The prevailing theory of semantic memory holds that conceptual memories are multimodal and partly represented in the sensory and motor systems through which they are acquired (Allport, 1985
; Martin, 2007
; Binder et al., 2009
; Kiefer and Pulvermüller, 2011
). Many functional activation findings support this theory, yet converging anatomical findings relating functional activation in healthy adults to cortical atrophy in individuals with impaired processing of sensory-motor concepts has been lacking. Here we tested this theory in individuals with lvPPA, a neurodegenerative condition associated with atrophy in superior temporal and inferior parietal cortices (Grossman, 2010
), including canonical auditory association cortex (Maeder et al., 2001
; Lewis et al., 2004
). We demonstrated for the first time that individuals with lvPPA have relative difficulty for words with strong sound associations. Participants with lvPPA were differentially impaired at recognizing Sound words relative to the performance of healthy controls and relative to their own performance on other word categories. Moreover, we provide the first evidence identifying difficulty with auditory-weighed words in a group with focal disease in auditory association cortex. These findings provide crucial evidence supporting the hypothesis that conceptual memories are represented in part in sensory-motor association cortices.
Structural MRI analyses in lvPPA revealed reduced gray matter density centered in posterior temporal and inferior parietal cortices. This included regions of auditory association cortex defined on the basis of an fMRI functional localizer administered to healthy adults. Furthermore, lvPPA participants had reduced gray matter density in a region of posterior temporal cortex that was activated by healthy participants for Sound words during a separate fMRI experiment. We directly examined this region and found that reduced gray matter density in lvPPA correlated with selective difficulty on Sound words. These findings demonstrate a differential impairment in modality-specific conceptual knowledge in lvPPA that is directly related to reduced gray matter density in modality-specific association cortex.
Though several studies have examined the neural representation of words with strong visual or motor associations (Hauk et al., 2004
; Bonner et al., 2009
; Desai et al., 2010
), we know of only one other study examining the neural representation of words with strong sound associations (Kiefer et al., 2008
). Using fMRI,Kiefer et al. (2008)
demonstrated that auditory association regions in superior and middle temporal gyri are activated when participants perform a word recognition task on words with strong sound associations. This is consistent with our findings relating performance on Sound words to auditory association cortex in the superior temporal lobe.
Previous investigations of semantic memory in individuals with stroke have emphasized dissociations across semantic categories such as animals, plants, tools, and so forth (Gainotti et al., 1995
; Gainotti, 2004
; Mahon and Caramazza, 2009
). The differential weighting of sensory and motor features may be important for some of these categories, but this information was not the basis for defining these categories. Here we examined knowledge of categories that were primarily defined by the weighting of their modality-specific features, matching the approach taken in recent functional neuroimaging investigations (Pulvermüller, 2005
; Martin, 2007
; Binder et al., 2009
) and directly examining the role of modality-specific features in conceptual memory.
Studies of stroke patients pointing to a role for sensory and motor regions in conceptual memory have largely examined controlled semantic retrieval processes, with tasks such as picture naming and category-sorting (Gainotti, 2004
) that differ from the simple lexical decision task we used to examine single word meaning. And though previous work has shown that apraxic patients may be impaired at recognizing the sounds of actions (Pazzaglia et al., 2008
), this work does not directly address the role of modality-specific representations in word meaning. Furthermore, the neuroimaging technique employed in these stroke studies uses a binary classification approach to label voxels as either lesioned or not lesioned. This differs from our gray matter density analysis, which can relate gradations of cortical tissue loss directly to behavior. Our study is the first to directly examine the sound features of concepts in a patient group with neurodegenerative disease affecting auditory association cortex, and it provides an informative and complementary approach to previous neuropsychological investigations of conceptual memory.
Our correlation analysis related atrophy to a behavioral measure specific for the impairment on Sound words relative to other words. Likewise, our auditory functional localizer was analyzed relative to a visual baseline. This minimized the confounds of associating Sound words with aspects of conceptual memory that may also contribute to Sight and Manipulation words, or that may underlie task performance, such as decision-making or lexical retrieval. This is important because some theories incorporate additional heteromodal components in semantic memory that are not specific to any sensory or motor modality (Koenig and Grossman, 2007
; Patterson et al., 2007
; Binder et al., 2009
; Bonner and Grossman, In Press
). From this perspective, severe semantic impairments may accompany damage to heteromodal regions, but damage to a single modality may produce a mild overall semantic impairment that is differentially worse for concepts with strong associations in that modality. This is consistent with the pattern of impairment we observed in the lvPPA participants.
This impairment was observed on a simple word recognition task. Although this task did not explicitly require semantic retrieval, we assume that conceptual representations are automatically activated during word recognition and, thus, likely play a role in the successful performance of this task. This is in line with characterizations of word recognition tasks by many other investigators (Binder et al. 2009
). Indeed, it was important to use a simple measure of conceptual knowledge so that we could examine the role of modality-specific representations without requiring participants to engage in mental imagery. This minimized potential confounds related to post-conceptual processing (Machery, 2007
; Mahon and Caramazza, 2008
) and task-specific effects of controlled semantic retrieval (Peelle et al., 2009
Although we have contrasted a multimodal account of conceptual representation with an abstract symbolic processing account, it is worth pointing out that even multimodal representations have abstract properties. Abstract symbolic accounts (Fodor, 1975
) argue that semantic representations have the same symbolic format regardless of their modality-specific associations and, thus, do not rely on the sensory-motor system. Alternatively, the sensory-motor account (Martin, 2007
) argues that concepts are represented as distributed networks of feature representations in sensory and motor association cortices. But these multimodal representations may also have important abstract properties. For one, the regions involved in perceiving a feature may not be identical to those involved in representing that feature in conceptual memory; rather, these regions may be anatomically adjacent (Chatterjee, 2010
). Furthermore, multimodal concepts may use information from a range of specific instances to form a prototypical representation. For example, not all apples look the same, but our concept for “apple” can be applied to most of them. Hence, multimodal representations must be compatible with abstract cognitive processing, and we cannot rule out the possibility that regions representing modality-specific semantic features are anatomically adjacent to regions involved in perceiving those features.
A related issue is that the region of superior temporal cortex associated with Sound word processing in both the behavioral and fMRI studies, may not be strictly unimodal. In fact, this area of cortex has also been associated with crossmodal integration of auditory and visual information (Beauchamp et al., 2004
). Our findings may be consistent with such an account, since a region that processes crossmodal auditory and visual information would likely be important for the representation of auditory-weighted concepts, which have both auditory and visual features.
Previous investigations of lvPPA have focused on difficulty with repetition of sentences and phrases (Gorno-Tempini et al., 2011
), likely stemming from an impaired phonological loop. This problem may contribute to difficulty understanding sentence-length verbal information, but is unlikely to explain selective difficulty for a specific word category on a task involving single word processing. Likewise, difficulty with lexical retrieval is unlikely to explain the pattern of modality-specific performance observed on our word recognition task, given that we would not expect a pure impairment of lexical retrieval to interact with the semantic categories in our study. While previous assessments of conceptual memory in lvPPA have generally shown preserved performance relative to assessments of verbal working memory (Gorno-Tempini et al., 2011
), our study is the first to evaluate modality-specific effects in conceptual memory in lvPPA. Our more fine-grained assessment of conceptual memory in lvPPA reveals disproportionate difficulty for concepts with strong sound associations, consistent with the pattern of cortical atrophy in lvPPA affecting canonical auditory association cortex.
In conclusion, we provide evidence from individuals with localized cortical atrophy that conceptual representations rely in part on modality-specific association cortex. We demonstrate that individuals with lvPPA have selective difficulty processing words with strong sound associations, and that this impairment is directly related to reduced gray matter density in a region of auditory association cortex that healthy adults activate when processing the same Sound words in a separate fMRI experiment. These findings suggest that concepts rely on feature representations in sensory association cortices.