The language-of-thought (LoT) framework enables us to formulate testable hypotheses about the representational formats, computational structures, and expressive power of infants' thoughts before they can effectively communicate via signed or spoken language. Quilty-Dunn et al.'s take on the LoT hypothesis provides a highly generative framework for operationalizing the relevant units in this hypothesis space. Here, I explain why cognitive developmentalists should embrace such an LoT approach.

LoT-framed hypotheses can already be found throughout foundational studies in infant cognitive science, although they are not necessarily explicitly formulated as such. They can be found in our attempts to understand the early formats of representations of objects, number, space, and agents (e.g., Baillargeon, Reference Baillargeon2004; Feigenson, Dehaene, & Spelke, Reference Feigenson, Dehaene and Spelke2004; Kibbe, Reference Kibbe2015; Leslie, Reference Leslie, Hirschfeld and Gelman1994; Spelke & Kinzler, Reference Spelke and Kinzler2007; Vasilyeva & Lourenco, Reference Vasilyeva and Lourenco2012), learning and reasoning (e.g., Denison & Xu, Reference Denison and Xu2019; Rabagliati, Ferguson, & Lew‐Williams, Reference Rabagliati, Ferguson and Lew-Williams2019), and social cognition (e.g., Kushnir, Reference Kushnir2022; Leslie, Reference Leslie1987), to name just a few examples. There is also a growing literature on infants' capacity for combinatorial thought (e.g., Cesana-Arlotti et al., Reference Cesana-Arlotti, Martín, Téglás, Vorobyova, Cetnarski and Bonatti2018; Piantadosi, Palmeri, & Aslin, Reference Piantadosi, Palmeri and Aslin2018). Although there is insufficient space here to engage with all of the nuances of these domains, I aim to draw a throughline: Across these domains infant researchers ask, what are the basic representational units of infants' mental lives? How do infants manipulate these representations in the absence of continued perceptual input? How might infants make new connections between representations, learn new concepts, or think new thoughts? To what extent might these early capacities form the basis of more complex thought or the acquisition of new knowledge domains (like physics or algebra)? Answering these questions requires formulating hypotheses around LoTs.

As a case study, consider the research on infants' capacity for “arithmetic.” In a now classic paper, Wynn (Reference Wynn1992) found that infants who were shown objects hidden one at a time behind an occluder were able to represent the total quantity of objects hidden. Wynn (Reference Wynn1992) suggested that infants' success was evidence that they grasp the numerical relationship between the inputs and outputs of an addition computation. This is a provocative suggestion – that infants have an LoT-like capacity for combinatorial thought over numerical representations – and the LoT framework allows us to set up testable conditions under which such capacities might be evidenced. Infants could be doing something that resembles the expressive power of arithmetic: Their representations of the objects could be formatted in a way that allows those representations to be used as operands in a mental function specifying arithmetic relations (i.e., f([object a], [object b]) = a + b = c) – consistent with an LoT. Or, infants could be doing something that is decidedly not LoT-like at all: For example, they could track the two sequentially hidden individual objects via separately deployed attentional indexes resulting in a representation of [index, index], and the total quantity is represented only implicitly by the number of indexes deployed. In fact, we do not yet know which of these potential explanations (if either) underlies infants' behavior in Wynn (Reference Wynn1992) (see Cheng & Kibbe, Reference Cheng and Kibbe2023, for related discussion). Identifying to what extent infants' early capacities have a computational structure or combinatorial capacity similar to formal arithmetic is particularly important because we care not only about what's going on in the infant mind, but also about whether formal numerical knowledge can emerge from infants' early capacities (see, e.g., Carey, Reference Carey2009). This is just one example, but there are many such LoT hypotheses out there just waiting to be tested.

For developmentalists who may be hesitant to explicitly formulate LoT hypotheses about infant cognition, Quilty-Dunn et al.'s approach to LoT has major advantages. It does not commit infant researchers to a single format for an LoT, into which disparate evidence must be proverbially crammed. Their approach also does not commit us to LoTs with full, recursive, natural-language-like expressive grammar, which would be difficult to square with infants' apparent capacities in a variety of domains. And their approach does not commit developmentalists to Fodor's radical nativism, which in the past has (somewhat unfairly, I would argue) marked the LoT hypothesis as incommensurate with development.

In fact, an LoT approach is developmental. Taking an LoT approach to infant cognition is compatible with efforts to understand how motor development, neurobiological development, and/or individual differences related to cognitive, emotional, educational, economic, or sociocultural factors may shape cognition in infancy and beyond. Indeed, it allows us to formulate hypotheses to identify potential computational structures over which these factors may operate across early development. It allows for the possibility of differential developmental trajectories for LoTs across domains and across infancy.

Taking an LoT approach also does not require that infants' early computational capacities must be quarantined from experience or learning (e.g., I think positing that LoT structures may be identifiable in infancy is compatible with a rational constructivist approach; see Xu, Reference Xu2019). Nor does it require us to find evidence for LoT(s) in every aspect of early cognition. There are plenty of instances in which iconic, noncombinatorial representations provide the best explanation for infants' (and, indeed, adults') behavior in some domain. Instead, it requires infant researchers to take seriously the possibility that infants can think before they articulate language, and to identify cases of such expressive capacities, and their functional utility for the developing mind.

Importantly, taking an LoT approach also does not entail that there is some developmental hierarchy in the expressive power of LoTs, with infants on the bottom and adults at the top. Infant researchers can and should formulate hypotheses around both continuity and change in the computational capacities of minds across domains and across development.

Quilty-Dunn et al. lay out the case for LoT-like structures in the mind, and a roadmap for how to go about looking for them. If human cognition includes LoTs in its fully developed state, then we need to formulate our hypotheses about the origins and development of human cognition within an LoT framework. Cognitive developmentalists should embrace this approach.