Computational Models Of Reading Research Articles

Prediction in reading: A review of predictability effects, their theoretical implications, and beyond.

Historically, prediction during reading has been considered an inefficient and cognitively expensive processing mechanism given the inherently generative nature of language, which allows upcoming text to unfold in an infinite number of possible ways. This article provides an accessible and comprehensive review of the psycholinguistic research that, over the past 40 or so years, has investigated whether readers are capable of generating predictions during reading, typically via experiments on the effects of predictability (i.e., how well a word can be predicted from its prior context). Five theoretically important issues are addressed: What is the best measure of predictability? What is the functional relationship between predictability and processing difficulty? What stage(s) of processing does predictability affect? Are predictability effects ubiquitous? What processes do predictability effects actually reflect? Insights from computational models of reading about how predictability manifests itself to facilitate the reading of text are also discussed. This review concludes by arguing that effects of predictability can, to a certain extent, be taken as demonstrating evidence that prediction is an important but flexible component of real-time language comprehension, in line with broader predictive accounts of cognitive functioning. However, converging evidence, especially from concurrent eye-tracking and brain-imaging methods, is necessary to refine theories of prediction.

Relationship Between Resting State Functional Connectivity and Reading-Related Behavioural Measures in 69 Adults

Abstract In computational models of reading, written words can be read using print-to-sound and/or print-to-meaning pathways. Neuroimaging data associate dorsal stream regions (left posterior occipitotemporal cortex, intraparietal cortex, dorsal inferior frontal gyrus [dIFG]) with the print-to-sound pathway and ventral stream regions (left anterior fusiform gyrus, middle temporal gyrus) with the print-to-meaning pathway. In 69 typical adults, we investigated whether resting state functional connectivity (RSFC) between the visual word form area (VWFA) and dorsal and ventral regions correlated with phonological (nonword reading, nonword repetition, spoonerisms), lexical-semantic (vocabulary, sensitivity to morpheme units in reading), and general literacy (word reading, spelling) skills. VWFA activity was temporally correlated with activity in both dorsal and ventral reading regions. In pre-registered whole-brain analyses, spoonerisms performance was positively correlated with RSFC between the VWFA and left dorsal regions (dIFG, superior parietal and intraparietal cortex). In exploratory region-of-interest analyses, VWFA-dIFG connectivity was also positively correlated with nonword repetition, spelling, and vocabulary. Connectivity between the VWFA and ventral stream regions was not associated with performance on any behavioural measure, either in whole-brain or region-of-interest analyses. Our results suggest that tasks such as spoonerisms and spellings, which are both complex (i.e., involve multiple subprocesses) and have high between-subject variability, provide greater opportunity for observing resting-state brain-behaviour associations. However, the complexity of these tasks limits the conclusions we can draw about the specific mechanisms that drive these associations. Future research would benefit from constructing latent variables from multiple tasks tapping the same reading subprocess.

The influence of oral vocabulary knowledge on individual differences in a computational model of reading

Studies have demonstrated systematic individual differences in the degree of semantic reliance (SR) when reading aloud exception words in adult skilled readers. However, the origins of individual differences in reading remain unclear. Using a connectionist model of reading, this study investigated whether oral vocabulary knowledge may affect the degree of SR as a potential source of individual differences in reading. Variety in oral vocabulary knowledge was simulated by training the model to learn the mappings between spoken and meaning word forms with different vocabulary sizes and quantities of exposure to these vocabularies. The model’s SR in the reading aloud task was computed. The result demonstrated that the model with varying amounts of oral exposure and vocabulary sizes had different levels of SR. Critically, SR was able to predict the performance of the model on reading aloud and nonword reading, which assimilated behavioural reading patterns. Further analysis revealed that SR was largely associated with the interaction between oral vocabulary exposure and oral vocabulary size. When the amount of exposure was limited, SR significantly increased with vocabulary sizes but decreased then with vocabulary sizes. Overall, the simulation results provide the first computational evidence of the direct link between oral vocabulary knowledge and the degree of SR as a source of individual differences in reading.

How you read affects what you gain: Individual differences in the functional organization of the reading system predict intervention gains in children with reading disabilities.

There is now considerable evidence regarding the types of interventions that are effective at remediating reading disabilities on average. It is generally unclear, however, what predicts the magnitude of individual-level change following a given intervention. We examine new predictors of intervention gains that are theoretically grounded in computational models of reading and focus on individual differences in the functional organization of the reading system. Specifically, we estimate the extent to which children with reading disabilities (n=118 3rd-4th graders) rely on two sources of information during an oral word reading task - print-speech correspondences and semantic imageability - before and after a phonologically-weighted intervention. We show that children who relied more on print-speech regularities and less on imageability pre-intervention had better intervention gains. In parallel, children who over the course of the intervention exhibited greater increases in their reliance on print-speech correspondences and greater decreases in their reliance on imageability had better intervention outcomes. Importantly, these two factors were differentially related to specific reading task outcomes, with greater reliance on print-speech correspondences associated with pseudoword naming, while (lesser) reliance on imageability related to word reading and comprehension. We discuss the implications of these findings for theoretical models of reading acquisition and educational practice.

Insights from the study of Arabic reading

AbstractCurrent reading models were largely designed to explain findings from experiments of the reading of English and other European languages (Reichle, 2020, Computational models of reading: A handbook). Recent evidence from studies of other languages and writing systems (e.g., Chinese) has demonstrated the need to critically evaluate the assumptions of these models, and whether they are sufficient to explain the full range of findings related to reading, as required, for example, to understand the universal and specific cognitive principles that support reading. In this article, we review the recent behavioural and cognitive‐neuroscience research on the reading of Arabic, a world language that until recently has received scant attention despite the fact that its writing system poses fundamental challenges for current models of reading. We also highlight the points of convergence and difference between what has been learned about the reading of Arabic and the reading of another, more widely studied Semitic language, Hebrew. We then discuss the theoretical implications of these findings for existing models of reading.

The Genetics of Reading and Language.

Recounts how our collaboration with Nick Martin was shaped over two decades, leading to the first studies of predictions from the 'Dual Route Cascaded' computational model of reading in twins, and extending into the molecular work, first linkage, fine mapping of genes identified in pedigree studies, into now the genomewide association study era and the first polygenic risk scores for reading and their potential in early clarifying causality and validating interventions, as well as for future global collaborations in improving these predictors and identifying causal variants. We highlight Nick's warm, future-focused optimism, support and inclusive approach without which none of this would have been possible. The circle of Nick asking, over half a century ago, 'What genes do you think make some kids get better grades?' has built a diverse scientific legacy involving thousands of papers and collaborations. The (heritable) traits of curiosity, boldness, warmth, interest in societally important questions, openness to new methods, ambition and collaborative skill to bring into being the infrastructure and samples needed for this research are rare, and we are grateful.

A computational model of reading across development: Effects of literacy onset on language processing

Abstract Cognitive development is shaped by interactions between cognitive architecture and environmental experiences of the growing brain. We examined the extent to which this interaction during development could be observed in language processing. We focused on age of acquisition (AoA) effects in reading, where early-learned words tend to be processed more quickly and accurately relative to later-learned words. We implemented a computational model including representations of print, sound and meaning of words, with training based on children’s gradual exposure to language. The model produced AoA effects in reading and lexical decision, replicating the larger effects of AoA when semantic representations are involved. Further, the model predicted that AoA would relate to differing use of the reading system, with words acquired before versus after literacy onset with distinctive accessing of meaning and sound representations. An analysis of behaviour from the English Lexicon project was consistent with the predictions: Words acquired before literacy are more likely to access meaning via sound, showing a suppressed AoA effect, whereas words acquired after literacy rely more on direct print to meaning mappings, showing an exaggerated AoA effect. The reading system reveals vestigial traces of acquisition reflected in differing use of word representations during reading.

Quantity and Diversity of Preliteracy Language Exposure Both Affect Literacy Development: Evidence from a Computational Model of Reading

ABSTRACTDiversity of vocabulary knowledge and quantity of language exposure prior to literacy are key predictors of reading development. However, diversity and quantity of exposure are difficult to distinguish in behavioural studies, and so the causal relations with literacy are not well known. We tested these relations by training a connectionist triangle model of reading that learned to map between semantic; phonological; and, later, orthographic forms of words. The model first learned to map between phonology and semantics, where we manipulated the quantity and diversity of this preliterate language experience. Then the model learned to read. Both diversity and quantity of exposure had unique effects on reading performance, with larger effects for written word comprehension than for reading fluency. The results further showed that quantity of preliteracy language exposure was beneficial only when this was to a varied vocabulary and could be an impediment when exposed to a limited vocabulary.

Orthographic and Phonological Representations in the Fusiform Cortex.

Mental and neural representations of words are at the core of understanding the cognitive and neural mechanisms of reading. Despite extensive studies, the nature of visual word representation remains highly controversial due to methodological limitations. In particular, it is unclear whether the fusiform cortex contains only abstract orthographic representation, or represents both lower and higher level orthography as well as phonology. Using representational similarity analysis, we integrated behavioral ratings, computational models of reading and visual object recognition, and neuroimaging data to examine the nature of visual word representations in the fusiform cortex. Our results provided clear evidence that the middle and anterior fusiform represented both phonological and orthographic information. Whereas lower level orthographic information was represented at every stage of the ventral visual stream, abstract orthographic information was increasingly represented along the posterior-to-anterior axis. Furthermore, the left and right hemispheres were tuned to high- and low-frequency orthographic information, respectively. These results help to resolve the long-standing debates regarding the role of the fusiform in reading, and have significant implications for the development of psychological, neural, and computational theories of reading.

Exploring the relations between word frequency, language exposure, and bilingualism in a computational model of reading

Individuals show differences in the extent to which psycholinguistic variables predict their responses for lexical processing tasks. A key variable accounting for much variance in lexical processing is frequency, but the size of the frequency effect has been demonstrated to reduce as a consequence of the individual’s vocabulary size. Using a connectionist computational implementation of the triangle model on a large set of English words, where orthographic, phonological, and semantic representations interact during processing, we show that the model demonstrates a reduced frequency effect as a consequence of amount of exposure to the language, a variable that was also a cause of greater vocabulary size in the model. The model was also trained to learn a second language, Dutch, and replicated behavioural observations that increased proficiency in a second language resulted in reduced frequency effects for that language but increased frequency effects in the first language. The model provides a first step to demonstrating causal relations between psycholinguistic variables in a model of individual differences in lexical processing, and the effect of bilingualism on interacting variables within the language processing system.

Computational Models of Reading: A Primer

AbstractThis article briefly reviews four computational models of reading and the types of empirical findings that they are designed to explain: (1) theDual‐Route modelof word identification (Coltheart et al., 2001); (2) theSimple‐Recurrent Network modelof sentence processing (Elman, 1990); (3) theConstruction–Integration modelof discourse representation (Kintsch, 1988); and (4) theE‐Z Reader modelof eye‐movement control in reading (Reichle et al., 2012). These particular models are reviewed because they provide comprehensive accounts of a large number of empirical findings in each of their respective domains, and because they have advanced our understanding of the cognitive processes involved in reading by motivating new empirical studies. Future models of reading will build upon the success of their predecessors by integrating models from two or more of the aforementioned domains, thereby providing a means to examine the compatibility of their theoretical assumptions and more comprehensive accounts of the cognitive processes involved in reading.

What can we learn from learning models about sensitivity to letter-order in visual word recognition?

Recent research on the effects of letter transposition in Indo-European Languages has shown that readers are surprisingly tolerant of these manipulations in a range of tasks. This evidence has motivated the development of new computational models of reading that regard flexibility in positional coding to be a core and universal principle of the reading process. Here we argue that such approach does not capture cross-linguistic differences in transposed-letter effects, nor do they explain them. To address this issue, we investigated how a simple domain-general connectionist architecture performs in tasks such as letter-transposition and letter substitution when it had learned to process words in the context of different linguistic environments. The results show that in spite of of the neurobiological noise involved in registering letter-position in all languages, flexibility and inflexibility in coding letter order is also shaped by the statistical orthographic properties of words in a language, such as the relative prevalence of anagrams. Our learning model also generated novel predictions for targeted empirical research, demonstrating a clear advantage of learning models for studying visual word recognition.

Neural correlates reveal sub-lexical orthography and phonology during reading aloud: a review.

The sub-lexical conversion of graphemes-to-phonemes (GPC) during reading has been investigated extensively with behavioral measures, as well as event-related potentials (ERPs). Most research utilizes silent reading (e.g., lexical decision task) for which phonological activation is not a necessity. However, recent research employed reading aloud to capture sub-lexical GPC. The masked priming paradigm avoids strategic processing and is therefore well suitable for capturing sub-lexical processing instead of lexical effects. By employing ERPs, the on-line time course of sub-lexical GPC can be observed before the overt response. ERPs have revealed that besides phonological activation, as revealed by behavioral studies, there is also early orthographic activation. This review describes studies in one’s native language, in one’s second language, and in a cross-language situation. We discuss the implications the ERP results have on different (computational) models. First, the ERP results show that computational models should assume an early locus of the GPC. Second, cross-language studies reveal that the phonological representations from both languages of a bilingual become activated automatically and the phonology belonging to the context is selected rapidly. Therefore, it is important to extend the scope of computational models of reading (aloud) to multiple lexicons.

Can evolution provide perfectly optimal solutions for a universal model of reading?

Frost has given us good reason to question the universality of existing computational models of reading. Yet, he has not provided arguments showing that all languages share fundamental and invariant reading universals. His goal of outlining the blueprint principles for a universal model of reading is premature. Further, it is questionable whether natural evolution can provide the optimal solutions that Frost invokes.

Old and new ways to study characteristics of reading disability: The case of the nonword-reading deficit

Theoretical and computational models of reading have traditionally been informed by specific characteristics of disabled readers. One of the most frequently studied marker effects of developmental dyslexia is the nonword-reading deficit. Disabled readers are generally believed to show a specific problem in reading nonwords. This study presents a survey of frequently cited methods used to examine this effect by controlling general reading ability in various ways. An extensive analysis, however, shows that the majority of these methods (grade equivalents scores, the reading-level match design, and interactions in a chronological-age match design) actually fail to account for confounding variables such as age and general slowing, potentially affecting the conclusions reached. To alleviate this problem, an alternative method is presented: i.e. state trace analysis. Applying this method in a sample of Dutch disabled and typical readers, the results revealed an absence of a nonword-reading deficit in the disabled readers. Furthermore, after controlling for their decoding ability, disabled readers showed inferior word reading performance, which strongly suggests that the fundamental problem of disabled readers does not relate to the reading of nonwords but concerns their (dis)ability to acquire orthographic (word-specific) knowledge. Further, predictions for disabled readers in an inconsistent orthography like English are formulated. Finally, based on a review of neurobiological studies, implications for theories of reading disability are discussed.

Lexical Decision in Children: Sublexical Processing or Lexical Search?

Length effects in the lexical decision latencies of children might indicate that children rely on sublexical processing and essentially approach the task as a naming task. We examined this possibility by means of the effects of neighbourhood size and articulatory suppression on lexical decision performance. Sixty-six beginning and 62 advanced readers performed a lexical decision task in a standard, articulatory suppression, or tapping condition. We found length effects on words and nonwords in the children's lexical decisions. However, the effects of neighbourhood size were similar to those reported for adult lexical decisions, rather than the effects previously found in children's naming. In addition, no effect was found of articulatory suppression. Both findings suggest that, despite clear length effects, children do not adopt a naming task approach but, like adults, base lexical decisions mainly on a lexical search. These results pose a challenge for several computational models of reading.

Top-down modulation of ventral occipito-temporal responses during visual word recognition

Although interactivity is considered a fundamental principle of cognitive (and computational) models of reading, it has received far less attention in neural models of reading that instead focus on serial stages of feed-forward processing from visual input to orthographic processing to accessing the corresponding phonological and semantic information. In particular, the left ventral occipito-temporal (vOT) cortex is proposed to be the first stage where visual word recognition occurs prior to accessing nonvisual information such as semantics and phonology. We used functional magnetic resonance imaging (fMRI) to investigate whether there is evidence that activation in vOT is influenced top-down by the interaction of visual and nonvisual properties of the stimuli during visual word recognition tasks. Participants performed two different types of lexical decision tasks that focused on either visual or nonvisual properties of the word or word-like stimuli. The design allowed us to investigate how vOT activation during visual word recognition was influenced by a task change to the same stimuli and by a stimulus change during the same task. We found both stimulus- and task-driven modulation of vOT activation that can only be explained by top-down processing of nonvisual aspects of the task and stimuli. Our results are consistent with the hypothesis that vOT acts as an interface linking visual form with nonvisual processing in both bottom up and top down directions. Such interactive processing at the neural level is in agreement with cognitive and computational models of reading but challenges some of the assumptions made by current neuro-anatomical models of reading.

List context effects in reading Italian words and nonwords: Can the word frequency effect be eliminated?

It has been suggested that when words and nonwords are presented mixed in the same blocks in transparent scripts the word frequency effect can be eliminated. This is either because in reading mixed blocks the lexical route is deemphasised in favour of the nonlexical route (route deemphasis) or because a homogenisation of reaction times occurs for words and nonwords by adjustment of a time criterion for articulation (time criterion). In five experiments using different list manipulations and experimental designs, we assessed the effects of frequency and length in reading aloud words and nonwords in the Italian transparent orthography. The effect of word frequency remained constant irrespective of context manipulation and nonword characteristics. As reading nonwords may be easier in Italian than in English, control over reading processing may be unnecessary for Italian readers. The results are discussed with respect to current computational models of reading.

Can the dual-route cascaded computational model of reading offer a valid account of the masked onset priming effect?

The masked onset priming effect (MOPE) refers to the empirical finding that target naming is faster when the target (SIB) is preceded by a briefly presented masked prime that starts with the same letter/phoneme (suf) than when it does not (mof; Kinoshita, 2000, Experiment 1). The dual-route cascaded (DRC) computational model of reading (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001) has offered an explanation for how the MOPE might occur in humans. However, there has been some empirical discrepancy regarding whether for nonword items the effect is limited to the first-letter/phoneme overlap between primes and targets or whether orthographic/phonological priming effects occur beyond the first letter/phoneme. Experiment 1 tested these two possibilities. The human results, which were successfully simulated by the DRC model, showed priming beyond the first letter/phoneme. Nevertheless, two recent versions of the DRC model made different predictions regarding the nature of these priming effects. Experiment 2 examined whether it is facilitatory, inhibitory, or both, in order to adjudicate between the two versions of the model. The human results showed that primes exert both facilitatory and inhibitory effects.

Computational modelling of the masked onset priming effect in reading aloud

The masked onset priming effect (MOPE) was first defined by Forster and Davis (1991) as the finding that human naming latencies are faster when a target word (e.g., BREAK) is preceded by a briefly presented masked prime word that shares its initial sound with the target (e.g., belly) compared to when it does not (e.g., merry) or when it rhymes with it (e.g., stake). The present paper presents a review of empirical findings on the MOPE in the English language and their simulations by the DRC computational model of reading (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001), which offers an explicit account of how the effect might occur in humans. A new version of DRC, called DRC 1.2, which differs from the previous downloadable DRC version mainly in the way its nonlexical route operates, has been recently developed. The performance of DRC 1.2 on simulating the MOPE is evaluated in this paper.