A multistream model of visual word recognition

Abstract

Since the beginning of experimental psychology, researchers have been interested in how humans recognize words during reading. For example, Cattell (1886) hypothesized that words were processed analytically, as component letters, whereas Pillsbury (1897) hypothesized that words were processed holistically, as whole units. Investigators have inferred from the results of many studies that both types of processing are involved (e.g., Rayner & Pollatsek, 1989). However, few contemporary models of word recognition are based on the physiology of the visual system.The experiments in this article were motivated by a hybrid model that is an integration of a physiologybased model of the visual system with an informationprocessing model; we call this the multistream model. This model contains a lexicon, the activation of which plays a central role in decisions about whether letter strings are words. However, it differs from most other models of visual word recognition in two important ways: The processing streams of the multistream model analyze spatial frequency information, not visual features; and the different processing channels of the multistream model are responsive to different aspects of the stimulus. That is, letters are not formed from features, and words are not necessarily formed from component letters. Instead, words and letters are formed separately through the use of spatial frequency analysis of different "grain sizes" (i.e., words are recognized from lower spatial frequencies, and letters are recognized from higher spatial frequencies). We elaborate on each of these differences and then describe four experiments in which we tested predictions of the multistream model. Spatial Frequency Filters Versus FeaturesAlthough many researchers have assumed that perceptual objects (e.g., words) are perceived as collec- University of Akron, Akron, OhioFour experiments are reported that test a multistream model of visual word recognition, which associates letter-level and word-level processing channels with three known visual processing streams isolated in macaque monkeys: the magno-dominated (MD) stream, the interblob-dominated (ID) stream, and the blob-dominated (BD) stream (Van Essen & Anderson, 1995). We show that mixing the color of adjacent letters of words does not result in facilitation of response times or error rates when the spatial-frequency pattern of a whole word is familiar. However, facilitation does occur when the spatial-frequency pattern of a whole word is not familiar. This pattern of results is not due to different luminance levels across the different-colored stimuli and the background because isoluminant displays were used. Also, the mixed-case, mixed-hue facilitation occurred when different display distances were used (Experiments 2 and 3), so this suggests that image normalization can adjust independently of object size differences. Finally, we show that this effect persists in both spaced and unspaced conditions (Experiment 4)-suggesting that inappropriate letter grouping by hue cannot acc...

Fast and slow readers and the effectiveness of the spatial frequency content of text: Evidence from reading times and eye movements.

“…This resonates with the view that lower spatial frequencies reach cortical areas before higher spatial frequencies (e.g., Allen et al, 2009;Jordan et al, 2012Jordan et al, , 2014Patching & Jordan 2005a,b), and so, compared to slow readers, fast readers may be especially able to use lower spatial frequencies to help gain lexical access more rapidly. Indeed, this ability for fast readers could provide an effective bottom-up basis for influences of predictability that many argue are also part of fast reading (e.g., Hawelka et al, 2015;Hersch & Andrews, 2012;Long et al, 1994;Murray & Burke, 2003).…”

Distinct role of spatial frequency in dissociative reading of ideograms and phonograms: An fMRI study

“…Kana words can be processed automatically on the basis of their word-level orthographic representation (Nakagawa and Sukigara, 2000). Therefore, if we apply the model described by Allen et al (2009) to our results, Kanji may be dominantly processed by the letter (or character)-level processing associated with the P-pathway, while word-level processing in the M-pathway may contribute more to Kana processing.…”

“…HSF and LSF information are processed distinctly in the P-and M-pathways, respectively (Tobimatsu and Celesia, 2006). Allen et al (2009) suggested a possible link between reading and the parallel visual pathways. When reading, letter-level processing is distributed to the P-pathway, which responds to HSF information, while word-level processing occurs in the M-pathway, which is sensitive to LSF information.…”

Effects of increased letter spacing on word identification and eye guidance during reading

“…There is a growing body of research that indicates an important role in word recognition for different bands of spatial frequency information, from coarse-scale (low-frequency) information that describes the overall extent of a word to more fine-scale (high-frequency) information necessary to specify individual letters and features (Allen & Madden, 1990;Allen, Smith, Lien, Kaut, & Canfield, 2009;Boden & Giaschi, 2000;Dakin & Morgan, 1999;Jordan, 1990Jordan, , 1995Jordan & Bevan, 1996;Jordan & de Bruijn, 1993;Leat & Munger, 1994;Legge et al, 1985;Patching & Jordan, 2005a, 2005b.The disruption to normal spatial frequency content caused by increased letter spacing may mean that coarse-scale information corresponding to the overall extent of words is particularly distorted, making words in these conditions afford poor quality information about word extent and location, which may also interfere with saccadic programming. Further research is needed to determine the nature of the spatial frequency information that is afforded by words and the role that it plays in facilitating eye guidance.…”