Computational Models Of Reading Research Articles

Stressing what is important: Orthographic cues and lexical stress assignment

Computational models of reading have typically focused on monosyllabic words. However extending those models to polysyllabic word reading can uncover critical points of distinction between competing models. We present a connectionist model of stress assignment that learned to map orthography onto stress position for English disyllabic words. We compared the performance of the connectionist model to Rastle and Coltheart's [(2000).] rule-based model of stress assignment for words and nonwords. The connectionist model performed well on predicting human performance in reading nonwords that both contained and did not contain affixes, whereas the Rastle and Coltheart model performed well only on nonwords with affixes. The connectionist model provides an important first step to simulating all aspects of polysyllabic word reading, and indicates that a probabilistic approach to stress assignment can reflect human performance on stress assignment for both words and nonwords.

Hemispheric dissociation and dyslexia in a computational model of reading

There are several causal explanations for dyslexia, drawing on distinctions between dyslexics and control groups at genetic, biological, or cognitive levels of description. However, few theories explicitly bridge these different levels of description. In this paper, we review a long-standing theory that some dyslexics’ reading impairments are due to impairments in hemispheric transfer. We test this theory in a computational model of reading, implementing anatomical features of the visual system. We demonstrate that, when callosal transfer is impaired, the model reads nonwords as well as an unimpaired model, but reads exception words poorly: a pattern of behaviour similar to surface dyslexia. This computational modelling provides a causal link between brain-based theories of dyslexia to cognitive-level theories that refer specifically to phonological impairments within the reading system.

A functional magnetic resonance imaging study of the tradeoff between semantics and phonology in reading aloud

Using functional magnetic resonance imaging, we explored the role of semantics in mediating orthographic-to-phonological processing in reading aloud, focusing on the interaction of imageability with spelling-to-sound consistency for low-frequency words. Behaviorally, high-imageable words attenuate the standard latency and accuracy disadvantage for low-frequency inconsistent words relative to their consistent counterparts. Neurobiologically, high-imageable words reduced consistency-related activation in the inferior frontal gyrus but increased posterior activation in the angular and middle temporal gyri, representing a possible neural signature of the tradeoff between semantics and phonology in reading aloud. We discuss implications for neurobiological models of reading in terms of understanding the interplay among areas associated with component processes and suggest that the results constitute an important step toward integrating neurobiological and computational models of reading.

Impaired oral reading in two atypical dyslexics: A comparison with a computational lexical-analogy model

A computational model of reading was developed based upon the notion that the structural relationship between orthography and phonology is of greater importance than the dimension of semantics for the reading aloud of single words. Degradation of this model successfully simulated the reading performance of two patients with atypical acquired dyslexia. The first patient CAV, studied and described in the literature by Warrington in 1981, presented with unusual concrete word dyslexia, i.e., he had a category-specific reading deficit for concrete words compared with abstract words. The second patient BG is a phonological dyslexic who, although displaying a strong concreteness effect (concrete words read better than abstract words), was able to read functors individually perfectly well, a pattern that is rarely if ever seen. The computational model was used to generate a set of words for which it was predicted that BG would show no concreteness effect. The results of BG’s reading of these words were consistent with this prediction, thereby providing greater support for the validity of the model. It is concluded that a computational approach that attempts not only to reproduce the core symptoms of the major varieties of acquired dyslexia but also to simulate clinical data from specific patients has much to contribute to the understanding of cognitive deficits and to the design of effective rehabilitation strategies.

What do letter migration errors reveal about letter position coding in visual word recognition?

Dividing attention across multiple words occasionally results in misidentifications whereby letters apparently migrate between words. Previous studies have found that letter migrations preserve within-word letter position, which has been interpreted as support for position-specific letter coding. To investigate this issue, the authors used word pairs like STEP and SOAP, in which a letter in 1 word could migrate to an adjacent letter in another word to form an illusory word (STOP). Three experiments show that both same-position and adjacent-position letter migrations can occur, as well as migrations that cross 2 letter positions. These results argue against position-specific letter coding schemes used in many computational models of reading, and they provide support for coding schemes based on relative rather than absolute letter position.

Age of Acquisition Effects in Word Reading and Other Tasks

Recent studies have suggested that age of acquisition (AoA) has an impact on skilled reading independent of factors such as frequency. This result raises questions about previous studies in which AoA was not controlled and about current theories in which it is not addressed. Analyses of the materials used in previous studies suggest that the observed AoA effects may have been due to other factors. We also found little evidence for an AoA effect in computational models of reading that used words that exhibit normal spelling-sound regularities. An AoA effect was observed, however, in a model in which early and late learned words did not overlap in terms of orthography or phonology. The results suggest that with other correlated properties of stimuli controled, AoA effects occur when what is learned about early patterns does not carry over to later ones. This condition is not characteristic of learning spelling-sound mappings but may be relevant to tasks such as learning the names for objects.

Cross-language computational investigation of the length effect in reading aloud.

The authors examined whether 2 computational models of reading, the dual-route cascaded model (M. Coltheart, K. Rastle, C. Perry, R. Langdon, & J. C. Ziegler, 2001) and the connectionist 2-layer model (M. Zorzi, G. Houghton, & B. Butterworth, 1998), were able to predict the pattern that the length effect found in reading aloud is larger in German than in English (J. C. Ziegler, C. Perry, A. M. Jacobs, & M. Braun, 2001). The results showed that the dual-route cascaded model, which uses a serial mechanism for assembling phonology, successfully predicted this cross-language difference. In contrast, the connectionist model of Zorzi et al. (1998) predicted the opposite: a larger length effect in English than in German. Both the success of one model and the failure of the other highlight fundamental differences between 2 major classes of computational models.

Is covert attention really unnecessary?

We are largely in agreement with the Findlay & Walker model. However, they appear to dismiss the role of covert spatial attention in tasks in which people are free to move their eyes. We argue that an account of the facts about the perceptual span in reading requires a window of attention not centered around the fovea. Moreover, a computational model of reading that we (Reichle et al. 1998) developed gives a good account of eye movement control in reading and would be unable to do so without relying heavily on covert attention.