The neural manifestation of the word concreteness effect: An electrical neuroimaging study

Abstract

Previous studies have provided controversial evidence about the way in which words with different degrees of concreteness are represented in the brain. The aim of the present study was to investigate whether the processing of abstract vs. concrete words differently affected the timing and topographical distribution of ERP components. Participants were engaged in a lexical decision task (word/non-word discrimination) while EEG was recorded from 128 scalp sites. Reaction times (RTs) to words were faster than RTs to pseudowords. Words were discriminated from pseudowords since larger N2 responses to words than to pseudowords were observed over the left occipito-temporal areas at 300 ms post-stimulus. Concrete words and abstract words were discriminated as early as 350 ms post-stimulus, with larger responses to concrete than to abstract words over the mesial occipital regions. Concreteness-related ERP differences were also observed in the amplitudes of the later anterior LP component (between 370 and 570 ms), with larger responses to abstract words than to concrete words. The LORETA source localization technique was also applied to identify the intra-cranial generators of surface potentials reflecting lexico-semantic processing. Results showed that words (both abstract and concrete) were associated with a stronger activation of the left fusiform gyrus and the left temporal cortex, as compared to pseudowords. Concrete word processing was associated with a stronger activation of the left extrastriate visual areas (namely BA 18 and BA 19) as compared to abstract word processing. By revealing the neural markers of the concreteness effect, our study contributes to the understanding of the neurogenesis of verbal semantic knowledge impairments and the incidence of these impairments in clinical populations.

Introduction

The process by which semantic information is stored and retrieved during word recognition is a central theme in cognitive psychology research. Neuropsychological, hemodynamic, and electrophysiological studies have provided evidence of dissociations in the way by which words belonging to different classes are represented in the brain (e.g. Damasio and Tranel, 1993, Dehaene, 1995, Pulvermüller, 1999). Category-specific differences in brain processes have been proposed for nouns vs. verbs (Luzzatti et al., 2006, Preissl et al., 1995, Pulvermüller et al., 1999), objects vs. actions (see the recent review by Vigliocco, Vinson, Druks, Barber, & Cappa, 2010), proper vs. common nouns (Proverbio et al., 2001, Proverbio et al., 2009), mass vs. count nouns (Mondini et al., 2008), living vs. nonliving entities (Martin, 2007, Perani et al., 1999b), or abstract vs. concrete nouns (Binder et al., 2005, Papagno et al., 2009b).

Concrete words differ from abstract words in that they refer to predominantly perceptually defined concepts, while the meaning of abstract words arises from their association with other usually abstract words. Concrete word advantage (concreteness effect) has been reported in numerous behavioral studies (for a review, see Paivio, 1991). The concreteness effect has also been demonstrated in a number of neuropsychological studies (for example, Franklin et al., 1995, Warrington, 1981), although clinical cases showing a reversal of the concreteness effect have also been reported (e.g. Bonner et al., 2009, Papagno et al., 2009a). In the study of Bonner et al. (2009), poor performance with concrete relative to abstract verbs was correlated with cortical thinning of the right anterior temporal lobe in semantic dementia (SD) patients, thus suggesting that this region may contribute to the storing and processing of visual semantic features. The authors accordingly advanced the hypothesis that degraded visual feature knowledge may be the underlying reason for the impaired comprehension of concrete words in patients with SD. An alternative influential model of SD, however, argues that conceptual representation arises from the interaction of modality-specific association regions (spokes) with a central, modality-invariant hub. Impairment in function of the hub accounts for the selective, multimodal semantic impairment seen in SD (Rogers et al., 2004). The inferior aspects of the anterior temporal lobes, the region that is typically atrophied in SD, is said to be the site of the semantic hub. A problem with this account, however, is that it provides no obvious explanation for the reversal of concreteness effect that is typically found in SD, since there is no reason to suppose that concrete concepts should depend on the hub to a greater extent than abstract concepts.

Different hypotheses about the neural underpinnings of concrete vs. abstract word processing have been advanced. Behavioral studies using visual hemifield stimulation (e.g. Day, 1977, Levine and Banich, 1982, Shibahara and Lucero-Wagoner, 2002), and clinical studies (e.g. Villardita, Grioli, & Quattropani, 1988) seem to suggest that processing of abstract words is confined to the left hemisphere, whereas concrete word processing also engages right-hemispheric brain areas.

Neither neuroimaging nor electrophysiological studies have provided definitive evidence on this matter. Perani, Cappa, et al. (1999) used PET to measure regional cerebral activity during visual lexical decision task with concrete and abstract nouns. They found that abstract word processing, when directly compared to concrete word processing, was associated with the selective activation of the right temporal pole, the right amygdala, and the bilateral inferior frontal cortex. No brain areas were more active in response to concrete words than abstract words. Kiehl et al. (1999), using the same paradigm in conjunction with fMRI, found that a direct comparison between abstract and concrete word processing yielded a greater activation in the right anterior temporal cortex in response to abstract words than in response to concrete words. They interpreted the results as support for a right hemisphere involvement in the processing of abstract words. Conversely, two subsequent fMRI studies (Binder et al., 2005, Fiebach and Friederici, 2004) using a lexical decision task found no clear-cut hemispheric differences between activations to concrete and abstract words. Fiebach and Friederici (2004) showed a stronger activation of the pars triangularis of the left inferior frontal lobe in response to abstract words, and a stronger activation of the basal regions of the left temporal lobe in response to concrete words. Binder et al. (2005) found that relative to abstract words, concrete words activated bilateral regions including the angular gyrus and the dorsal prefrontal cortex. Relative to concrete words, abstract words activated left inferior frontal regions previously linked with phonological and verbal working memory processes. In a recent meta-analysis, Binder, Desai, Graves, and Conant (2009) reported thirteen functional neuroimaging studies that showed areas of stronger activation for concrete compared with abstract words, with overlap in bilateral angular gyrus, left mid-fusiform gyrus, left dorsomedial prefrontal cortex, and left posterior cingulate cortex. When compared to concrete words, abstract words seemed to be associated with higher activation in the left areas that are usually involved in semantic processing such as the left inferior frontal gyrus (mainly pars orbitalis) and the left anterior superior temporal sulcus (STS). In summary, neuroimaging studies provide no clear evidence about the neural underpinnings of concrete vs. abstract word processing. Some studies suggest a right hemispheric involvement in abstract word processing (Kiehl et al., 1999, Perani et al., 1999a), thus being in direct contradiction with the hypothesis suggested by behavioral studies that processing of abstract words involves only the left hemisphere, whereas concrete words would engage right-hemispheric brain areas. Other neuroimaging studies (Binder et al., 2005, Fiebach and Friederici, 2004) suggest a strong involvement of the left hemisphere both in response to abstract and concrete word processing. Results of neuroimaging studies differ even when the same type of task is used (i.e., visual lexical decision). This may depend, at least in part, on the stimuli features, such as the degree of imageability: although concrete words are mostly imageable, abstract words present a high degree of variability within this dimension (Martín-Loeches, Hinojosa, Fernández-Frías, & Rubia, 2001).

A number of ERP studies using different tasks have found that concrete words elicit more negative ERPs than abstract words in the time window of N400 component (300–500 ms extending to 800 ms post-stimulus). For example, Kounios and Holcomb (1994) investigated the concreteness effect using a priming paradigm with a lexical decision and concrete/abstract classification task. In this study, concrete words elicited a more negative response than abstract words in the time window of N400 component, between 300 and 500 ms after stimulus onset. This effect was greater during the concrete/abstract classification task (that requires deeper semantic processing than the lexical decision task), and was greater over the right anterior sites. In a following study by the same research group (Holcomb, Kounios, Anderson, & West, 1999) participants read sentences in a congruency judgment task. Different amplitudes of the N400 to concrete and abstract final words were found for anomalous and neutral sentences, but not for congruent sentences. The authors proposed that both sentence context and imageability of the words may facilitate word processing. In another ERP study, West and Holcomb (2000) investigated these effects by manipulating task demands. Three groups of subjects performed a sentence verification task in which the final word of each sentence was either concrete or abstract. For each group, the truthfulness judgment required image generation or semantic decision or evaluation of surface characteristics of the words. Concrete and abstract words produced similar ERPs in the task that required attending to the surface characteristics of the words. Concreteness effects were, however, detected in the image generation and semantic decision task. In both tasks concrete words elicited more negative ERPs than abstract words between 300 and 550 ms (N400). ERPs were also more negative for concrete than abstract words between 550 and 800 ms. This effect was more frontally distributed and was strongest in the imagery task. Another research group (Tsai et al., 2009) manipulated the word concreteness in tasks that required different depths of semantic processing, namely a lexical decision task and a semantic relatedness judgment task. The results replicated the concreteness effect described in previous studies, indicating that concrete nouns elicited larger N400 responses than abstract nouns. No effect of task demands was found. Larger negativities in response to concrete words than in response to abstract words were also replicated while participants were engaged in an incidental memory paradigm (Nittono, Suehiro, & Hori, 2002). These ERP components were more left-lateralized for low imagery words than for high imagery words. This led the authors to suggest that high imagery words activated more extensively a semantic network distributed across the left and right hemispheres than low imagery words. They also suggested that right hemisphere activation probably dealt with imagery-related information, which was not available for low imagery words. Significant effects of word concreteness on N400-like ERPs were also found in a study by Renoult, Brodeur, and Debruille (2010), in which 6 concrete and 6 abstract words were repeatedly presented in two concrete–abstract classification tasks, and in three studies making use of the lexical decision task (Kanske and Kotz, 2007, Tolentino and Tokowicz, 2009, Zhang et al., 2006). In summary, ERP results typically demonstrate that abstract and concrete words are differently processed in the brain. Nevertheless, the effects described have a broad distribution over the scalp, generally greater over anterior than posterior sites, so that no clear evidence is traceable about the cerebral regions involved or about the lateralization of the effects. Some authors have reported a slight right lateralization (Holcomb et al., 1999, Kounios and Holcomb, 1994) in favor of concrete words. Others have reported a left lateralization (Tolentino and Tokowicz, 2009, Zhang et al., 2006) or no clear lateralization (Kanske and Kotz, 2007, Nittono et al., 2002, Renoult et al., 2010, Tsai et al., 2009, West and Holcomb, 2000).

Fewer ERP studies have shown concreteness effects of earlier latency or in more identifiable scalp regions. For example, Wirth et al. (2008) manipulated context (in terms of semantic association strength) and concreteness during passive reading of paired words. The authors found an enhanced early context effect in response to abstract paired words as compared to concrete words (in the latency of P1–N1) centered over the left inferior prefrontal cortex. More interestingly for the purposes of the present study, Martín-Loeches et al. (2001) using a visual lexical decision task found larger word Recognition Potential (RP) in response to concrete words than in response to abstract words, and by means of BESA they located the possible intracortical RP generator in the left occipito-temporal regions. RP is a negative-going wave that is sensitive to lexico-semantic content and thought to originate in the left visual extrastriate brain regions. In that study, both concrete and abstract words elicited an RP as compared to pseudowords. Moreover, concrete words elicited greater RP as compared to abstract words, thus leading the authors to conclude that this effect revealed a unitary semantic system that is specialized for highly imageable concrete stimuli and also subserves the processing of abstract stimuli.

In summary, the analysis of the literature shows how neuroimaging studies provide no clear evidence about the neural underpinnings of concrete vs. abstract word processing, even when the same type of task is used (visual lexical decision). This may depend, at least in part, on the stimuli features. This is the reason why in the present study, particular attention was paid to rule out the confounding effects of psycholinguistic variables such as word length, written frequency, subjective frequency of occurrence and to control for the concreteness and degree of imageability of the stimuli. ERP studies typically demonstrate that abstract and concrete words are differently processed in the brain, as concrete words elicit more negative ERPs than abstract words in the time window corresponding to N400 component. Nevertheless, the effects described have a broad distribution over the scalp, so that no clear evidence is traceable about the cerebral regions involved or about the lateralization of the effects.

In the present study, we aimed to shed further light on how the processing of abstract and concrete words may differently affect the timing and topographical distribution of ERPs. The use of a dense-array EEG system offered the opportunity to test the existence of topographically defined ERP components. Source localization by means of LORETA allows drawing inferences about the intra-cranial generators of surface potentials reflecting orthographic and lexico-semantic word processing and reveal the concreteness effect, and it offers an opportunity to match electrophysiological results with neuroimaging results.

The main purpose was to investigate the spatial and temporal dynamics of activation of the brain regions that are known to be part of the lexico-semantic processing system. We considered the findings of the only ERP study (to our knowledge), which evidenced a concreteness effect over the left occipito-temporal regions (Martín-Loeches et al., 2001), in our study design. As previously mentioned, these authors argued in favor of a unitary semantic system that is specialized for highly imageable concrete stimuli and also subserves the processing of abstract stimuli. In light of available neuroimaging literature about the neural underpinnings of lexico-semantic processing (for a review see Binder et al., 2009) we were not convinced by this interpretation, and therefore aimed to test the hypothesis that (1) a word's lexical and semantic status (traceable, respectively, in words vs. pseudowords and abstract vs. concrete words comparisons) would differently affect the latency, amplitude, and scalp topographic distribution of ERP components. We also predicted that (2) concrete words are associated with higher activation in brain regions that are usually involved in visual object processing; (3) abstract words are associated (or not) with higher activation in the left temporal and prefrontal areas that are usually involved in verbal semantic processing; and (4) concrete words are associated (or not) with selective activation of right hemispheric regions.