence

727 Erlbaum Associates Publishers.

728 van den Broek, P., Risden, K., Fletcher, C. R., & Thurlow, R. (1996). A “landscape” view of

729 reading: Fluctuating patterns of activations and the construction of a stable memory

730 representation. In. B. K. Britton & A. C. Graesser (Eds.), Models of Understanding Text (pp.

731 165-187). Lawrence Erlbaum Associates, Inc.

29
732 van Deursen, A., Helsper, E. J., & Eynon, R. (2016) Development and validation of the Internet Skills

733 Scale (ISS), Information, Communication & Society, 19:6, 804-

734 823, [Link]

735 van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. Academic Press.

736 Van Meter, P., List, A., Lombardi, D., & Kendeou, P. (2020). Handbook of Learning from Multiple

737 Representations and Perspectives. Routledge. [Link]

738 van Oostendorp, H. (1996). Updating situation models derived from newspaper articles.

739 Medienpsychologie, 8, 21–33.

740 van Oostendorp, H. (2014). The ambivalent effect of focus on updating mental representations. In D. N.

741 Rapp & J. L. G. Braasch (Eds.), Processing inaccurate information: Theoretical and applied

742 perspectives from cognitive science and the educational sciences (p. 223-243). MIT Press.

743 van Oostendorp, H., & Bonebakker, C. (1999). Difficulties in updating mental representations during

744 reading news reports. In H. van Oostendorp & S. R. Goldman (Eds.), The construction of mental

745 representations during reading (p. 319–339). Lawrence Erlbaum Associates Publishers.

746 VanSledright, B. A. (2002). Fifth graders investigating history in the classroom: Results from a

747 researcher-practitioner design experiment. The Elementary School Journal, 103(2), 131–

748 160. [Link]

749 VanSledright, B. A., & Frankes, L. (2000). Concept-and strategic-knowledge development in historical

750 study: A comparative exploration in two fourth-grade classrooms. Cognition and Instruction,

751 18(2), 239-283. https;//[Link]/10.1207/S1532690XCI1802_04

752 Von der Mühlen, S., Richter, T., Schmid, S., Schmidt, E. M., and Berthold, K. (2016). The use of source-

753 related strategies in evaluating multiple psychology texts: a student–scientist comparison. Reading

754 and Writing, 29, 1677–1698. [Link]

755 Wallace, L. E., Wegener, D. T., & Petty, R. E. (2020). When sources honestly provide their biased

756 opinion: Bias as a distinct source perception with independent effects on credibility and persuasion.

757 Personality and Social Psychology Bulletin, 46(3), 439-453.

30
758 Wiley, J., Jaeger, A. J., & Griffin, T. D. (2018). Effects of task instructions on comprehension from

759 multiple sources in history and science. In J. L. G. Braasch, I. Braten, & M. T. McCrudden

760 (Eds.) Handbook of multiple source use (pp. 341-361). Routledge.

761 Wiley, J., Goldman, S. R., Graesser, A. C., Sanchez, C. A., Ash, I. K., & Hemmerich, J. A. (2009).

762 Source evaluation, comprehension, and learning in Internet science inquiry tasks. American

763 Educational Research Journal, 46(4), 1060–1106. [Link]

764 Wiley, J., & Voss, J. F. (1999). Constructing arguments from multiple sources: Tasks that promote

765 understanding and not just memory for text. Journal of Educational Psychology, 91(2), 301–

766 311. [Link]

767 Wilkes, A. L., & Leatherbarrow, M. (1988). Editing episodic memory following the identification of

768 error. The Quarterly Journal of Experimental Psychology A: Human Experimental Psychology,

769 40A(2), 361–387. [Link]

770 Wineburg, S. S. (1991). Historical problem solving: A study of the cognitive processes used in the

771 evaluation of documentary and pictorial evidence. Journal of Educational Psychology, 83(1), 73–

772 87. [Link]

773 World Health Organization. (2020, February 2). Novel Coronavirus (2019-nCoV). Situation report -

774 13. Retrieved from [Link]

775 reports/[Link]

776 Zwaan, R. A., & Radvansky, G. A. (1998). Situation models in language comprehension and

777 memory. Psychological Bulletin, 123(2), 162–185. [Link]

778
31
779 Figure captions

780

781 Figure 1. Components of a documents model (adapted from Britt & Rouet, 2012).

782 S-S: source-to-source; S-C: source-to-content

783

784
32

A
B
TEXTS (contents +
information about the sources)

Reader’s mental representation

Document node S-C Link Mental Model

(SOURCE A) Text A

S-S Link

Document node S-C Link Mental Model Text

(SOURCE B) B

Intertext Model (document nodes Integrated Mental

and links) Model

785