Inhibition and facilitation in visual word recognition: Prefrontal contribution to the orthographic neighborhood size effect

Abstract

The recognition of words is a central component of language processing. A major role for visual word recognition has been attributed to the orthographic neighbors of a word, i.e., words that are orthographically similar to a target word. It has been demonstrated that the presence of orthographic neighbors facilitates the recognition of words, but hinders the rejection of nonwords. It is therefore assumed that representations of orthographic neighbors are at least partially activated during word recognition, and that they influence word recognition depending on the specific task context. In the present study, we used fMRI to examine the neural bases of the effect of orthographic neighborhood size on speeded lexical decisions to words and nonwords. Our results demonstrate lexicality × neighborhood size interactions in mid-dorsolateral and medial prefrontal cortex, suggesting the involvement of a domain-general, extra-lexical process for orthographic neighborhood effects on word and nonword processing. This result challenges computational models that offer purely lexical accounts of the orthographic neighborhood effect and suggests an important role for executive control functions during visual word recognition.

Introduction

The recognition of written words is a basic component of human language processes and of universal importance to modern everyday life. Recent investigations of the functional neuroanatomy of visual word recognition indicate that accessing the representations of words stored in the mental lexicon primarily involves the temporal cortex of the left hemisphere. More specifically, it was suggested that basal occipitotemporal brain regions are involved in early processing stages of word recognition, including the extraction of abstract word form representations from orthographic input (e.g., Cohen et al., 2000, Dehaene et al., 2002, Gaillard et al., 2006, McCandliss et al., 2003) (but see Price and Devlin, 2003, Price and Devlin, 2004) and the access to lexical representations (e.g., Beauregard et al., 1997, Bellgowan et al., 2003, Fiebach et al., 2002, Herbster et al., 1997, Howard et al., 1992, Kuo et al., 2003). Later stages of the word recognition process involve the activation of word meaning in lateral temporal cortex (e.g., Price, 2000).

Functional neuroimaging studies often focused on the localization of visual word recognition processes, and made few or no assumptions concerning the specific sequence of processing steps involved. Thus, it is often assumed implicitly that lexical representations of words (i.e., a hypothetical mental lexicon) are sequentially searched through until a target word is identified. This assumption, however, appears unrealistic given that behavioral and computational evidence indicates that several lexical entries sharing orthographic features are initially activated during visual word processing (e.g., Grainger and Jacobs, 1996, McClelland and Rumelhart, 1981). Orthographic similarity has been operationalized as the subset of words that share all but one letter with a target word. This set of words is referred to as the ‘orthographic neighborhood’ of that target word (Coltheart et al., 1977). It is thus assumed that the representations of orthographic neighbors of a word are partially activated during the initial stages of word recognition, before lexical competition and lateral inhibition finally result in the identification of the correct lexical entry (e.g., Grainger and Jacobs, 1996). In a task such as lexical decision, the presence of a large neighborhood of orthographically similar words facilitates target word identification but makes it more difficult to classify letter strings as nonwords (e.g., Andrews, 1989, Carreiras et al., 1997, Forster and Shen, 1996, Sears et al., 1995). Interestingly, the facilitatory effect of neighborhood size on word recognition is task-dependent. Facilitatory effects of neighborhood size are observed in the lexical decision task under speeded conditions but not under instructions stressing accuracy (e.g., Grainger and Jacobs, 1996). Also, large neighborhoods can have inhibitory effects on word recognition in other measures such as reading-related gaze durations (Pollatsek et al., 1999) and performance in semantic tasks (Forster and Shen, 1996).

It is assumed that the amount of summed activation present in the lexicon, which results from the simultaneous partial activation of orthographic neighbors of the target word, positively influences the identification of the target word entry in the mental lexicon (e.g., Andrews, 1989). For example, in their Multiple Read-Out Model (MROM), Grainger and Jacobs (1996) postulate two intra-lexical sources of information that can generate a ‘yes’ response in the lexical decision task: (1) the activation level of the representational unit of the presented word, as well as (2) the global activation in the orthographic lexicon induced by the partial activation of orthographic neighbors of the target word. If either of these activation levels exceeds a criterion, a ‘word’ response is generated. Thus, the larger the neighborhood of the target word, the more activation is globally present in the lexical system, and the faster can a positive response be generated. The inhibitory effect of large neighborhoods on nonword responses in the lexical decision task is accounted for as follows: because the global lexical activation induced by the orthographic neighbors of the nonword generates a strong tendency to respond “word”, this preponderant response tendency has to be inhibited and a “nonword” response has to be generated. The MROM assumes an extra-lexical time criterion: if the criterion is reached before a ‘word’ response has been initiated, the stimulus is categorized as nonword. In the case of nonwords with large orthographic neighborhoods, lateral inhibition between the intra-lexical word units usually cannot converge on a single word representation before the time criterion is reached, resulting in slowed response times for such nonwords. However, global lexical activation resulting from the large neighborhood may generate more incorrect ‘word’ responses than would be seen for nonwords with few neighbors.

To summarize, behavioral and computational investigations of the orthographic neighborhood size effect indicate that the similarity relations among orthographic word representations can automatically exert facilitatory or inhibitory influences on word and nonword processing. The neural mechanisms underlying these important component processes of word recognition, however, are not yet clear. Using event-related brain potentials (ERPs) and a lexical decision task, Holcomb et al. (2002) show that the size of the orthographic neighborhood influences postlexical semantic processing stages. In their study, orthographic neighborhood size modulated the amplitude of the N400 component, an ERP effect indexing semantic processing (Kutas and Federmeier, 2000). Holcomb and colleagues conclude that their results support the assumption of a greater level of global lexical–semantic activation for words and nonwords with large orthographic neighborhoods, because larger neighborhoods led to greater N400 amplitudes for both types of stimuli. This proposal is broadly consistent with the account of the orthographic neighborhood effects described above—although it is not unproblematic that the effect of a prelexical variable, orthographic similarity, is attributed to a postlexical semantic processing stage. It is noteworthy that in this ERP study, no electrophysiological correlate of the interaction of lexicality and neighborhood size was found even though this effect is evident in behavioral data. A second ERP study by Braun et al. (2006) supports the finding of a neighborhood size effect on the N400 for nonwords but not words in the lexical decision task. In this study, the MROM (Grainger and Jacobs, 1996) was used to compute the theoretical global lexical activation for each stimulus. Nonwords with greater predicted global lexical activation indeed produced greater N400 amplitudes than nonwords that generate less global lexical activation.

The orthographic neighborhood size effect has also been studied using functional magnetic resonance imaging (fMRI). Binder et al. (2003) report that words without neighbors elicited greater activation than words with many neighbors in the left middle and superior frontal gyri, the left angular gyrus, left middle temporal gyrus, and in midline structures. Nonwords without neighbors also activated midline structures to a greater degree than nonwords with many neighbors. No brain regions showed signatures of facilitatory effects of neighborhood size. These findings are not easily reconciled with the ERP data reported by Holcomb et al. (2002) and Braun et al. (2006). Furthermore, no brain regions showed the interaction effects expected for a lexical decision task, i.e., facilitatory effects for words and inhibitory effects for nonwords. Binder et al. (2003) discuss their findings – in particular the left angular and middle temporal activations for words without neighbors – as a likely result of the way the participants were instructed to perform the task. By deliberately emphasizing the importance of accuracy over speed, the instruction induced a more semantic processing strategy that relied on the careful semantic evaluation of the processed letter strings, as demonstrated by response time and accuracy differences between a pilot study and the fMRI experiment (Binder et al., 2003). The behavioral results of lexical decision performance in the fMRI study showed no facilitatory effects of neighborhood size on word recognition, while these were indeed seen in a behavioral pilot study in which speed was stressed over accuracy. The effect of orthorgraphic neighborhood size on auditory word recognition was also investigated using fMRI (Prabhakaran et al., 2006). However, the results of that study cannot easily be integrated with the work cited above, as neighborhood size has an inhibitory, rather than facilitatory, effect on auditory word recognition as compared to visual word recognition.

Given the inconsistencies among neurophysiological data and between neurophysiological data and conclusions drawn from behavioral studies of the orthographic neighborhood size effect, the present study set out to investigate the neural bases of orthographic neighborhood size effects under instruction conditions identical to those used in behavioral studies, i.e., under instruction conditions that typically induce facilitatory neighborhood size effects on word recognition and inhibitory effects on nonword rejection. fMRI was used to measure brain activation while participants read letter strings, half of which were words and half nonwords, and performed a lexical decision task on each item. The number of orthographic neighbors of the words and nonwords varied between zero and ten, and was matched between words and nonwords. Stimuli were selected such that subsets of items from each category could be used to form extreme groups, thereby allowing for statistical analysis in a 2 × 2 factorial design with 40 stimuli per condition [i.e., lexicality (words vs. nonwords) × neighborhood size (small: 0–2 neighbors vs. large: 5–10 neighbors)]. In addition, by including words and nonwords with neighborhood sizes falling between the extreme groups, the fMRI data could be analyzed using a parametric approach.

We were particularly interested in identifying brain areas that evince an activation pattern analogous to the lexicality × neighborhood size interaction seen in behavioral data, i.e., with a facilitatory effect of orthographic neighborhood size for words and an inhibitory neighborhood size effect for nonwords. A brain region sensitive to such an interaction effect should show greatest activation for nonwords with a large neighborhood, least activation for words with many neighbors, and intermediate activation levels for the other two experimental conditions. We expected three possible alternative results, defined by reference to the anatomical location where such an interaction effect may be expressed. (1) If such a lexicality × neighborhood size interaction at the neural level was expressed in extrastriate, higher level visual areas, this would be suggestive of a prelexical basis of the orthographic neighborhood size effect (e.g., Cohen et al., 2002). (2) Alternatively, if a lexicality × neighborhood size interaction in brain activation were seen in lateral temporal regions associated with semantic processing, this result would lend support to the interpretation of Holcomb et al. (2002) and Binder et al. (2003) of a semantic nature of orthographic neighborhood size effects. (3) Finally, if prefrontal cortical areas showed this kind of lexicality × neighborhood size interaction, one would have to assume an involvement of domain-general executive control mechanisms in the effect of orthographic neighborhood size on word recognition. To determine the specificity of observed interaction effects for the lexical decision task, a supplementary fMRI experiment was performed during which participants passively read words and nonwords with large vs. small orthographic neighborhoods.