Distributed Code for Semantic Relations Predicts Neural Similarity

Abstract

The ability to generate and process semantic relations is central to many aspects of human cognition. Theorists have long debated whether such relations are coded as atomistic links in a semantic network, or as distributed patterns over some core set of abstract relations.The form and content of the conceptual and neural representations of semantic relations remains to be empirically established. The present study combined computational modeling and neuroimaging to investigate the representation and comparison of abstract semantic relations in the brain. By using sequential presentation of verbal analogies, we decoupled the neural activity associated with encoding the representation of the first-order semantic relation between words in a pair from that associated with the second-order comparison of two relations. We tested alternative computational models of relational similarity in order to distinguish between rival accounts of how semantic relations are coded and compared in the brain. Analyses of neural similarity patterns supported the hypothesis that semantic relations are coded, in the parietal cortex, as distributed representations over a pool of abstract relations specified in a theory-based taxonomy. These representations, in turn, provide the immediate inputs to the process of analogical comparison, which draws on a broad frontoparietal network. This study sheds light not only on the form of relation representations but also on their specific content.Significance: Relations provide basic building blocks for language and thought. For the past half century, cognitive scientists exploring human semantic memory have sought to identify the code for relations. In a neuroimaging paradigm, we tested alternative computational models of relation processing that predict patterns of neural similarity during distinct phases of analogical reasoning. The findings allowed us to draw inferences not only about the form of relation representations, but also about their specific content. The core of these distributed representations is based on a relatively small number of abstract relation types specified in a theory-based taxonomy. This study helps to resolve a longstanding debate concerning the nature of the conceptual and neural code for semantic relations in the mind and brain. Distributed Code for Relations

The Internal Structure of Semantic Relations: Effects of Relational Similarity and Typicality

“…A recent model of relational learning and reasoning, the Bayesian Analogy with Relational Transformations model (BART; Chiang et al, 2019;Lu et al, 2012Lu et al, , 2019b implements both distributed role-based representations, as in Hummel and Holyoak's LISA model (1997), but also distributed representations of relational features, as in Chaffin's (1992) relational element theory. BART learns semantic relations from exemplars of those relations by associating a relational vector and two feature vectors representing the two words by learning a set of relational weights that connects them (Figure 3).…”

“…While there have been theoretical debates concerning whether or not individual relations are represented explicitly, i.e. independently of the concepts they connect (Doumas & Hummel, 2005; for a review, see Popov, Hristova, & Anders, 2017), empirical research has only recently started to address these questions systematically (Asmuth & Gentner, 2016;Chiang, Peng, Lu, Holyoak, & Monti, 2019;Estes & Jones, 2006;Gagné, Spalding, & Ji, 2005;Popov et al, 2017).…”

“…The present study aims to explore how the typicality of the relational exemplars affect explicit relational processing in a range of tasks, such as analogical verification, associative recognition, relational exemplar generation and similarity ranking. Typicality/similarity norms derived from studies of relational reasoning have been used to predict relational priming in short-term memory (Estes & Jones, 2006), to predict relational luring in longterm memory (Popov et al, 2017), to predict patterns of neural responses to semantic relations in fMRI research (Chiang et al, 2019), and to train and test models of relational representations (Jurgens, Turney, Mohammad, & Holyoak, 2012;Lu, Wu, & Holyoak, 2019b). Thus, just like with entity categories, data for typicality and similarity of semantic relations has proven to be useful for model testing and development.…”