Reading normal and degraded words: Contribution of the dorsal and ventral visual pathways
Introduction
Whenever human adults read words, a cortical network is activated which involves left occipito-temporal cortex as a central node. This region, which has been termed the “Visual Word Form Area”, is thought to house what psycholinguists and neuropsychologists have termed the “visual word form system” (Warrington and Shallice, 1980), a hierarchy of neural mechanisms for invariant visual word recognition (Cohen et al., 2000, Dehaene et al., 2005, McCandliss et al., 2003). The purpose of the present research is to further clarify the limits of this ventral reading system, and the conditions under which it needs to be supplemented by dorsal serial (“letter-by-letter”) reading mechanisms.
We have proposed that the ability to read words stems from the more general ability of the ventral visual system to identify complex multipart objects. Encoding the abstract identity and the relative position of letters is loosely similar to recognizing the drawing of a car on the basis of its component parts and of their spatial relationships. Following this intuition, we recently proposed an attempt at modelling the neural processes of visual word perception, derived from general principles governing the organization of the primate visual system (the Local Combination Detector or LCD model, Dehaene et al., 2005). Neurophysiological observations in the macaque ventral visual regions have led to construe this system as a hierarchy of converging neural detectors with progressively larger receptor fields, tuned to increasingly complex objects (Booth and Rolls, 1998, Riesenhuber and Poggio, 1999, Rolls, 2000, Ullman, 2007). Encoding of complex shapes then results from the convergence of elementary contour detectors (Brincat and Connor, 2004, Tsunoda et al., 2001). The LCD model thus proposes that words are encoded through a posterior to anterior hierarchy of neurons tuned to increasingly larger and more complex word fragments, such as visual features, single letters, bigrams, quadrigrams, and possibly whole words.
This system reaches its optimal level of expertise only after years of practice. Children initially decipher words slowly and letter by letter, as indexed by a large effect of word length on reading latencies (Aghababian and Nazir, 2000). Over years of practice, speed increases and the length effect eventually disappears, at least for words of about 3 to 6 letters. Within the LCD framework, this adult pattern of performance is thought to reflect the parallel encoding of letters through the hierarchy of converging detectors. Through perceptual learning mechanisms, the ventral visual system progressively becomes attuned to the regularities of the writing system at all of the above hierarchical levels.
The goal of the present paper is to provide an empirical test of part of the LCD model using fMRI in adults. We propose to progressively degrade words in three different ways and to examine at which threshold of degradation the ventral reading hierarchy ceases to support parallel visual recognition. Perceptual learning mechanisms can only ensure that the visual system of the expert reader is attuned to the perception of normal print: horizontally aligned words presented in the foveal region in a usual font. When words depart from this standard format, reading becomes more difficult, as revealed by lower reading speed, and by the emergence of an effect of word length on reading latencies: response time becomes a linear function of the number of letters. For instance, reading low-contrast words (Legge et al., 1997), words printed in mIxEd case (Lavidor, 2002), vertically presented words (Bub and Lewine, 1988), or words displayed in the left visual field (Lavidor and Ellis, 2002), induces a word length effect.
We suggest that this length effect reflects a failure of parallel letter processing in the ventral pathway, and indicates the deployment of a serial mechanism based on a coordination of dorsal and ventral processes. Those strategies would generally involve the deployment of serial attention to letters or groups of letters within the target string, and therefore induce an effect of length. In terms of functional anatomy, such serial reading would involve both parietal structures driving spatial–attentional processes, and a modulation by this top-down attention of ventral occipito-temporal structures coding for word fragments such as single letters. Dorsal attention orienting (Gitelman et al., 1999, Husain and Rorden, 2003, Kanwisher and Wojciulik, 2000b, Mesulam, 1999) and ventral top-down modulation (Chawla et al., 1999, Kastner et al., 1998, Somers et al., 1999) have been previously observed in many visual tasks, but their contribution to reading remains understudied (Cohen et al., 2003, Henry et al., 2005, Mayall et al., 2001).
Understanding the mechanisms engaged in reading degraded words is thus instrumental in addressing two related questions. First, what are the limits of the perceptual expertise of the ventral visual system for reading? Second, what are the mechanisms of the interplay of the ventral and dorsal pathways during word reading?
In a first attempt to address those issues, we recently studied a simultanagnosic patient with bilateral parietal atrophy (Vinckier et al., 2006). The patient was excellent at reading normally printed foveal words, but she was severely impaired at reading words which were mirror reversed, or rotated by angles larger than 50°, or whose letters were separated by at least two blank spaces, or words displayed in her left hemifield. We proposed that above those critical thresholds, i.e. when stimulus degradation exceeds the perceptual tolerance of the ventral system, reading normally requires the intervention of the parietal lobes to pilot the attention-driven exploration of stimuli. This neuropsychological study thus provided a first estimate of the limits of the perceptual expertise of the ventral system, and pointed to the involvement of parietal cortex in reading words exceeding such limits. Hall et al. (2001) found converging evidence of the deleterious influence of letter spacing and case alternation on reading performance in a patient with parietal (but also occipitotemporal) lesions.
Here we address those issues using behavioral measures and fMRI during word reading by normal subjects. We transformed stimulus words according to three modes of degradation: rotating words by 0° to 90° (rotation mode), separating letters with 0 to 3 blank spaces (spacing mode), or displacing words from 100% in the left hemifield to 100% in the right hemifield (position mode). Within each degradation mode, 5 successive levels of degradation were contrasted, from optimally displayed words (i.e. horizontal foveal words with contiguous letters) to heavily degraded words, thus allowing for a fine-grained assessment of a putative threshold effect on reading difficulty and brain activation (Fig. 1).
Some previous studies have addressed the influence of word degradation on brain activations, mostly showing modulation of the posterior ventral cortex by physical parameters such as visual contrast (Mechelli et al., 2000), visual noise (Helenius et al., 1999, Jernigan et al., 1998), stimulus rate and duration (Price and Friston, 1997, Price et al., 1996), and of the right parietal cortex by case alternation (Mayall et al., 2001). The LCD model, however, goes one step further by generating precise expectations concerning the degradation thresholds above which reading performance should be expected to deteriorate. Concerning rotation, behavioral measures suggest that for angles larger than 45–60° readers abandon the normal fast and parallel reading pattern (Koriat and Norman, 1985, Koriat and Norman, 1989, Lavidor et al., 2001a). Accordingly, Vinckier et al.’s (2006) patient’s reading performance dropped above a threshold angle larger than 50°. A tentative parallel might be drawn with IT neurons in macaques, which show an invariant response for object rotations up to about 45° (Logothetis and Pauls, 1995). There are thus converging indications that the ventral visual system cannot maintain rotation-invariant processing for angles above a limit of about 40–60°.
Concerning letter spacing, normal subjects begin to show a word length effect when reading words with letters separated by two spaces or more (Vinckier et al., in preparation). This value matches a prediction of the LCD framework, based on the principle that letter detector neurons with a local receptive field converge onto open bigram detector neurons (Dehaene et al., 2005). Given the increase of receptive fields in IT cortex by a factor of about 2.5 from one neural level to the next, the LCD model proposes that bigram detectors integrate letter information over a range of 2–3 letter positions, and should therefore fail to detect their preferred letter pairs whenever the component letters are separated by blank spaces of at least two letter widths. Indeed, in the above simultanagnosic patient, we observed a drop in reading performance whenever 2 or more spaces were inserted between letters (Vinckier et al., 2006). A similar threshold of about 2 spaces was expected in the present study.
Concerning the impact of word position in the visual field, it has been shown in normal subjects that while there is no length effect for words displayed in the right visual field, at least close to the fovea, such an effect emerges whenever words are displayed in the left visual field (for a review see Ellis, 2004). When words extend across central fixation, only their left part induces a length effect (Lavidor et al., 2001b). Within the LCD model, at least two factors may contribute to this effect: the left-hemispheric lateralization of the visual word form area, which implies that right visual field letters enjoy more direct and more numerous projections to invariant letter and word recognition processes than their left counterparts (Dehaene et al., 2005); and the left-to-right direction of reading which, combined with the leftward bias of the preferred eye landing, may induce greater perceptual learning in the right visual field (Nazir et al., 2004). We therefore expected that optimal fast and parallel reading would be disrupted as soon as the major part of words would be displayed in the left visual field.
For all three degradation modes, our behavioral and brain imaging predictions were similar. First, above a critical threshold of stimulus degradation, we expected the onset both of a slowing down of reading, and of a word length effect on response times, revealing the loss of parallel word identification and a switch to letter-by-letter reading. Second, we predicted the sudden onset of attention-related posterior parietal activations at the same critical threshold. Third, above this threshold we expected the parietal activation to drive an amplification of the posterior part of the ventral visual word form area, at a location devoted to letter-level coding. In addition to those specific predictions, we intended to explore whether some regions would be differently modulated by the three modes of degradation, and to investigate the neural bases of the word length effect.
Section snippets
Subjects
Twelve right-handed native French speakers with normal or corrected to normal vision participated in this experiment (4 men and 8 women, mean age 22 years). All subjects gave written informed consent and were naive about the aims of the experiment.
Stimuli
We selected 3 sets of 210 4-, 5- and 6-letter words, matched for frequency (p = 0.83). All were high-frequency nouns which did not refer to animals (frequency: 5–125 per million; mean: 29 per million) (http://www.lexique.org, New et al., 2004). To
Behavioral results
Overall subjects made 1.7% errors and had a mean correct RT of 833 ms. For each subject, we computed the median RT for each cell of the crossed design Mode × Degradation level × Length, and those values were entered in an ANOVA with Subjects as random factor (Fig. 2). Error rates, which will not be discussed further, followed the same pattern as RTs, although only the main effects of Degradation level and Mode reached significance (Table 1).
Basic reading network
Within each fMRI run, one fifth of trials consisted of normal presentation of words (i.e. horizontal, foveal, and with normally spaced letters). Examination of the basic contrast of normal words versus the baseline resting trials (voxelwise threshold p < 0.001, threshold for cluster extent p < 0.05 corrected) replicated the usual network of activation observed during word reading (Fig. 3, top row). Activation was observed in bilateral occipital cortex (posterior to Montreal Neurological Institute
Word degradation and reading threshold
Our first prediction was that whenever word degradation does not allow the Visual Word Form system to encode words in a fast and parallel manner, reading performance should deteriorate, as indexed primarily by slower reading latencies. Moreover in such circumstances readers were expected to engage in effortful serial reading strategies, as revealed by the emergence of a word length effect. Behavioral data gathered during MRI scanning conformed to those expectations, as latencies increased
Conclusion
On the basis of these converging results from adults, patients, and children, we propose that the bilateral posterior parietal regions play an essential role whenever serial reading strategies are deployed, either during normal reading (of words by young readers, or of pseudo-words by expert adults), or during compensatory reading when parallel word recognition is impossible (due to word degradation or to a ventral visual brain lesion). Our results confirm that the ventral visual word form
Acknowledgments
This research was funded by the Institut National de la Santé et de la Recherche Médicale (INSERM), the Commissariat à l’Energie Atomique (CEA), and the Agence Nationale pour la Recherche (ANR, CORELEX project).
References (83)
- et al.
Developing normal reading skills: aspects of the visual processes underlying word recognition
J. Exp. Child Psychol.
(2000) - et al.
Parietal cortex and attention
Curr. Opin. Neurobiol.
(2004) - et al.
A selective effect of parietal damage on letter identification in mixed case words
Neuropsychologia
(2007) - et al.
Different modes of word recognition in the left and right visual fields
Brain Lang
(1988) - et al.
The pathophysiology of letter-by-letter reading
Neuropsychologia
(2004) - et al.
The neural code for written words: a proposal
Trends Cogn. Sci.
(2005) Length, formats, neighbours, hemispheres, and the processing of words presented laterally or at fixation
Brain Lang.
(2004)- et al.
Direct intracranial, fMRI and lesion evidence for the causal role of left inferotemporal cortex in reading
Neuron
(2006) From thought to action: the parietal cortex as a bridge between perception, action, and cognition
Neuron
(2007)- et al.
Brain activations during letter-by-letter reading: a follow-up study
Neuropsychologia
(2005)