We endorse Thomas’s call to understand the development of systematic reasoning about relationships. We are not yet convinced, however, that relational models should be central to an account of infants’ relationship reasoning. Thomas presents these categories as the basis for infants’ ability to interpret and predict the social behaviors of interacting agents. However, many behaviors that are central to social relationships – e.g. giving, helping, cooperating, communicating, and imitating – occur regularly across many relationship types. This many-to-many mapping between behaviors and relationships suggests that immediately categorizing a relationship from observed behavior may be more difficult than Thomas proposes. This may be especially true for infants who lack substantial contextual and cultural knowledge, and for relationship categories defined by patterns of reciprocity that unfold over time.

Consider giving. Thomas suggests that infants use acts of giving to recognize equality matching relationships, thus explaining their expectation that one act of giving will be followed by another, reciprocal act (Tatone & Csibra, Reference Tatone and Csibra2020). But giving is not exclusive to equality matching relationships. Close family and friends give one another gifts and resources; leaders and followers may give provisions or tributes, respectively. The fact that one act of giving ought to be insufficient evidence of an equality matching relationship casts doubt on whether infants’ expectations of reciprocity stem from this specific relational model, rather than a broader expectation that social regard is often mutual. A broader reciprocal relationship representation would also help account for infants’ expectation that the giver will keep giving to, rather than taking from, the recipient, even when exchange is unbalanced (Tatone, Geraci, & Csibra, Reference Tatone, Geraci and Csibra2015). Outside the lab, it would also be forgiving of fragmented observations and ignorance of the value of different resources (e.g., money).

Imitation is another behavior Thomas maps onto a specific relational model, proposing that it is an expression of “oneness” characteristic of communal sharing relationships. But imitation is ubiquitous in human social life: imitative alignment occurs in interactions between parents and infants, teachers and students, customers and salespeople, and research participants and confederates who have only just met (see Chartrand & Lakin, Reference Chartrand and Lakin2013, for review). Should observers conclude that all these relationships are to some degree communal sharing relationships? An alternative is that imitative alignment is seen as evidence of a valuing of others’ goals or perspectives, which can be a feature of many relationship types (Powell & Winkielman, Reference Powell, Winkielman, Genschow and Cracco2024). Of note, infants do not always expect group members to imitate one another (for evidence that strong imitative expectations depend upon the perception of behaviors as conventional or ritualistic, see Liberman et al., Reference Liberman, Woodward, Sullivan and Kinzler2016; Hennessy, Spelke, & Powell, Reference Hennessy, Spelke and Powell2025)

A general alternative to Thomas’s proposal is that people, including infants, can reason about the features and statistics of relationships without necessarily categorizing them first. This would be in line with other intuitive theories: observers can reason about the solidity of an object or goal of an agent without deciding what kind of object or agent they have encountered (Spelke, Reference Spelke2022). Similarly, people may infer features of a relationship – defined as the relation between any set of agents – and reason about how those features will impact the relevant parties’ actions or emotions, without necessarily assigning the relationship to a particular relational model or other category.

As Thomas discusses in Box 2, Powell (Reference Powell2022) proposed that infants represent the relationship feature of care or “affiliation” as the weighted value that one individual places on another’s rewards or welfare. This representation can be integrated with an intuitive psychology that explains agents’ behavior and emotions in terms of what they believe and value (Jara-Ettinger et al., Reference Jara-Ettinger, Gweon, Schulz and Tenenbaum2016; Doan, Ong, & Wu, Reference Doan, Ong and Wu2024). This integration allows observers 1) to use others’ behavior to infer strength of interpersonal care in the form of the magnitude of this weight, and 2) to use information about strength of care to predict interpersonal actions and emotions, in a graded fashion, within and across relationship categories (Jiang et al., Reference Jiang, Smith-Flores, Liang, Ong and Powell2025; Quillien, Reference Quillen2023; Smith-Flores & Powell, Reference Smith-Flores and Powell2023; Tompkins et al., Reference Tompkins, Vasquez, Gerdin, Dunham and Liberman2024). As Thomas notes, this need not entail relationship partners supporting all one another’s goals; they may disagree on what is best for one individual’s overall welfare, as in the case of a parent who hinders a child’s dangerous or unhealthy goal. Children understand and approve of such protective hindering (Tompkins & Powell, Reference Tompkins and Powell2025), and infants may in some cases as well. Finally, like communal sharing models, interpersonal care is inconsistent with exact exchange; this pattern suggests behavior motivated by expected reciprocal rewards, rather than valuing of the other’s welfare.

While strength of care is an important feature for understanding many social relationships, we agree with Thomas that it does not define relationships alone. Another key feature that seems to be represented from infancy is relative difference in agentic capacity or power between relationship participants (Mascaro & Csibra, Reference Mascaro and Csibra2012). These and other features can be combined to flexibly form relationship categories, which are indeed a component of mature social reasoning (Earp et al., Reference Earp, McLoughlin, Monrad, Clark and Crockett2021). Some categories could be innate, but many categories would be learned from the statistical co-occurrence of features, as well as events and contexts.

Relatedly, the role of experience with specific social interactions in early social reasoning is underexplored. For example, might infants’ putative expectations about saliva sharing and comforting (Thomas et al.) be based on the association of multiple actions or events with the same relationship partner? Caregivers comfort and feed infants; they are also the most likely people to reach into infants’ mouths, to feed them, clean them, or remove foreign objects. To fully consider associative explanations, studies should test for comforting expectations when an adult feeds or probes an infant or puppet without first putting the food or finger in their own mouth.

In sum, we propose that humans’ intuitive theory of relationships is based on relationship features, supplemented by categories that combine these features, rather than core relational models. This multifaceted system can best explain how people flexibly reason about social information, spanning from isolated interaction observations to the histories of lifelong relationships.