Ashley Thomas highlights identifying, categorizing, and communicating social relationships but does not fully address mechanisms for processing structural regularities – i.e., the systematic organization of social relationships among three or more individuals. These mechanisms are vital to a comprehensive account of core cognition of social relationships. I argue that processing such regularities requires (i) combinatorial mechanisms, which enable the integration of dyadic relationships into broader social networks, and (ii) relational reasoning that takes into account systematic relations between relationships. Both are essential to processing social relationships and their organization into larger structures.

Social structural regularities are widespread. Throughout the animal kingdom, dominance relationships tend to be transitive – if A dominates B and B dominates C, then A is likely to dominate C (Tibbetts et al., Reference Tibbetts, Pardo-Sanchez and Weise2022), and to be organized in pyramids – one dominant over several subordinates is more common than the reverse (Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023). The prevalence of these regularities provides evolutionary leverage for the emergence of cognitive mechanisms designed to process them.

Mechanisms for processing social structures share key features of core cognition. They emerge early, persist throughout development, and are found across species. For instance, human adults, children, and infants draw inferences consistent with the pyramidal shape of hierarchies (Hok et al., Reference Hok, Vasquez, Barakzai and Shaw2024; Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023). Similarly, transitive inferences over dominance relationships are found both in humans – from infancy (Gazes et al., Reference Gazes, Hampton and Lourenco2017; Mascaro & Csibra, Reference Mascaro and Csibra2014), and across species including birds, fish, and insects (Grosenick et al., Reference Grosenick, Clement and Fernald2007; Paz-y-Miño et al., Reference Paz-y-Miño, Bond, Kamil and Balda2004; Tibbetts et al., Reference Tibbetts, Agudelo, Pandit and Riojas2019). Assumptions about social structures are abstract – they cannot be reduced to a purely sensory vocabulary. They also support inferences beyond the sensory input, about relationships between individuals who have never been observed interacting. Thus, core cognition might very well include mechanisms for representing social structures.

Assumptions about the shape of social structures have been observed in infants with months of social experience (e.g., Gazes et al., Reference Gazes, Hampton and Lourenco2017; Mascaro & Csibra, Reference Mascaro and Csibra2014; Spokes & Spelke, Reference Spokes and Spelke2017). Thus, it remains unclear whether these assumptions are innate or learned through experience. Indeed, computational models show how learners can navigate a hypothesis space of structural forms and choose the most appropriate one in a given domain, even with limited evidence (Kemp & Tenenbaum, Reference Kemp and Tenenbaum2008). However, even flexible models of structural learning require pre-existing cognitive abilities, such as the capacity to combine dyadic relationships into larger structures (combinatoriality) and to process structural shapes (relational reasoning). These foundational capacities appear to be prerequisites for learning about social structures, rather than products of learning, and are likely part of the start-up kit for processing social relationships.

First, processing social structures requires combinatorial processes to combine separate representations of relationships into larger networks. Evidence suggests they operate early. Infants effectively memorize hierarchical relationships presented sequentially (first A>B, then B>C, and last C>D), making it possible to build a unified representation of the hierarchy incrementally (first A>B, then A>B>C, and finally A>B>C>D). In contrast, infants struggle to memorize the same relationships presented in disjointed sequences (first A>B, then C>D, and finally B>C). This order requires holding separate relationships in mind before integrating three relationships at once into a unified structure (first A>B, then A>B and C>D, and finally A>B>C>D). Thus, infants attempt to combine dyadic relationships into a unified representation of a social structure, rather than memorizing them in isolation (Mascaro & Csibra, Reference Mascaro and Csibra2014). Combinatoriality is key to processing social structures, which are rarely – if ever – directly observed. Instead, learners typically witness interactions between individuals, on the basis of which relationships might be inferred, and can, in turn, be combined in larger representations of social structures. Combinatorial processes allow learners to generate a wide range of social structure representations from a limited repertoire of social relationships – supporting flexible learning across diverse social environments.

The second key component for processing social structures is relational reasoning – the ability to process relations between relationships. This capacity is required to form assumptions about the way sets of relationships are organized, enabling learners to process patterns and regularities within social structures. For instance, assumptions about the transitivity or pyramidal organization of hierarchical relationships do not refer to relationships between pairs of specific individuals – they specify how hierarchical relationships in general are organized. Relational reasoning may be achieved through mechanisms that explicitly represent social structures (e.g., specifying structural forms and the processes that generate them) or rely on more implicit mechanisms yielding computations consistent with structural regularities without directly representing them.

Mechanisms supporting relational reasoning are key to organizing and inferring social relationships not only within each relational domain but also across domains. They are required, for instance, to think about the systematic ways in which alliances and hierarchical relationships interact – a capacity evident from infancy (Pun et al., Reference Pun, Birch and Baron2016, Reference Pun, Birch and Baron2022). Moreover, they are likely to play a central role in helping learners infer their relationships to others, from observing how others relate to them and to each other (Thomas et al.).

Data on the processing of structural regularities offer key insights to test classic theoretical accounts. For instance, they reveal that infants’ social inferences about dominance relationships do not align with predictions from the authority ranking model (Fiske, Reference Fiske1992). This model assumes a linear ordering in which everyone’s rank can be compared with everyone else’s, making transitive inferences obligatory. However, infants’ inferences over dominance relationships are not obligatory (Mascaro & Csibra, Reference Mascaro and Csibra2012). Moreover, infants’ assumptions about the pyramidal shape of hierarchies cannot be captured by a simple linear ordering (Mascaro et al., Reference Mascaro, Goupil, Pantecouteau, Depierreux, Van der Henst and Claidière2023). Infants’ assumptions about the shape of social structures also illuminate why core cognition of social relationships is domain specific, complementing Ashley Thomas’s argument. They suggest that the representation of social relationships cannot be reduced to either core agency cognition or, more broadly, to reasoning based on naïve psychology. The distinct patterns in social structures necessitate a specialized cognitive framework separate from other domains of core cognition.