In the spirit of the predictive nature of cognition, we agree with Gilead and colleagues that a predictive brain framework for abstract representations, contemplated as a hierarchy ranging from the tangible to the intangible, could be salutary. However, it is important to recognize that although a crucial aspect of the ontology proposed by Gilead and colleagues is the dichotomy between tangible and intangible entities, the latter remains rather ill-defined despite the formal treatment (sect. 2.1, para. 3). In particular, Gilead and colleagues define “intangible abstracta” (often called “abstract representations/concepts” in the literature on semantic representations) as categories whose concreta are not detected by our senses, but mainly transmitted from mind to mind using language. However, they also propose that some intangible dimensions of the intangible abstracta “may have an innate basis, or may be emergent properties discovered via personal experience” (sect. 2.1, para. 3), properties also relevant for the modality-specific and multimodal abstracta (both based on sensorimotor features) (sect. 2.1, para. 1 and 2). Consequently, the distinctions between the different kinds of representations are obscure and Gilead and colleagues’ definition of “intangible abstracta” seems somewhat contradictory to us. Therefore, it is important to get a clear idea of how personal experience and social interaction combine to produce intangible abstracta.
In light of these theoretical considerations, we propose that many intangible representations could be intangible abstracta with affective content. The plausibility of this view has been supported by many studies demonstrating the crucial role of emotion for intangible abstracta (Crutch et al. Reference Crutch, Troche, Reilly and Ridgway2013; Kousta et al. Reference Kousta, Vigliocco, Vinson, Andrews and Del Campo2011). In particular, although tangible entities have direct sensory referents (Crutch & Warrington Reference Crutch and Warrington2004; Montefinese et al. Reference Montefinese, Ambrosini, Fairfield and Mammarella2013; Paivio Reference Paivio1971), intangible abstracta tend to be more emotionally valenced (Crutch et al. Reference Crutch, Troche, Reilly and Ridgway2013; Kousta et al. Reference Kousta, Vigliocco, Vinson, Andrews and Del Campo2011; Vigliocco et al. Reference Vigliocco, Kousta, Della Rosa, Vinson, Tettamanti, Devlin and Cappa2013) and have low sensorimotor grounding (for a concise review, see Montefinese Reference Montefinese2019). In line with the idea that affective content is particularly relevant for intangible abstracta representation, a number of neuroimaging studies showed that intangible abstracta processing increases activation in brain regions involved in emotion processing (Vigliocco et al. Reference Vigliocco, Kousta, Della Rosa, Vinson, Tettamanti, Devlin and Cappa2013; Wang et al. Reference Wang, Wu, Ling, Xu, Fang, Wang, Binder, Men, Gao and Bi2018), such as the rostral anterior cingulate cortex.
Very recently it has been proposed that interoception (the perception of the internal state of the body) contributes to the perceptual grounding of intangible abstracta. Crucially, interoception is the most important perceptual modality in the experience of emotions, especially the negative ones (e.g., fear and sadness), over and above the traditional five sensory modalities (Connell et al. Reference Connell, Lynott and Banks2018). An exploration of emotion and of its perceptual grounding via interoception seems like a necessary step in building a comprehensive theory of abstract representational capacities.
Still, taking affective information into account might not suffice to capture representation of intangible abstracta. In this regard, recent multimodal models suggest that supplementing affective information with information related to the statistical distribution of concepts in language (i.e., distributional models of semantic representation; Landauer & Dumais Reference Landauer and Dumais1997) drastically improves prediction of human affective judgments (Bestgen & Vincze Reference Bestgen and Vincze2012; Recchia & Louwerse Reference Recchia and Louwerse2015; Vankrunkelsven et al. Reference Vankrunkelsven, Verheyen, Storms and De Deyne2018). More importantly, recent work by Lenci et al. (Reference Lenci, Lebani and Passaro2018) reveals a strong link between distributional statistics and emotion: intangible representations have more affective content and tend to co-occur with contexts with higher emotive value. However, it is worth noting that the contribution of the distributional models to semantic representation goes beyond that of affective intangible abstracta. Indeed, it has been shown that these models can successfully account for semantic and linguistic judgments, as well as higher-level judgments such as probability judgments and risk perception in a human-like manner (Bhatia et al. Reference Bhatia, Richie and Zou2019; Rotaru et al. Reference Rotaru, Vigliocco and Frank2018). As is the case for emotion, the importance of distributional information for intangible abstracta is also supported by neuroimaging studies. Intangible abstracta reliably engage neural systems associated with linguistic processing (especially, left anterior temporal cortex and left inferior frontal gyrus) to a greater extent than tangible abstracta (Wang et al. Reference Wang, Conder, Blitzer and Shinkareva2010). Increased activity for intangible abstracta in networks associated with language processing appears to be specifically associated with distributional similarity, versus other aspects of intangible representations which do not appear to be localized to language-related networks (Wang et al. Reference Wang, Wu, Ling, Xu, Fang, Wang, Binder, Men, Gao and Bi2018). As intangible abstracta are mainly acquired through verbal experience (as Gilead and colleagues acknowledge in sect. 2.1, para. 3) and the distributional theory represents one of the main theoretical frameworks in the semantic literature, it is surprising that such a role of language is not addressed directly. Given the importance of these models in explaining both intangible and tangible representations, we think that Gilead and colleagues should incorporate them in their theory. Moreover, by revealing the statistical relations between abstract entities, distributional models represent a powerful tool to integrate Gilead and colleagues’ account and predictive brain theories, which assume the brain as a statistical inferential machine.
In short, we believe that Gilead and colleagues have missed a chance to “provide cognitive scientists with an accurate ontology of the representational entities that exist in our mind – and that subserve predictive cognition” (sect. 5.1, para. 5). What we think is missing from their analysis is how emotion and distributional information fits in with the proposed ontology. In keeping with a metaphor used by the authors, interoceptive experience and linguistic distribution would represent two additional “tricks” used by our brain both to build the different layers of the representational hierarchy and to “transcend the here-and-now,” and we think that the authors’ model could benefit from integrating them.
In the spirit of the predictive nature of cognition, we agree with Gilead and colleagues that a predictive brain framework for abstract representations, contemplated as a hierarchy ranging from the tangible to the intangible, could be salutary. However, it is important to recognize that although a crucial aspect of the ontology proposed by Gilead and colleagues is the dichotomy between tangible and intangible entities, the latter remains rather ill-defined despite the formal treatment (sect. 2.1, para. 3). In particular, Gilead and colleagues define “intangible abstracta” (often called “abstract representations/concepts” in the literature on semantic representations) as categories whose concreta are not detected by our senses, but mainly transmitted from mind to mind using language. However, they also propose that some intangible dimensions of the intangible abstracta “may have an innate basis, or may be emergent properties discovered via personal experience” (sect. 2.1, para. 3), properties also relevant for the modality-specific and multimodal abstracta (both based on sensorimotor features) (sect. 2.1, para. 1 and 2). Consequently, the distinctions between the different kinds of representations are obscure and Gilead and colleagues’ definition of “intangible abstracta” seems somewhat contradictory to us. Therefore, it is important to get a clear idea of how personal experience and social interaction combine to produce intangible abstracta.
In light of these theoretical considerations, we propose that many intangible representations could be intangible abstracta with affective content. The plausibility of this view has been supported by many studies demonstrating the crucial role of emotion for intangible abstracta (Crutch et al. Reference Crutch, Troche, Reilly and Ridgway2013; Kousta et al. Reference Kousta, Vigliocco, Vinson, Andrews and Del Campo2011). In particular, although tangible entities have direct sensory referents (Crutch & Warrington Reference Crutch and Warrington2004; Montefinese et al. Reference Montefinese, Ambrosini, Fairfield and Mammarella2013; Paivio Reference Paivio1971), intangible abstracta tend to be more emotionally valenced (Crutch et al. Reference Crutch, Troche, Reilly and Ridgway2013; Kousta et al. Reference Kousta, Vigliocco, Vinson, Andrews and Del Campo2011; Vigliocco et al. Reference Vigliocco, Kousta, Della Rosa, Vinson, Tettamanti, Devlin and Cappa2013) and have low sensorimotor grounding (for a concise review, see Montefinese Reference Montefinese2019). In line with the idea that affective content is particularly relevant for intangible abstracta representation, a number of neuroimaging studies showed that intangible abstracta processing increases activation in brain regions involved in emotion processing (Vigliocco et al. Reference Vigliocco, Kousta, Della Rosa, Vinson, Tettamanti, Devlin and Cappa2013; Wang et al. Reference Wang, Wu, Ling, Xu, Fang, Wang, Binder, Men, Gao and Bi2018), such as the rostral anterior cingulate cortex.
Very recently it has been proposed that interoception (the perception of the internal state of the body) contributes to the perceptual grounding of intangible abstracta. Crucially, interoception is the most important perceptual modality in the experience of emotions, especially the negative ones (e.g., fear and sadness), over and above the traditional five sensory modalities (Connell et al. Reference Connell, Lynott and Banks2018). An exploration of emotion and of its perceptual grounding via interoception seems like a necessary step in building a comprehensive theory of abstract representational capacities.
Still, taking affective information into account might not suffice to capture representation of intangible abstracta. In this regard, recent multimodal models suggest that supplementing affective information with information related to the statistical distribution of concepts in language (i.e., distributional models of semantic representation; Landauer & Dumais Reference Landauer and Dumais1997) drastically improves prediction of human affective judgments (Bestgen & Vincze Reference Bestgen and Vincze2012; Recchia & Louwerse Reference Recchia and Louwerse2015; Vankrunkelsven et al. Reference Vankrunkelsven, Verheyen, Storms and De Deyne2018). More importantly, recent work by Lenci et al. (Reference Lenci, Lebani and Passaro2018) reveals a strong link between distributional statistics and emotion: intangible representations have more affective content and tend to co-occur with contexts with higher emotive value. However, it is worth noting that the contribution of the distributional models to semantic representation goes beyond that of affective intangible abstracta. Indeed, it has been shown that these models can successfully account for semantic and linguistic judgments, as well as higher-level judgments such as probability judgments and risk perception in a human-like manner (Bhatia et al. Reference Bhatia, Richie and Zou2019; Rotaru et al. Reference Rotaru, Vigliocco and Frank2018). As is the case for emotion, the importance of distributional information for intangible abstracta is also supported by neuroimaging studies. Intangible abstracta reliably engage neural systems associated with linguistic processing (especially, left anterior temporal cortex and left inferior frontal gyrus) to a greater extent than tangible abstracta (Wang et al. Reference Wang, Conder, Blitzer and Shinkareva2010). Increased activity for intangible abstracta in networks associated with language processing appears to be specifically associated with distributional similarity, versus other aspects of intangible representations which do not appear to be localized to language-related networks (Wang et al. Reference Wang, Wu, Ling, Xu, Fang, Wang, Binder, Men, Gao and Bi2018). As intangible abstracta are mainly acquired through verbal experience (as Gilead and colleagues acknowledge in sect. 2.1, para. 3) and the distributional theory represents one of the main theoretical frameworks in the semantic literature, it is surprising that such a role of language is not addressed directly. Given the importance of these models in explaining both intangible and tangible representations, we think that Gilead and colleagues should incorporate them in their theory. Moreover, by revealing the statistical relations between abstract entities, distributional models represent a powerful tool to integrate Gilead and colleagues’ account and predictive brain theories, which assume the brain as a statistical inferential machine.
In short, we believe that Gilead and colleagues have missed a chance to “provide cognitive scientists with an accurate ontology of the representational entities that exist in our mind – and that subserve predictive cognition” (sect. 5.1, para. 5). What we think is missing from their analysis is how emotion and distributional information fits in with the proposed ontology. In keeping with a metaphor used by the authors, interoceptive experience and linguistic distribution would represent two additional “tricks” used by our brain both to build the different layers of the representational hierarchy and to “transcend the here-and-now,” and we think that the authors’ model could benefit from integrating them.