Visual Word Recognition

The goal of research on visual word recognition is to understand the kinds of capacities that underlie the rapid and almost effortless comprehension of words in READING, how these capacities are acquired, and the impairments that occur developmentally and following brain injury (DYSLEXIA). Visual word recognition has also provided a domain in which to explore broader theoretical issues concerning knowledge representation, learning, perception, and memory; for example, it played a significant role in the development of both modular (MODULARITY OF MIND) and connectionist (COGNITIVE MODELING, CONNECTIONIST) approaches to cognition.

Studies of EYE MOVEMENTS in reading indicate that most words are fixated once for durations ranging from 50 to 250 ms. Short function words are sometimes skipped and longer words may be fixated more than once. Word recognition speeds vary depending on reading skill, the type of text, and how carefully it is being read; large increases in reading speed can only be achieved with significant loss of comprehension, as in skimming. The main bottleneck is perceptual: the perceptual span is approximately four letters to the left of fixation and fifteen to the right when reading from left-to-right (it is asymmetrical to the left in reading languages such as Hebrew). Letter identities can be determined only over a smaller range, approximately five to six letters; further from fixation only letter shape and length are perceived (Pollatsek and Rayner 1990).

A long-standing issue for reading researchers and educators is whether words are recognized on a visual basis or by first computing a phonological representation (see PHONOLOGY). Using visual information might seem to be more efficient because it involves a direct mapping from spelling to meaning; using phonology (translating from orthography to phonology to meaning) involves an extra step. However, a compelling body of research suggests that skilled readers compute phonological information as part of the recognition process (e.g., Van Orden 1987). Studies of learning to read have also highlighted the important role of phonological information (Wagner and Torgesen 1987). The quality of prereading children's knowledge of the structure of spoken language is a good predictor of later reading skill; children who are good readers are better able to translate from spelling to sound; and many dyslexic persons exhibit minor deviations in their representation of spoken language that disrupt reading acquisition (e.g., Bradley and Bryant 1983). Despite this evidence, reading education in most English-speaking countries attempts to discourage children from using phonological information on the mistaken view that it discourages reading efficiency. There is also strong evidence that learning to read an orthography has a reciprocal impact on phonological representation (Morais et al. 1986).

One barrier to using phonology in reading English and many other writing systems would seem to be the quasi- regular (Seidenberg and McClelland 1989) character of orthographic-phonological correspondences: most words can be pronounced "by rule" (e.g., gave, mint) but there are many exceptions that deviate from the rules in differing degrees (e.g., have, pint). This observation led to the development of "dual-route" models in which there are separate mechanisms for reading rule-governed words and exceptions (Coltheart 1978). Connectionist models provide an alternative approach in which a single network consisting of distributed representations of spelling, sound, and meaning is used for all words. Such networks can encode both "rule-governed" forms and "exceptions," while capturing the overlap between them. Whereas the older models involved parallel, independent visual and phonological recognition pathways, connectionist models permit continuous pooling of information from both sources until a word's meaning has been computed.

Research on WRITING SYSTEMS organized along different principles (see papers in Frost and Katz 1992) suggests that there may be more commonalities in how they are read than the differences among them might otherwise suggest. One major difference among writing systems is in how transparently they represent phonological information. For example, whereas the pronunciations of orthographic patterns in Finnish and Serbo-Croatian are highly predictable, many English words are irregularly pronounced, and the nonalphabetic Chinese writing system provides only partial cues to pronunciation. These differences have often led to suggestions that one or another writing system is optimal for learning to read. Writing systems exhibit trade-offs among other design features, however, that tend to level the playing field (Seidenberg 1992). For example, English has many irregularly pronounced words but they tend to be very short and to cluster among the highest-frequency words in the language; hence they are likely to be easy to learn and process. Serbo-Croatian is more transparent at the level of letters and phonemes but there are few monosyllabic words and there is also a complex system governing syllabic stress. The pronunciations of words in Hebrew can be reliably predicted from their spellings except that the vowels are normally omitted. Studies of reading acquisition in different writing systems do not suggest large differences in the average rate at which children learn to read.

A major unresolved issue concerns the role of subword units such as syllables and morphemes (see MORPHOLOGY) in word recognition. Does reading a word such as farmer involve parsing it into the morphemes [farm] + [er] or merely using orthographic and phonological information? Although several studies have provided evidence for lexical decomposition, the extent to which it occurs in reading is not known. Any decomposition scheme runs up against what to do with cases like corner or display, which appear to be morphologically complex but are not. Connectionist models have also begun to provide an alternative account in which morphological structure reflects an emergent, interlevel representation mediating correlations among orthography, phonology, SEMANTICS, and aspects of grammar.

Other research has addressed how readers determine the meanings of words and integrate them with the contexts in which they occur. Words in texts tend not to be very predictable, which makes using context to guess them an inefficient strategy. The computation of a word's meaning is nonetheless constrained by context, as is clearly the case for ambiguous words such as rose and plane but also relatively unambiguous words such as cat. For example, in a sentence about petting, the word cat may activate the feature <fur>; in a context about getting scratched, cat will activate <claws> (Merrill, Sperber, and MacCauley 1981).

Impairments in word recognition are characteristic of developmental dyslexia. Dyslexia is often associated with phonological impairments that interfere with learning the relationship between the written and spoken forms of language (Liberman and Shankweiler 1985). In other cases, dyslexic persons have normal phonology but are developmentally delayed: they read like much younger children. This delay may reflect impoverished experience or other deficits in perception or learning (Manis et al. 1996).

Dyslexia also occurs as a consequence of neuropathologic discorders such as Alzheimer's disease or herpes encephalitis. Three major subtypes have been identified: phonological dyslexia, in which the main impairment is in pronouncing novel letter strings such as nust; surface dyslexia, in which the main impairment is in reading irregularly pronounced words such as pint; and deep dyslexia, in which the patient makes semantic paraphasias such as pronouncing sympathy "orchestra" (Shallice 1988). Current research focuses on using computational models of normal word recognition to explain how these patterns of impairment could arise (see MODELING NEUROPSYCHOLOGICAL DEFICITS). For example, connectionist models of normal performance can be "lesioned" to create the reading impairments seen in several types of patients (Plaut et al. 1996). A growing body of neuroimaging evidence is beginning to clarify how the representations and processes specified in these models are realized in the brain.

See also

-- Mark S. Seidenberg


Bradley, L., and P. E. Bryant. (1983). Categorizing sounds and learning to read -- a causal connection. Nature 301:419-421.

Coltheart, M. (1978). Lexical access in simple reading tasks. In G. Underwood, Ed., Strategies of Information Processing. London: Academic Press.

Coltheart, M., B. Curtis, P. Atkins, and M. Haller. (1993). Models of reading aloud: Dual-route and parallel distributed processing approaches. Psychological Review 100:589-608.

Frost, R., and L. Katz, Eds. (1992). Orthography, Phonology, Morphology, and Meaning. Amsterdam: North-Holland.

Liberman, I. Y., and D. Shankweiler. (1985). Phonology and the problem of learning to read and write. Remedial and Special Education 6:8-17.

Manis, F., M. S. Seidenberg, L. Doi, C. McBride-Chang, and A. Petersen. (1996). On the basis of two subtypes of developmental dyslexia. Cognition 58:157-195.

Merrill, E. C., R. D. Sperber, and C. McCauley. (1981). Differences in semantic encoding as a function of reading comprehension skill. Memory and Cognition 9:618-624.

Morais, J., P. Bertelson, L. Cary, and J. Alegria. (1986). Literacy training and speech segmentation. Cognition 24:45-64.

Plaut, D. C., J. L. McClelland, M. S. Seidenberg, and K. E. Patterson. (1996). Understanding normal and impaired word reading: Computational principles in quasiregular domains. Psychological Review 103:56-115.

Pollatsek, S., and K. Rayner. (1990). Eye movements in reading: A tutorial review. In D. Balota, F. D'arcais, and K. Rayner, Eds., Comprehension Processes in Reading. Hillsdale, NJ: Erlbaum.

Seidenberg, M. S. (1992). Beyond orthographic depth: Equitable division of labor. In R. Frost and L. Katz, Eds., Orthography, Phonology, Morphology, and Meaning. Springer.

Seidenberg, M. S., and J. L. McClelland. (1989). A distributed, developmental model of visual word recognition and naming. Psychological Review 96:523-568.

Shallice, T. (1988). From Neuropsychology to Mental Structure. Cambridge: Cambridge University Press.

Van Orden, G. C. (1987). A ROWS is a ROSE: Spelling, sound, and reading. Memory and Cognition 15:181-198.

Wagner, R. K., and J. K. Torgesen. (1987). The nature of phonological processing and its causal role in the acquisition of reading skills. Psychological Bulletin 101:192-212.

Further Readings

Coltheart, M., B. Curtis, P. Atkins, and M. Haller. (1993). Models of reading aloud: Dual-route and parallel distributed processing approaches. Psychological Review 100:589-608.

Gough, P., L. Ehri, and R. Treiman, Eds. (1992). Reading Acquisition. Hillsdale, NJ: Erlbaum.

Harm, M., and M. S. Seidenberg. (Forthcoming). Phonology, reading acquisition, and dyslexia: Insights from connectionist models. Psychological Review.

Plaut, D. C., and T. Shallice. (1993). Deep dyslexia: A case study of connectionist neuropsychology. Cognitive Neuropsychology 10:377-500.

Seidenberg, M. S. (1995). Visual word recognition: An overview. In J. L. Miller, and P. D. Eimas, Eds., Speech, Language, and Communication. San Diego: Academic Press.

Van Orden, G. C., B. F. Pennington, and G. O. Stone. (1990). Word identification in reading and the promise of a subsymbolic psycholinguistics. Psychological Review 97:488-522.