Object lessons: Spelke principles and psychological explanation

Abstract

There is general agreement that from the first few months of life, our apprehension of physical objects accords, in some sense, with certain principles. In one philosopher's locution, we are ‘perceptually sensitive’ to physical principles describing the behavior of objects. But in what does this accordance or sensitivity consist? Are these principles explicitly represented or merely ‘implemented’? And what sort of explanation do we accomplish in claiming that our object perception accords with these principles? My main goal here is to suggest answers to these questions. I argue that the object principles are not explicitly represented, first addressing some confusion in the debate about what that means. On the positive side, I conclude that the principles supply a competence account, at Marr's computational level, and that they function like natural constraints in vision. These are among their considerable explanatory benefits—benefits endowed by rules and principles in other cognitive domains as well. Characterizing the explanatory role of the object principles is my main project here, but in pursuing certain sub-goals I am led to other conclusions of interest in their own right. I address an argument that the object principles are explicitly represented which assumes that object perception is substantially thought-like. This provokes a jaunt off the main path which leads to interesting territory: the boundary between thought and perception. I argue that object apprehension is much closer to perception than to thought on the spectrum between the two.

Acknowledgments

Many thanks to José Bermúdez, Jonathan Cohen, Frances Egan, Robert Matthews, Robert McMurray, Brian Scholl, and three anonymous referees of this journal for helpful comments. Thanks also to Jerry Fodor and Philip Robbins for helpful discussion, and thanks to Philip for supplying the catchy part of the title.

Notes

 For example, if infants perceive objects in accord with Cohesion, then if they see no spatial separation between two stationary surfaces, they infer that the surfaces make up a single object—even if the surfaces differ in color and texture.

 In the canonical example, formulas that belong to an axiom set's deductive closure are implicitly represented in the set. The idea that what is explicitly represented is more readily accessible than what is implicitly represented is also found in the popular associations between being explicit and being conscious and verbalizable.

 To my knowledge nobody has actually made this argument.

 Her arguments for this claim were justified in part by the observations that object representations are unlike perceptual representations in being persistent, and free of commitments to any specific sensory modality.

 This pronouncement is provoked by Dennett's objection to RTM that the contents of beliefs needn’t be explicitly represented any more than the instruction to ‘get your queen out early’ in a chess program (Dennett, 1981). Such a rule may, he observes, just fall out of the design of the program without corresponding to any particular line of code.

 A fuller list might include the notion that concepts are the basis of word meanings, and that thoughts are structured, obeying something like Evans’ Generality Constraint (Evans, 1982).

 Another objection: what about the concept BLUE? First, notice that this is no counterexample to the claim that conceptual representations tend to be cross-modal or amodal. Second, compare BLUE to a perception of blue: the former is available for combining with other concepts into thoughts, which are not modally specific—say, upon hearing my alarm I rolled over and saw that the sky was a bright blue.

 Bob McMurray is the author of this objection.

 What is more, judgments of category membership can be very fast and effortless, but they can also be slowed down, as in tricky borderline cases; and there is no reason to think that object perception may be slowed down in this way.

 The full story about encapsulation is more complicated, to be sure. For example, at 12 months, infants start regarding the duck–truck as separate objects. This result and other similar ones suggest, perhaps, that object perception becomes less encapsulated as development progresses.

 On the other hand, they endorse Scholl and Leslie's claim that infant object representations are representations of mid-level vision. They call these representations ‘object files’. Their files differ from Scholl and Leslie's indexes only in that they possess Carey and Xu's concept-making features.

 It was their discussion, however, that alerted me to the need for such a list; for this I am grateful.

 This is illustrated by Stich's example. Let r be a rule of syntax. On the one hand, you believe that if r, then Chomsky is mistaken. On the other hand, you have r stored in your language faculty—so you know it not in the special linguist's way, but just like everyone else. But because r is trapped inside your language faculty, it will not lead, in combination with your belief, to the conclusion that Chomsky is mistaken. If you instead believed r as the linguist does, then you would probably draw the conclusion that Chomsky is mistaken, as beliefs tend to generate further beliefs via inference.

 Scholl and Leslie acknowledge this in passing, but not quite as forcefully as I think is appropriate. In personal correspondence, Brian Scholl has amplified this acknowledgment.

 In the first category we find demonstrations that young infants use common motion, but not static featural information, as a clue to object-hood (Spelke, 1988). In the second category we find demonstrations that very young infants are sensitive to the heights of objects, contrary to what Indexing would predict (Baillargeon & Graber, 1987).

 One might instead defend this sort of picture by appeal to Occam's Razor, as follows. If we ought to explain cognitive capacities by the simplest means available, and if unthought-like mechanisms are assumed to be simpler than thought-like ones, then we should explain a given cognitive capacity with the least thought-like mechanism that is adequate to the task. This line is not without adherents (cf. Pylyshyn, 1999), but it has vulnerabilities. One might object to any appeal to simplicity in psychological explanation, on grounds that Nature the Tinkerer can hardly be expected to have done things in the simplest way. Or one might object that it is far from obvious that thought is more complicated than perception (cf. Bermúdez, 2003; Spelke, 1988); or that a single powerful mechanism would be more parsimonious than a patchwork of weak ones.

 One might ask here whether there is any meaningful difference between object apprehension by a confederacy of dunces or by one very clever mechanism: might not a clever mechanism be realized by a confederacy of dunces? Yes, but this is beside the point. Generally speaking, a confederacy as a whole may possess marks of thought or lack them, depending on engineering details. But I argued that the mechanisms of object perception taken together, whatever they are—not just the Indexing mechanism—appear to lack the marks of thought (§4.2). So in this case, the confederacy as a whole lacks the marks of thought.

 Many thanks to an anonymous reviewer of this journal for helping me clarify my strategy here.

 Notice that while Spelke's (1988) view and Indexing differ in what they say about performance—one has it that the object apprehension is underwritten by thought-like mechanisms, the other not—on both views our object-apprehending competence is captured by the Spelke principles.

 See Matthews (1991) for a strong case for the sort of interpretation of Chomsky that I recommend.

 As one might expect, the full story about the rigidity constraint is actually more complicated: human beings are able to construct interpretations that are consistent with elastic motion, for instance. The rigidity constraint must be qualified (Hildreth & Ullman, 1989).

 Natural constraints are generally understood to be at level 1, so this argument that the S-principles are like natural constraints offers further support for (c).