Computational Models Of Word Recognition Research Articles

Neural correlates of affective contributions to lexical decisions in children and adults

The goal of the present study was to investigate whether 6–9-year old children and adults show similar neural responses to affective words. An event-related neuroimaging paradigm was used in which both age cohorts performed the same auditory lexical decision task (LDT). The results show similarities in (auditory) lexico-semantic network activation as well as in areas associated with affective information. In both age cohorts’ activations were stronger for positive than for negative words, thus exhibiting a positivity superiority effect. Children showed less activation in areas associated with affective information in response to all three valence categories than adults. Our results are discussed in the light of computational models of word recognition, and previous findings of affective contributions to LDT in adults.

A model-guided dissociation between subcortical and cortical contributions to word recognition

Neurocognitive studies of visual word recognition have provided information about brain activity correlated with orthographic processing. Some of these studies related the orthographic neighborhood density of letter strings to the amount of hypothetical global lexical activity (GLA) in the brain as simulated by computational models of word recognition. To further investigate this issue, we used GLA of words and nonwords from the multiple read-out model of visual word recognition (MROM) and related this activity to neural correlates of orthographic processing in the brain by using functional magnetic resonance imaging (fMRI). Words and nonwords elicited linear effects in the cortex with increasing BOLD responses for decreasing values of GLA. In addition, words showed increasing linear BOLD responses for increasing GLA values in subcortical regions comprising the hippocampus, globus pallidus and caudate nucleus. We propose that these regions are involved in the matching of orthographic input onto representations in long-term memory. The results speak to a potential involvement of the basal ganglia in visual word recognition with globus pallidus and caudate nucleus activity potentially reflecting maintenance of orthographic input in working memory supporting the matching of the input onto stored representations by selection of appropriate lexical candidates and the inhibition of orthographically similar but non-matching candidates.

The need for a universal computational model of bilingual word recognition and word translation

Dijkstra, Wahl, Buytenhuijs, van Halem, Al-jibouri, de Korte, and Rekké (2018) present in their keynote article a promising computational model of word recognition and word production in monolinguals and bilinguals, called Multilink. We agree with the authors that the model is a “basis for the development of a more general computational model of word retrieval” (Dijkstra et al., 2018). However, it is also important that such a model is universal.

"To Name or Not to Name: That is the Question": The Role of Response Inhibition in Reading.

Reading is a complex process that includes the integration of information about letters (graphemes) and sounds (phonemes). In many circumstances, such as noisy environments, response inhibition is an additional factor that plays a marked role in successful oral reading. Response inhibition can take the form of task relevant inhibition (i.e., foils in a go/no-go task) and task irrelevant inhibition (i.e., distractor information). Here we investigated task relevant inhibition by having participants (N = 30) take part in two tasks: go/no-go naming with nonwords foils (GNG-NW) and go/no-go naming with pseudohomophones foils (GNG-PH). Also, we investigated the addition of task irrelevant inhibition by having participants (N = 28) take part in two tasks: GNG-NW + information masking and GNG-PH + information masking. We provide evidence that during a task relevant inhibition task, sub-word sound level information can be successfully inhibited, as evidenced by comparable response times for regular words and exception words, provided the foils do not contain familiar sound-based information (GNG-NW). In contrast, regular words were read aloud faster than exception words in a GNG-PH task, indicating that sub-word level interference occurs when the foils contain familiar sound-based information. The addition of task irrelevant inhibition (i.e., information masking at the phoneme level), served to increase response time overall, but did not impact the pattern of response times between regular words and exception words. Together these findings provide useful information regarding the role of response inhibition in word recognition and may be useful in computational models of word recognition and future work may benefit from accounting for the effects outlined in this paper.

Drifting through Basic Subprocesses of Reading: A Hierarchical Diffusion Model Analysis of Age Effects on Visual Word Recognition.

Reading is one of the most popular leisure activities and it is routinely performed by most individuals even in old age. Successful reading enables older people to master and actively participate in everyday life and maintain functional independence. Yet, reading comprises a multitude of subprocesses and it is undoubtedly one of the most complex accomplishments of the human brain. Not surprisingly, findings of age-related effects on word recognition and reading have been partly contradictory and are often confined to only one of four central reading subprocesses, i.e., sublexical, orthographic, phonological and lexico-semantic processing. The aim of the present study was therefore to systematically investigate the impact of age on each of these subprocesses. A total of 1,807 participants (young, N = 384; old, N = 1,423) performed four decision tasks specifically designed to tap one of the subprocesses. To account for the behavioral heterogeneity in older adults, this subsample was split into high and low performing readers. Data were analyzed using a hierarchical diffusion modeling approach, which provides more information than standard response time/accuracy analyses. Taking into account incorrect and correct response times, their distributions and accuracy data, hierarchical diffusion modeling allowed us to differentiate between age-related changes in decision threshold, non-decision time and the speed of information uptake. We observed longer non-decision times for older adults and a more conservative decision threshold. More importantly, high-performing older readers outperformed younger adults at the speed of information uptake in orthographic and lexico-semantic processing, whereas a general age-disadvantage was observed at the sublexical and phonological levels. Low-performing older readers were slowest in information uptake in all four subprocesses. Discussing these results in terms of computational models of word recognition, we propose age-related disadvantages for older readers to be caused by inefficiencies in temporal sampling and activation and/or inhibition processes.

No one way ticket from orthography to semantics in recognition memory: N400 and P200 effects of associations

Computational models of word recognition already successfully used associative spreading from orthographic to semantic levels to account for false memories. But can they also account for semantic effects on event-related potentials in a recognition memory task? To address this question, target words in the present study had either many or few semantic associates in the stimulus set. We found larger P200 amplitudes and smaller N400 amplitudes for old words in comparison to new words. Words with many semantic associates led to larger P200 amplitudes and a smaller N400 in comparison to words with a smaller number of semantic associations. We also obtained inverted response time and accuracy effects for old and new words: faster response times and fewer errors were found for old words that had many semantic associates, whereas new words with a large number of semantic associates produced slower response times and more errors. Both behavioral and electrophysiological results indicate that semantic associations between words can facilitate top-down driven lexical access and semantic integration in recognition memory. Our results support neurophysiologically plausible predictions of the Associative Read-Out Model, which suggests top-down connections from semantic to orthographic layers.

Lexical is as lexical does: computational approaches to lexical representation

In much of neuroimaging and neuropsychology, regions of the brain have been associated with ‘lexical representation’, with little consideration as to what this cognitive construct actually denotes. Within current computational models of word recognition, there are a number of different approaches to the representation of lexical knowledge. Structural lexical representations, found in original theories of word recognition, have been instantiated in modern localist models. However, such a representational scheme lacks neural plausibility in terms of economy and flexibility. Connectionist models have therefore adopted distributed representations of form and meaning. Semantic representations in connectionist models necessarily encode lexical knowledge. Yet when equipped with recurrent connections, connectionist models can also develop attractors for familiar forms that function as lexical representations. Current behavioural, neuropsychological and neuroimaging evidence shows a clear role for semantic information, but also suggests some modality- and task-specific lexical representations. A variety of connectionist architectures could implement these distributed functional representations, and further experimental and simulation work is required to discriminate between these alternatives. Future conceptualisations of lexical representations will therefore emerge from a synergy between modelling and neuroscience.

The subjective frequency of word n-grams

When asked to think about the subjective frequency of an n-gram (a group of n words), what properties of the n-gram influence the respondent? It has been recently shown that n-grams that occurred more frequently in a large corpus of English were read faster than n-grams that occurred less frequently (Arnon & Snider, 2010), an effect that is analogous to the frequency effects in word reading and lexical decision. The subjective frequency of words has also been extensively studied and linked to performance on linguistic tasks. We investigated the capacity of people to gauge the absolute and relative frequencies of n-grams. Subjective frequency ratings collected for 352 n-grams showed a strong correlation with corpus frequency, in particular for n-grams with the highest subjective frequency. These n-grams were then paired up and used in a relative frequency decision task (e.g. Is green hills more frequent than weekend trips?). Accuracy on this task was reliably above chance, and the trial-level accuracy was best predicted by a model that included the corpus frequencies of the whole n-grams. A computational model of word recognition (Baayen, Milin, Djurdjevic, Hendrix, & Marelli, 2011) was then used to attempt to simulate subjective frequency ratings, with limited success. Our results suggest that human n-gram frequency intuitions arise from the probabilistic information contained in n-grams.

How to say “no” to a nonword: A leaky competing accumulator model of lexical decision.

We describe a leaky competing accumulator (LCA) model of the lexical decision task that can be used as a response/decision module for any computational model of word recognition. The LCA model uses evidence for a word, operationalized as some measure of lexical activity, as input to the YES decision node. Input to the NO decision node is simply a constant value minus evidence for a word. In this way, evidence for a nonword is a function of time from stimulus onset (as in standard deadline models) modulated by lexical activity via the competitive dynamics of the LCA. We propose a simple mechanism for determining the value of this constant online during the first trials of a lexical decision experiment, such that the model can rapidly optimize speed and accuracy in discriminating words from nonwords. Further optimization is achieved via trial-by-trial adjustments in response criteria as a function of task demands and list context. We show that the LCA model can simulate mean response times and response distributions for correct and incorrect YES and NO decisions for a number of benchmark experiments that have been shown to be fatal for deadline models of lexical decision. Finally, using lexical activity calculated by a computational model of word recognition as input to the LCA decision module, we provide the first item-level simulation of both word and nonword responses in a large-scale database.

Modelling word recognition and reading aloud

Computational modelling has tremendously advanced our understanding of the processes involved in normal and impaired reading. The present Special Issue highlights some new directions in the field of word recognition and reading aloud. These new lines of research include the learning of orthographic and phonological representations in both supervised and unsupervised networks, the extension of existing models to multisyllabic word processing both in English and in other languages, such as Italian, French, and German, and the confrontation of these models with data from masked priming. Some of the contributors to the Special Issue also address hotly debated issues concerning the front-end of the reading process, the viability of Bayesian approaches to understanding masked and unmasked priming, as well as the longstanding debate about the role of rules versus statistics in language processing. Thus, the present Special Issue provides a critical analysis and synthesis of current computational models of reading and cutting edge research concerning the next generation of computational models of word recognition and reading aloud.

Modeling the use of durational information in human spoken-word recognition

Evidence that listeners, at least in a laboratory environment, use durational cues to help resolve temporarily ambiguous speech input has accumulated over the past decades. This paper introduces Fine-Tracker, a computational model of word recognition specifically designed for "tracking" fine-phonetic information in the acoustic speech signal and using it during word recognition. Two simulations were carried out using real speech as input to the model. The simulations showed that the Fine-Tracker, as has been found for humans, benefits from durational information during word recognition, and uses it to disambiguate the incoming speech signal. The availability of durational information allows the computational model to distinguish embedded words from their matrix words (first simulation), and to distinguish word final realizations of [s] from word initial realizations (second simulation). Fine-Tracker thus provides the first computational model of human word recognition that is able to extract durational information from the speech signal and to use it to differentiate words.

Is the orthographic/phonological onset a single unit in reading aloud?

Two main theories of visual word recognition have been developed regarding the way orthographic units in printed words map onto phonological units in spoken words. One theory suggests that a string of single letters or letter clusters corresponds to a string of phonemes (Coltheart, 1978; Venezky, 1970), while the other suggests that a string of single letters or letter clusters corresponds to coarser phonological units, for example, onsets and rimes (Treiman & Chafetz, 1987). These theoretical assumptions were critical for the development of coding schemes in prominent computational models of word recognition and reading aloud. In a reading-aloud study, we tested whether the human reading system represents the orthographic/phonological onset of printed words and nonwords as single units or as separate letters/phonemes. Our results, which favored a letter and not an onset-coding scheme, were successfully simulated by the dual-route cascaded (DRC) model (Coltheart, Rastle, Perry, Langdon, & Ziegler, 2001). A separate experiment was carried out to further adjudicate between 2 versions of the DRC model.

Fine‐phonetic variation in a computational model of word recognition

There is now considerable evidence from psycholinguistic and phonetic research that fine‐phonetic variation in the speech signal modulates human speech processing (HSP), and helps the listener segment a speech signal into syllables and words [e.g., Salverda et al., Cognition 90, 51‐89 (2003)]. This kind of information also appears to help the human perceptual system distinguish short words (like 'ham') from the longer words in which they are embedded (like 'hamster'). Salverda et al. showed that the lexical interpretation of an embedded sequence is related to its duration; a longer sequence tends to be interpreted as a monosyllabic word more often than a shorter one. Until recently, no computational models of HSP existed that are able to model this fine‐phonetic variation [Hawkins, J. of Phonetics 31, 373‐405 (2003)]. In this paper, we present Fine‐Tracker, a novel computational model of human word recognition that it is able to capture and use this fine‐grained acoustic‐phonetic variation during speech r...

The Bayesian reader: Explaining word recognition as an optimal Bayesian decision process.

This article presents a theory of visual word recognition that assumes that, in the tasks of word identification, lexical decision, and semantic categorization, human readers behave as optimal Bayesian decision makers. This leads to the development of a computational model of word recognition, the Bayesian reader. The Bayesian reader successfully simulates some of the most significant data on human reading. The model accounts for the nature of the function relating word frequency to reaction time and identification threshold, the effects of neighborhood density and its interaction with frequency, and the variation in the pattern of neighborhood density effects seen in different experimental tasks. Both the general behavior of the model and the way the model predicts different patterns of results in different tasks follow entirely from the assumption that human readers approximate optimal Bayesian decision makers.

Effects of syllable-frequency in lexical decision and naming: An eye-movement study

In three experiments we explored task-specific effects of syllable-frequency, following Perea and Carreiras' (1998) findings of a facilitative effect during naming and an inhibitory effect during lexical decision. In Experiment 1, an inhibitory effect of first syllable-frequency on articulation duration suggested a process-specific effect during naming: a facilitative effect on construction of phonological output, but an inhibitory effect on lexical access. Eye-movement recording was used in Experiment 2 to disentangle these two processes. An inhibitory effect on eye-movement parameters during both lexical decision and naming supported the assumption that higher first syllable-frequency inhibits lexical access. Implications of a general inhibitory effect of syllable-frequency for computational models of word recognition, specifically for the multiple-trace memory (MTM) model of polysyllabic word naming ( Ans, Carbonnel, & Valdois, 1998), are discussed.

Can CANISO activate CASINO? Transposed-letter similarity effects with nonadjacent letter positions

Nonwords created by transposing two adjacent letters (i.e., transposed-letter (TL) nonwords like jugde) are very effective at activating the lexical representation of their base words. This fact poses problems for most computational models of word recognition (e.g., the interactive-activation model and its extensions), which assume that exact letter positions are rapidly coded during the word recognition process. To examine the scope of TL similarity effects further, we asked whether TL similarity effects occur for nonwords created by exchanging two nonadjacent letters (e.g., caniso- CASINO) in three masked form priming experiments using the lexical decision task. The two nonadjacent transposed letters were consonants in Experiment 1 (e.g., caniso- CASINO), vowels in Experiment 2 ( anamil-ANIMAL) and both consonants and vowels in Experiment 3. Results showed that nonadjacent TL primes produce priming effects (in comparison to orthographic controls, e.g., caviro- CASINO), however, only when the transposed letters are consonants. In a final experiment we examined latencies for nonwords created by nonadjacent transpositions of consonants versus vowels in a lexical decision task. Both types of nonwords produced longer latencies than matched controls, with consonant TL nonwords being more difficult than vowel TL nonwords. The implications of these findings for models having “position-specific” coding schemes as well as for models proposing alternative coding schemes are discussed.

Replicating syllable frequency effects in Spanish in German: One more challenge to computational models of visual word recognition

Two experiments tested the role of syllable frequency in word recognition, recently suggested in Spanish, in another shallow orthography, German. Like in Spanish, word recognition performance was inhibited in a lexical decision and a perceptual identification task when the first syllable of a word was of high frequency. Given this replication of the inhibitory effect of syllable frequency in a second language, we discuss the issue whether and how computational models of word recognition would have to represent a word’s syllabic structure in order to accurately describe processing of polysyllabic words.

See Dick RAN: Rapid Naming and the Longitudinal Prediction of Reading Subskills in First and Second Graders

The rapid automatic naming (RAN) task has been widely used in studies of reading acquisition and found to be reliably related to reading achievement. This study addressed the basis for this observed relation, providing new empirical evidence and an analysis of the RAN task within a computational model of word recognition in reading (Seidenberg & McClelland, 1989). The contributions of RAN, verbal ability, and phonological awareness to the prediction of phonological and orthographic skills from the 1st to the 2nd grade were examined. Both RAN (digits and letters) and phoneme awareness accounted for independent variance in later reading scores, even when vocabulary and prior reading skill were entered 1st in the regression analysis. RAN was a stronger predictor than phoneme awareness for 3 tasks in which orthographic information is critical (orthographic choice, word-likeness judgment, and exception word pronunciation), whereas the opposite held true for nonword reading and paragraph comprehension. The reading model suggests that the RAN task accounts for distinct variance in reading when compared to phoneme awareness because RAN involves arbitrary associations between print and sound (e.g., a digit and its name), whereas phoneme awareness is more related to the learning of systematic spelling-sound correspondences. In addition to the arbitrariness factor, RAN contains a number of other components that overlap with reading and, hence, collectively make it a good predictor of reading skills. In this article, we illustrate how explicit computational models of reading can clarify relations among tasks commonly used in reading research.