Integrative processes in reading comprehension

Integrative processes in reading comprehension

First Author: Ludo Verhoeven -- Radboud University Nijmegen
Abstract / Summary: 

Comprehending what is read requires the dynamic interplay of reading and language subsystems that bring about word identification, word meaning retrieval and integration, sentence parsing, referential-binding, and text memory guided by relevant knowledge and general cognitive resources. Central in this interplay are the integrative processes that result in comprehension. In the present symposium, we bring together research on different levels of integration and different research approaches to integration in diverse groups of learners. The symposium combines four lines of research: (1) modeling the linguistic components in text comprehension, (2) studying the self-paced reading times and individual differences in text integration processes, (3) identifying ERP indicators of  word-to-text integration, and (4) uncovering the temporal flow in word-to-text integration processes via eye tracking and imaging. The outcomes of these complementary research paradigms will be discussed in perspective of a neurocognitive account of integrative processes in reading comprehension.

Symposium Papers: 

Linguistic components in early first and second language reading comprehension

First Author/Chair:Ludo Verhoeven -- Radboud University Nijmegen
Additional authors/chairs: 
Marinus Voeten

Purpose: According to the simple view of reading, reading comprehension is the product of word decoding and linguistic comprehension. Against this background, we examined the additional role of early lexical quality in the prediction of reading comprehension, either directly or indirectly via word decoding or linguistic comprehension in early Dutch first and second language learners. 

Method: Following a longitudinal design, 566 children learning to read Dutch as L1 and 463 children learning to read Dutch as L2 were tested on speech decoding, morphological knowledge and vocabulary as indicators of lexical quality in kindergarten; word decoding and listening comprehension in first grade; and then reading comprehension in second grade. 

Results: The results showed L2 learners to consistently lag behind L1 readers on all measures except for word decoding. Both word decoding and listening comprehension predicted later reading comprehension for not only L1 but also L2 learners. However, later reading comprehension was also directly predicted by the children’s early morphological and vocabulary knowledge, on the one hand, and indirectly by speech decoding and morphological knowledge via word decoding and indirectly by morphological and vocabulary knowledge via listening comprehension. 

Conclusions: Although the lexical quality and listening comprehension of beginning L2 readers generally lags behind beginning L1 readers, the reading comprehension for both groups appears to build upon the same set of linguistic predictors. For both L1 and L2 learners it was found to be largely predicted by the quality of their early lexicons as measured by speech decoding, morphological knowledge and vocabulary.

Role of word-to-text integration and lexical predictors in ESL reading comprehension

First Author/Chair:Evelien Mulder -- Radboud University Nijmegen
Additional authors/chairs: 
Maco van de Ven; Eliane Segers; Alexander Krepel; Elise de Bree; Peter de Jong; Ludo Verhoeven

Purpose: Word-to-text integration which can be seen as a dynamic aspect of reading comprehension has received only scant attention in the literature on second language learning. The present study aimed to examine to what extent word-to-text integration (WTI) as a dynamic measure predicts reading comprehension in novice ESL learners after controlling for static lexical quality measures such as decoding, vocabulary ad morphological awareness. 

Method:Seventh grade ESL students (N = 449) in the Netherlands performed a self-paced reading task with simple and complex anaphora, inferencing, and anomaly sentences. For each of the three sentence types a WTI-index was created by dividing reading times on complex sentences between the reading times on simple sentences. Separate reading times for each sentence type were also included in the analyses. Lexical predictors were measured using standardized tasks. 

Results: Linear regression models were used to disentangle the contributions of dynamic and static measures of WTI to explaining reading comprehension ability. Static measures indicated that students with strong word decoding and syntactic skills and large vocabulary sizes had better reading comprehension skills. On top of these static measures, dynamic measures indicated that students who showed larger processing costs for complex relative to simple inferencing sentences showed better reading comprehension skills. 

Conclusions: It can thus be concluded that the dynamic measure of WTI predicts ESL reading comprehension, over and above static lexical quality predictors. Importantly, the depth of WTI processing on the part of the learners showed a positive relationship with their reading comprehension outcomes.

Incremental processes in text comprehension: Sources for word-to-text integration captured by Event Related Potentials

First Author/Chair:Anne Helder -- Leiden University
Additional authors/chairs: 
Charles Perfetti

Purpose: The study of word-to-text integration provides a window on incremental processes that either link the meaning of words at sentence beginnings to the preceding text or require the reader to build a new structure. In a series of ERP experiments we examined how conditions favor one (integration) over the other (structure building). 

Method: Undergraduate students read short passages (2-4 sentences) while their EEGs were recorded. ERPs (N400) on sentence-initial nouns were segmented and compared across conditions that differed in the way a passage provides a source for integration; the preceding sentence (a co-referential antecedent), the preceding text (the central theme), or not at all (e.g., at text beginnings). 

Results: When sentence-initial nouns trigger the retrieval of a text memory from the preceding sentence, a reduction in the N400 occurs, reflecting meaning integration across the sentence boundary. Additional integration with the preceding text, shown in N400 reductions for words that were related to the central theme, occurred only when instructions required readers to engage in global integration. N400s were most negative for words with no integration opportunity, reflecting readers built a new structure without integration. 

Conclusions: Results across experiments show that integration is determined by a) availability of a retrieval cue, b) the reader’s goals, and c) the requirement for structure building. The structure building requirement is present at the beginning of every sentence but is the only option at the beginning of a text. The theoretical implication is that integration and structure building can occur together at sentence beginnings.

Assessing combinatorial processing in skilled- and less-skilled comprehenders

First Author/Chair:Julie van Dyke -- Haskins Laboratories
Additional authors/chairs: 
Luca Campanelli; John Hale; Roeland Hancock; Nicole Landi

Purpose: Combinatorial processing refers to the ability to combine the meaning of a word with its surrounding context.  The crucial issue for assessing skilled comprehension is "How accurately and completely are these combinatorial processes executed?"  We use computational modelling to assess the effort necessary at each level (i.e., syntax, semantics, discourse) during integration. 

Method: Skilled- and less-skilled comprehenders (age 13-17) read naturally occurring texts during fMRI scanning with simultaneous eye-tracking (current N=16; data collection ongoing).  We use corpus-based surprisal, the inverse probability of a word occurring in a specific context, to link word-by-word predictions from our computational models to BOLD signal.  Through model fitting, we will identify the neural regions associated with low and high combinatorial processing in relation to skill.  We compare syntactic models of minimal combinatorial processing that encode only the linear order of words vs. those that encode information about constituents and hierarchical relations. For semantics, we compare low- vs. high-dimensional vector representations. 

Results:  Previous work implicates left anterior temporal and inferior frontal regions with combinatorial processing in adults.  Preliminary analyses show these regions are related to high and low combinatorial processing, respectively; skill-related analyses have not yet been done.  Eye-movements of skilled readers are better modeled by high-combinatorial models, whereas those of less-skilled readers are modeled by low-combinatorial models. 

Conclusions: Results indicate the quality of word-by-word combinatorial processing, formalized in our computational models, and the eye-movement patterns and neural regions that support it, in relation to comprehension ability.

Discussion on integrative processes in reading comprehension

First Author/Chair:DISCUSSANT Charles Perfetti -- University of Pittsburgh

The contributions of the four symposium presentations will be discussed in relation to the broad concept of integration in reading. Because the presentations range widely in goals, methods and in what is meant by integration, an attempt will be made to provide an integrative perspective on key conclusions. It will be concluded that, fundamentally, reading is integration across multiple levels of text and reader knowledge.