Visual word recognition

Listen to audio

Reading begins in the retina, whose structure imposes severe constraints on visual word recognition. Only the central part of the retina, the fovea, has sufficient resolution for visual identification of small letters. This is why our gaze constantly shifts during reading. Classic experiments by Rayner and colleagues (Rayner, 1998) have determined the amount of information acquired during gaze fixation. By masking letters at a certain distance from the fovea, they showed that preservation of around 4 letters to the left and 15 to the right of the fixation point leads to normal reading speed. In reality, only the identity of around 3-4 letters to the left and 7-8 letters to the right of fixation seems to be extracted. This visual span is therefore very narrow. Overall, these results suggest that reading proceeds essentially by sequential acquisition of information during each saccade, an acquisition which takes place virtually word by word, even if some parafoveal information seems to be extracted concerning the following word. These experiments may be seen as partial justification for the concentration of psycholinguistic research on isolated word processing - even though in-depth research into sentence and text processing remains all too rare.

In expert readers, behavioral studies have revealed several key features of visual word recognition:

No effect of word length. All other things being equal (frequency, regularity, etc.), word reading time is virtually independent of the number of letters in the word, at least when the word length does not exceed 7 or 8 letters. This lack of length effect indicates that the entire chain of letters is processed in parallel. This property is the result of expertise: in children, a strong influence of length exists, but gradually disappears with learning. It reappears in adults when words are degraded or when reading pseudowords.
Exploiting visual redundancy. The classic work of Miller, Bruner and Postman (1954), extended in particular by Reicher (1969) and Rumelhart & McClelland (1982), indicates that the visual system of the expert reader has internalized and exploits the distributional statistics of letters. The identification of a letter in a string of characters is facilitated in direct proportion to the mutual information provided by neighboring letters. This effect suggests an implicit or explicit representation of letter sets and their statistical distribution.
Representation of units larger than the letter. Behavioral experiments by Rey, Ziegler and Jacobs (2000) indicate that complex graphemes - groups of letters corresponding to a phoneme, such as "ch" or "oi" - are treated as units by the visual system. Bigrams, syllables and morphemes are also coded.
Effect of frequency of use of the word in the language. More frequent words are recognized more quickly. Reading speed also varies with subjective familiarity and age of acquisition.
Word neighborhood effects. Words that differ by just one letter are called "orthographic neighbors" (e.g. "chat" and "char"; the word "drap" has no neighbors). The processing of a word is generally slowed down when it has one or more higher-frequency neighbors (lexical inhibition). The number of neighbors also plays a role: word processing often (but not always) speeds up with the number of neighbors, particularly in fast lexical decision making.
Consistency" effects of grapheme-phoneme conversion. This refers to the consistency with which a letter or group of letters is transcribed into phonemes ( grapho-phonological consistency), and vice versa ( phono-grapheme consistency). The recognition time for a chain of letters slows down when its links with pronunciation are ambiguous.

All these behavioral effects underline the fact that expert reading is associated with the development of a variety of levels of visual representation that interact during word recognition.

-->

The visual word form area

As early as 1892, Déjerine described the syndrome of pure alexia or alexia without agraphia, a selective inability to recognize written words with no impairment of language, writing or visual recognition of objects and faces. He associates it with a disconnection of visual projections to the angular gyrus, putative seat of a "visual word image center". His patient's lesion, however, mainly affects the left ventral occipitotemporal region. This location is replicated in many contemporary studies based on anatomical MRI. Computerized intersection of lesions suggests that the essential region for pure alexia is located in the left lateral occipitotemporal sulcus (Cohen et al., 2003).

Functional MRI of normal subjects now confirms the essential role of this region in visual handwriting recognition (Cohen & Dehaene, 2004). All good readers activate this region when presented with written words, whereas it does not activate in response to spoken words. It occupies a reproducible location with respect to responses to other visual categories (faces, objects, houses; work by Puce et al., 1996; Ishai et al., 2000; Hasson et al., 2002, 2003). Evoked potentials, magnetoencephalography and intracranial recordings confirm its selective activation around 170-200 ms after word presentation. A rare case of left occipitotemporal surgery with pre- and post-operative imaging has recently demonstrated the causal role of this activation in reading ability (Gaillard et al., 2006).

The function of this region could be to provide the linguistic regions of the temporal lobe with a compact visual code of the letter string, invariant for font, size and word position. In fact, cognitive neuroimaging has made it possible to specify the degree of abstraction of the left occipitotemporal region's responses to written words. This region is the first visual area to respond invariantly to words presented to the right or left of the visual field, a spatial invariance that requires information to be transferred through the corpus callosum. It also exhibits a subliminal priming effect independent of "case" (upper or lower case), suggesting that it is the first to respond invariantly to the identity of a string of characters regardless of the exact shape of its letters. It is not sensitive to the difference between words and pseudowords, but its activation varies with orthographic regularity: as the frequency of bigrams increases, so does activation (Binder, Medler, Westbury, Liebenthal, & Buchanan, 2006). Recent data suggest that it is not homogeneous, but has an anterior-posterior organization with an increasing degree of invariance and coding of units of increasing size, from isolated letters to bigrams, morphemes and words (Dehaene et al., 2004; Vinckier et al., 2007).