The target article provides a welcome corrective to many of the over-hasty conclusions that have been drawn about the cognitive abilities of nonlinguistic creatures. Penn et al. provide detailed critical analyses of many of the experiments routinely cited as evidence of cognitive commonalities between human and nonhuman animals. As the authors point out, many of these experiments warrant a far more parsimonious interpretation than they generally receive. The case that they make for a profound discontinuity between the human mind and the nonhuman mind is powerful indeed (although I imagine that other commentators will have much to say about the details of their arguments).
Nonetheless, the authors are categorical in rejecting the thesis that nonhumans are purely associative, stimulus-response machines. They claim (surely correctly) that nonhuman animals are capable of representing particular objects in ways that allow those objects to be re-identified and that involve sensitivity to some of the relations in which those objects can stand. From this it follows, they claim (again, surely correctly) that the cognitive abilities of nonlinguistic creatures must be compositional and systematic in some substantive sense. They propose that the representational systems of nonlinguistic creatures are functionally compositional and featurally systematic. These systems are functionally compositional in that they can represent structured entities in such a way that there are effective and reliable processes both for building representations from their constituents and for decomposing representations into their constituents. Representations can be functionally compositional without being concatenatively compositional – that is, without being literally built up in a way that reflects the structure of what they represent. Nonhuman representations are featurally systematic in that they permit a limited form of recombination: “The kind of systematicity manifested by nonhuman animals is limited to perceptually based relations in which the values that each argument can take on in the relation are constrained only by observable features of the constituents in question” (sect. 10.6, para. 4). This is much weaker than the much discussed classical notion of systematicity discussed by authors such as Fodor and Pylyshyn (Reference Fodor and Pylyshyn1988).
Suppose that this is all true as a description of the differences between human and nonhuman cognition. It raises two obvious questions:
1. What is the relation between functional compositionality and featural systematicity?
2. What explains the fact that animal representational systems cannot be concatenatively compositional and classically systematic?
The relational reinterpretation hypothesis is intended as a first step in answering both these questions. Penn et al. write, “The discontinuity between human and nonhuman minds extends much farther: to any cognitive capability that requires reinterpreting perceptual relations in terms of higher-order, structural, role-governed relations” (sect. 11, para. 4). The basic idea is that nonhuman animals are capable of identifying and reasoning about certain classes of objects, properties, and relations – namely, those that can be identified through perceptible similarities and perceptible statistical regularities. What characterizes human cognition is the ability to represent higher-order, structural relations that, for example, permit objects to be classified in terms of their abstract roles and functions.
As far as the first question is concerned, the relation between functional compositionality and featural systematicity is that both are required for sensitivity to perceptual relations. And, with regard to the second question, concatenative compositionality and classical systematicity are ruled out for nonlinguistic creatures because they involve relational reinterpretation. The first answer has some plausibility. Sensitivity to perceptual relations requires the ability to track objects and to have minimally structured representations. The second answer, however, is less persuasive. The problem is that the relational reinterpretation hypothesis seems to only redescribe the phenomena it is trying to explain. It is hard to see what the difference is among the following:
(a) Being able to represent higher-order relations, abstract roles, and functions (the capacity for relational reinterpretation)
(b) Having representations that have components corresponding to higher-order relations, abstract roles, and functions (concatenative compositionality)
(c) Having representations that permit constituents to be recombined in ways constrained only by the abstract roles and functions of those constituents (classical systematicity)
If there is any difference at all between (a), on the one hand, and (b) and (c), on the other, it is simply that (b) and (c) are special cases of (a). This does not really help us to see why (b) and (c) are uniquely human cognitive achievements.
What would be needed to turn a redescription into an explanation is an account of why exactly (a) is unavailable to nonhuman animals. The authors do not have much to say about this. The LISA (Learning and Inference with Schemas and Analogies) model of relational reasoning that they propose at the end is a model of the mechanisms that enable (a), but again the model appears to recapitulate the basic description rather than do much explanatory work. As the authors describe it, LISA “implements the distinctive higher-order relational capabilities of a PSS [physical symbol system] via an additional representational system that has been grafted onto a simpler system … [providing] conjunctive representations that are functionally, but not concatenatively, compositional” (sect. 11.1, para. 1). Of course, this is interesting and important. But it fails to address the fundamental questions of why the representational systems of nonhuman animals must be of the “simpler” type and what it is about humans that allows them to have the “additional” system grafted onto the “simpler system.”
One obvious way of answering these questions is to highlight the distinctiveness of human linguistic abilities – either by way of the “rewiring hypothesis” proposed by Dennett (Reference Dennett1996), Mithen (Reference Mithen1996), and Bermúdez (2003; Reference Bermúdez2005) or by Carruthers's appeal to the role of representations in logical form in domain-general, abstract thinking (Carruthers Reference Carruthers2002). Penn et al. reject these proposals. Whatever their ultimate merits, however, these proposals quite plainly offer explanatory hypotheses. If Penn et al. are to offer a genuine alternative, they need to make clear just how their account is an explanation of the uniqueness of human cognition, rather than simply a description of that uniqueness.
