The Neural Representation of Concrete and Abstract Concepts

Abstract

concepts elicited greater activation than concrete concepts in such verbal processing areas. By contrast, concrete concepts elicited greater activation than abstract concepts in visuospatial processing (precuneus, posterior cingulate, and fusiform gyrus). This meta-analysis was limited to univariate comparisons of categories of concepts and did not have access to the activation patterns evoked by individual concepts. This limitation potentially overlooks nuanced distinctions in the representational structure. Univariate contrasts potentially overlook critical relationships across neural states and neural regions (Mur et al., 2009). Through the use of MVPA techniques, more recent studies have begun to examine the underlying semantic structure of sets of abstract concepts. The next section focuses on various imaging studies examining the neural activation patterns associated with abstract concepts and explores the possible semantic structures that are specific to abstract concepts. Neurosemantic Dimensions of Abstract Meaning As in the case of concrete concepts, the semantic dimensions underlying abstract concept categories can be identified from their activation patterns. One of the first attempts to decode the semantic content of abstract semantic information was conducted by Anderson, Kiela, Clark, and Poesio (2017). A set of individual concepts that belonged to various taxonomic categories (tools, locations, social roles, events, communications, and attributes) were decoded from their activation patterns. Whether a concept belonged to one of two abstract semantic categories (i.e., Law orMusic) was also decoded from the activation patterns of individual concepts. Although these abstract semantic categories could be decoded based on their activation patterns, the localization of this dissociation is unclear. Neurally-based semantic dimensions underlying abstract concepts differ from the dimensions underlying concrete concepts. Vargas and Just (2019) investigated the fMRI activation patterns of 28 abstract concepts (e.g., ethics, truth, spirituality) focusing on individual concept representation and the relationship between the activation profiles of these concept representations. Factor analyses of the activation patterns evoked by the stimulus set revealed three underlying semantic dimensions. These dimensions corresponded to 1) the degree to which a concept was Verbally Represented, 2) whether a concept was External (or Internal) to the individual, and 3) whether the concept contained Social Content. The Verbal Representation dimension was present across all participants and was the most salient of the semantic dimensions. Concepts with large positive factor scores for this factor included compliment, faith, and ethics, while concepts with large negative scores for this factor included gravity, force, and acceleration. The former three concepts seem far less perceptual than the latter three. For the Externality factor, a concept that is external is one that requires the representation of the world 456 r. vargas and m. a. just https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108635462.029 Downloaded from https://www.cambridge.org/core. Carnegie Mellon University, on 15 Sep 2021 at 00:03:10, subject to the Cambridge Core terms of use, available at outside oneself and the relative non-involvement of one’s own state. An internal concept is one that involves the representation of the self. At one extreme of the dimension lie concepts that are external to the self (e.g., causality, sacrilege, and deity). At the other extreme lie concepts that are internal to the participant (e.g. spirituality and sadness). The last semantic dimension was interpreted to correspond to Social Content. The concepts at one extreme of the dimension included pride, gossip, and equality, while the concepts at the other extreme included heat, necessity, and multiplication. Together these semantic dimensions underlie the neural representations of the 28 abstract concepts. One surprising finding was that the regions associated with the Verbal Representation dimension were the same as those found in the meta-analysis conducted by Wang et al. (2010) that contrasted the activation between concrete and abstract concepts. Activation in the LIFG (a region clearly involved in verbal processing) was evoked by concepts such as faith and truth, while the left supramarginal gyrus (LSMG) and left lateral occipital complex (LOC), both of which are involved in different aspects of visuospatial processing, were associated with concepts such as gravity and heat. Moreover, the output of the factor analysis (i.e., factor scores) for the Verbal Representation factor also suggested that the abstractness of the neural patterning in these regions for an individual concept is represented as a point on a continuum between language systems and perceptual processing systems. This interpretation corresponds to the intuition that abstractness is not a binary construct but rather a gradient-like translation of a concept into a more verbal encoding. This conclusion is somewhat surprising given that the set of 28 concepts are all qualitatively abstract, in that they have no direct perceptual referent. The amount of activation in LIFG evoked by a given abstract concept corresponds to its Verbal Representation factor score. These results raise an interesting theoretical and psychological question regarding the role of neural language systems, particularly LIFG, in the verbal representation of abstract concepts. That is, what does it mean, neurally and psychologically, for an abstract concept to be verbally represented? Abstract Concepts as Verbal RepresentationsConcepts as Verbal Representations What does it mean for an abstract concept to be represented in regions involved in verbal processing and to evoke activation in the LIFG? When the LIFG is artificially lesioned through the repeated use of transcranial magnetic stimulation (TMS), healthy participants show a 150 ms slower response time for comprehending abstract concepts (e.g., chance) (Hoffman, Jefferies, & Lambon Ralph, 2010). This same TMS-based lesioning procedure showed no influence in the amount of time needed to respond to concrete concepts (e.g., apple). However, these differences in the impact of TMS were nullified when the abstract concepts were presented within a context (e.g., The Neural Representation of Concrete and Abstract Concepts 457 https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108635462.029 Downloaded from https://www.cambridge.org/core. Carnegie Mellon University, on 15 Sep 2021 at 00:03:10, subject to the Cambridge Core terms of use, available at “You don’t stand a chance”). These results suggest that the abstractness of a concept is dependent on whether it requires integration of meaning across multiple contexts (Crutch & Warrington 2005; 2010; Hoffman 2016; Hayes & Kraemer, 2017). Moreover, the LIFG seems to be involved in the contextdependent integration of meaning. Given that LIFG appears to be involved in the contextualization of the meaning of abstract concepts (Hoffman et al., 2010) and that the magnitude of activation in LIFG is directly proportional to the degree that it is verbally represented (Vargas & Just, 2019), taken together these results suggest that the activation in LIFG reflects the magnitude of mental activity required to contextualize the meaning of a lexical concept. LIFG has been shown to elicit greater activation for sentence-level representations as compared to word-level concepts (Xu, Kemeny, Park, Frattali, & Braun, 2005). It may be the case that the central cognitive mechanism underlying the neural activation in LIFG represents the integration of meaning across multiple representations in order to form a new representation that is a product of its components. That is, the components of meaning of apple require less computation (in LIFG) to generate a composite representation than the concept of chance. Also, providing a context for chance, as in “You don’t stand a chance”, reduced the cognitive workload by providing a more explicit version of its meaning. A similar mechanism can account for the greater activation in LIFG for sentences than for individual words, because constructing a sentence-level representation requires combining the meanings of individual concept representations in a mutually context-constraining way. As previously discussed, another region involved in the integrating of meaning for concepts is the anterior temporal lobe (ATL). ATL has been implicated in the integration of semantic features to form a composite representation of object concepts (Coutanche & Thompson-Schill, 2015). However, unlike LIFG, ATL does not appear to differentiate between abstract concepts that vary based on the degree that they are verbally represented (as defined by their factor scores in Vargas & Just (2019)). In sum, the integration of abstract concept representations with other concepts in a sentence seems to require additional computation. However, it is unclear whether these integrating computations are processing some episodic contexts (as suggested by the results of Hoffman et al., 2010), or some specific concept representations, or use some more general amodal representational format. Hybrid Concepts: Neither Completely Concrete nor Completely Abstract Hybrid concepts are concepts that can be experienced directly but require additional processing beyond the five basic perceptual faculties to be 458 r. vargas and m. a. just https://www.cambridge.org/core/terms. https://doi.org/10.1017/9781108635462.029 Downloaded from https://www.cambridge.org/core. Carnegie Mellon University, on 15 Sep 2021 at 00:03:10, subject to the Cambridge Core terms of use, available at evoked. These concepts do not neatly fit within the dichotomy of concrete vs. abstract. For example, the concept envy cannot be tasted, seen, heard, smelled, or touched, b