Spelke has done a masterful job of describing six early emerging domain-specific systems of core knowledge that appear in infancy not only in children, but also in other animals. I focus here on two aspects of knowledge that Spelke argues are not present at birth and that humans develop over the first year of life – the ability to treat others as social agents, which develops at 10 months, and the ability to treat themselves and other social agents as having shareable experiences of objects and events, which develops at 12 months. Around their first birthday, children begin to construct a new system of knowledge of themselves and others as actors, collaborators, and sentient beings. How do these new concepts, which go beyond core knowledge, develop?

Spelke begins by considering whether infants' concepts of social agents and their mental states are, in fact, not a newly developed skill at all, but rather present from the start. Under this view, the concepts are late-emerging because their appearance depends on other late-emerging capacities. She entertains, and rejects, three abilities that could potentially account for the development of these concepts. (1) The ability to share experiences of objects with others – shared intentionality – which onsets around the end of the first year but comes so naturally to human children that it might be considered innate (Tomasello, Reference Tomasello2018). Spelke applies the criteria she uses to argue for core systems to shared intentionality. But she finds that it does not have the characteristics of a core system, nor is there an obvious explanation for its appearance at 12 months. (2) The ability to learn from others by interpreting the information directed to them as generic (Csibra & Gergely, Reference Csibra and Gergely2009). Children do gain different information when an adult actively draws their attention to objects (thus creating a pedagogical atmosphere), compared to when an adult acts on objects without looking, speaking, or gesturing to the child. But a bias to interpret information from an adult as generic is just as likely to get in the way of learning about a social partner's mental states as to foster discovery of those states (cf. Powell & Spelke, Reference Powell and Spelke2018). (3) The ability to interact with objects symbolically; for example, to recognize an object from its picture. Interacting with people over pictorial symbols might lead infants to view those people as social agents who share their experiences of objects. Spelke rejects this explanation because infant's new conceptions of people, their actions, and their mental states at 12 months precede, rather than follow, children's understanding of pictures as symbols. The emergence of concepts of social agents and their mental states is therefore not likely to depend on any of these three abilities.

The fourth hypothesis is, for me, the most interesting. Spelke argues that learning a language brings children a new understanding of people and their mental states. Her proposal is that this understanding arises when children “learn enough of their language to interpret people's speech as simultaneously social and object-directed (at about 10 months) and to view their acts of speaking as invitations to share their experiences of the objects and events that they speak about (at about 12 months)” (Spelke, Reference Spelke2022, p. 403). Mastery of a specific natural language is thus thought to underlie the onset of social agents and their mental states.

Language may indeed play an essential role in the development of these late-emerging skills. But learning language from another cannot be the whole story. The evidence comes from homesigners – deaf children whose profound hearing losses prevent them from learning spoken language, and whose hearing parents have not exposed them to sign language. These children construct gestures, called homesigns, which have the properties of language that Spelke considers essential to the acquisition of social agents and their mental states – an open-ended lexicon, a productively combinatorial grammar generating abstract structured representations, and a compositional semantics (see Goldin-Meadow, Reference Goldin-Meadow2020). Even though homesigners do not have a model for a conventional language, the gestures that their hearing parents produce when they talk to them might provide a model for combinatoriality or compositionality. But they don't (Goldin-Meadow, Mylander, & Butcher, Reference Goldin-Meadow, Mylander and Butcher1995, Reference Goldin-Meadow, Mylander and Franklin2007), confirming that homesigners do not learn these linguistic properties from others.

Given these facts about homesign, a good test of Spelke's hypothesis is to ask whether homesigners are able to treat people as social agents and to treat themselves and others as having mental states about shareable experiences of objects. They might, of course, develop these skills late, particularly because homesigners are delayed in the development of some linguistic skills (e.g., communicating about the non-here-and-now, Morford & Goldin-Meadow, Reference Morford and Goldin-Meadow1997). But the crucial question is do they have them at all? It's hard to imagine that adult homesigners, who can do relatively sophisticated things with their home-made languages (e.g., Coppola & Newport, Reference Coppola and Newport2005; Goldin-Meadow, Brentari, Coppola, Horton, & Senghas, Reference Goldin-Meadow, Brentari, Coppola, Horton and Senghas2015), do not have these concepts. But Pyers and Senghas (Reference Pyers and Senghas2009) have found that the homesigners who initially created Nicaraguan Sign Language have difficulty attributing false beliefs to others and thus are not proficient at understanding other peoples' minds. Moreover, as Spelke points out, nonhuman animals can appear to have an understanding of social agents in predictable contexts but show no understanding in novel contexts. In other words, homesigners might look more socially adept than they actually are. So this is not merely a thought-experiment – the test needs to be carried out.

If it turns out that adult homesigners are not able to treat people as social agents and treat themselves and others as having mental states about shareable experiences, this would provide strong support for Spelke's hypothesis – that learning language from another plays an important role in the development of these late-emerging concepts.

But if adult homesigners do display an understanding of these concepts, then learning language from another is not necessary to develop these abilities – creating language works too. Note, however, that learning some aspects of language might be necessary for children to develop other concepts; for example, homesigners have difficulty developing concepts of large exact numbers (Spaepen, Coppola, Spelke, Carey, & Goldin-Meadow, Reference Spaepen, Coppola, Spelke, Carey and Goldin-Meadow2011) and certain spatial relations (Gentner, Özyurek, Gurcanli, & Goldin-Meadow, Reference Gentner, Özyurek, Gurcanli and Goldin-Meadow2013), presumably because they lack a model from others for how to express these notions.

One final, clarifying point is worth making. Whether homesigners are truly a good test of Spelke's theory depends on what is central to the theory. Is it crucial that children learn from others only that they can share ideas about objects and people? If so, then homesigners might not be a good test case for the theory simply because their hearing parents gesture when they talk and use those gestures to point out objects and share attention to the objects with their children. Alternatively, is it crucial to the theory that children learn a compositional language from others? If so, homesigners do present a potential problem for the theory because their hearing parents do not provide them with a usable model for compositionality – the children create rather than learn these structures.