CORE PHYSICAL KNOWLEDGE AND THE MARKS OF THOUGHT

Core Physical Knowledge and the Marks of Thought

1. Introduction

There's good evidence that we represent individual objects starting in the first few months of life. From the age of about 2 months, infants have some capacity to track objects, and they have certain expectations about how they will behave and move. Over the past 15 years or so, much excellent research has explored the questions of what infants know about objects, when they know it, and how their knowledge should be characterized. On one view, infants are seen as possessors of an intuitive physical theory, whose expectations about objects manifest genuine understanding [Spelke 1988]. On a second, some of the infant object data is seen as resulting from the operation of a mechanism of visuospatial attention [Scholl and Leslie 1999]. On a third, the representations of that mechanism are still identified with infants' object representations, but are taken to have more in common with conceptual representations [Carey and Xu 2001]. On all these accounts, the infant's physical knowledge is taken to form the "core" of the adult's, and much of that core is taken to be innate [cf. Spelke 1994].

Here I'll focus on the second and the third accounts. I'll find that some key contrasts that have been claimed between them are in fact spurious. Sorting that out will lead to discussion of the differences between perceptual and conceptual processes and representations, which will provide me with the materials for laying out my own view of object apprehension. I'll start with some basic data on infant object cognition.

2. Some Basic Data

Based on numerous experimental inquiries into the physical knowledge of infants—most of them on the "preferential looking" model—Elizabeth Spelke has developed a case that infants perceive and reason about objects in a way that "accords with" a small set of principles:

i. Continuity. Objects exist continuously and move on paths that are connected over space and time.

ii. Cohesion. Objects are cohesive: they are internally connected and externally bounded entities that maintain both their connectedness and their boundaries over time and space.

iii. Contact. Objects influence each others' motions if and only if they touch.

Consider the principle of Continuity (the evidence for which is represented schematically in Fig. 1). Infants were habituated either to continuous motion of an object behind two parallel occluders ((a), continuous condition), or to discontinuous motion behind these occluders ((b), discontinuous condition). Habituation was followed by two trials: one replicated the motion of the continuous condition, but without occluders ((c)), while the other replicated the motions of the discontinuous condition, also without occluders ((d)). Infants in the continuous condition looked longer at the latter trial, while infants in the discontinuous condition looked longer at the former. Making the usual assumptions about preferential looking, it thus appears that if infants observe continuous motion, they expect one object; while if they observe discontinuous motion, they expect two [Spelke, Kestenbaum, Simons, and Wein 1995]. Accordance with the principles of Cohesion and Contact is supposed to entail a disposition to make certain inferences in response to certain cues, as in the case of Continuity.[1]

Another study that will be particularly important here concerned cross-modal transfer of information about an object. It produced evidence that infants can pick up information from what they feel which then gives them certain expectations about what they’ll see. Infants found out by manual exploration that two objects—ends of a sort of rattle—are connected (see Fig. 2). In one habituation condition ((a)), they were connected by a rigid bar, while in the other ((b)) they were connected by an elastic band that permitted independent motion. They handled these objects under a cloth, without being able to see them. They were subsequently shown a visual display of either a rigidly connected object ((c)), or two entirely disconnected objects ((d)). In the rigid motion condition, infants show a looking preference for the disconnected ones, suggesting that they expected a proper, connected object; while in the independent motion condition they showed the opposite preference (see Fig. 2) [Streri and Spelke 1988]. This suggests that object representations reside in a "cross–modal space"—at least, a space in which visual representations may engage with haptic ones.

3. Object Indexing

On the Object Indexing theory (OI), object perception is seen as underwritten by no genuine thinking, but rather by a fairly simple visual mechanism [Scholl and Leslie 1999]. The key idea in this model is that of an object index: a representation of mid–level vision, and a "mental pointer"—a sort of mental demonstrative—which serves to reference an item in the visual field without representing any of its distinguishing features. The assignment of indexes to items in the visual field depends only on spatiotemporal information—about the location, boundaries, and continuous motion of objects. Once assigned to an item, an index will 'stick': it will move with the object, and it will survive brief occlusion.

Scholl and Leslie propose that some of the data on infant object cognition may be explained by the Indexing mechanism. To see how such an explanation would work, consider again the evidence for Continuity. In the continuous condition, an object is seen to move between the two occluders. Here one index is established when the object first appears, which index survives occlusion and "re–acquires" the object upon its reemergence. Longer looking occurs in the two–object than the one–object trial because "increased attention is allocated in the form of a new index": some processing time is associated with the assignment of a new index. In the discontinuous condition, the first index just dies instead of re–acquiring the object, as it cannot jump the gap over which no motion occurs, so a second index is assigned during habituation. Infants look longer at the one–object test in this condition because "increased attention is allocated to search for the 'missing' object corresponding to the original" [Scholl and Leslie 1999, 54–55].

4. The Object File System

The key idea of the previous account, that object perception is underwritten by a mechanism of visuospatial attention, has been taken up and developed by Susan Carey and Fei Xu [Carey and Xu 2001]. However, they emphasize that object files have certain things in common with conceptual representations.[2] Already in OI we find that object indexes are unlike perceptual representations, occupying some sort of middle ground between perception and conception, in that they survive occlusion and individuate objects. Carey and Xu suggest that object files are closer to the conceptual side of this middle ground. I'll call this the Visual Concept account (VC).

5. How is an object file like a concept?

Carey and Xu suggest that infant object representations possess features that should lead us to regard them conceptual in certain respects. In particular, they suggest that an object file has both the wide and the narrow content one would expect from the concept PHYSICAL (SPELKE) OBJECT: the wide content of an object file is the property of Spelke–objecthood, while its narrow content is given by a conceptual role appropriate to the concept SPELKE OBJECT: it "articulate[s] physical reasoning, enter[s] into number–relevant computations, and support[s] intentional action" ([Carey and Xu 2001], 28). I think Carey and Xu fail to articulate any representational features that are distinctively conceptual.

5.1 Carey and Xu's Case

At issue in the claim about wide content is, presumably, the standard notion of informational content: one thing contains information about another if there are causal or counterfactual dependencies between the states of one and the states of the other. Here Carey and Xu are concerned to demonstrate that the Indexing system tracks proper Spelke objects (bounded, 3D, etc.) rather than a wider class of things encompassing, for example, 2D blips on a screen and piles of sand. This is worth noting, and Carey and Xu make a good case; but since informational content may be carried by things other than concepts—perceptual representations and thermometers, for example—it is no reason to think that object files are conceptual. Thus, any concept–making features would have to reside in the alleged narrow content of an object file.

The narrow content of a concept, if there is such a thing, is a matter of which computational processes it figures in. Carey and Xu give two reasons to suppose that object files have a specifically conceptual role: object files "guide volitional action", and they "articulate physical knowledge". It is difficult to make out what is meant by these claims, and what the arguments for them are supposed to be. I shall try my best.

The claim that object files guide volitional action in a way that suggests they are concept–like is based on the following data. First, at 8 months infants can retrieve objects hidden under cloths or behind barriers, suggesting that they have enduring representations of these objects. Second, it's been found that 10–month–old infants who witness 1 cracker being placed in one box, and 2 in a second—or 2, then 3—will crawl towards and reach for the box with more crackers in it [Feigenson et al 2001]. Carey and Xu claim that in order to do that, infants must have a representation of one set of objects stored in memory with which they compare the one they are currently attending to. They conclude from all this that object files—currently active ones, and ones stored in short–term memory—"guide actions directed towards the physical world", and that "[i]nsofar as being available to guide volitional action (informational promiscuity) is evidence that a representation is conceptual, these studies suggest that object files are" (208, emphasis added).

Now, all that Carey and Xu do by way of characterizing this "guidance" and "informational promiscuity" is to point to the presumed role of object representations in the studies just cited. But I presume, based on the label and that limited characterization, that what is meant by "informational promiscuity" is something like inferential promiscuity, a label Stephen Stich has given to the feature whereby a belief may play a role in an inference to any other belief [Stich 1978].[3] The basic idea of the latter is that of presence on a central mental stage, where representations may mingle freely. An inferentially non–integrated or encapsulated belief, by contrast, fails to combine with beliefs outside the module wherein it is housed, or the limited set of modules with which that module may exchange information.

I take it, then, that Carey and Xu are suggesting something like this. The data just described suggests that a representation of a set of objects may move from an attentional register to a memory register, and (in the Feigenson studies, at least) then combine freely with what next appears in that register. In claiming that object files are promiscuous, Carey and Xu suggest that these representations are generally available, rather than available to only a few processes.

As for the claim that object files articulate physical knowledge, Carey and Xu suggest that infants' knowledge of physical objects is considerably richer than what Indexing can account for, encompassing features about which that framework says nothing; and that this very richness suggests that the mechanisms involved in object apprehension are more thought–like than that framework acknowledges. More specifically, they point to findings indicating that very young infants expect objects to be solid, and to be subject to the laws of contact causality [Spelke et al 1995]; and that infants learn, in a series of well–defined stages, that unsupported objects fall [Baillargeon 1995]. Findings like these suggest, according to Carey and Xu, that "—the conceptual role of the infant's object concept is that of 3D Spelke–objects; objects are represented as solid entities in spatial relations with each other that cannot pass through other objects, and which move only upon contact” (209, emphasis added).

5.2 Carey and Xu's Concept–Making Features

About each of Carey and Xu's alleged conceptual features we can ask whether the data offered in its support really suggest that infant object representations possess it, and whether it is in fact a plausible mark of concept–hood.

The first feature, that of "guiding volitional action" was a matter of "informational promiscuity"—which was, as far as one could discern, a matter of a representation's being available to a large range of processes, rather than just those belonging to a module. Are infant object representations in fact promiscuous? Promiscuity is a matter of degree. I'll argue that very many perceptual representations are promiscuous to the same degree that Carey and Xu have shown object representations to be; so promiscuity to just that degree does not deserve a place in a good list of concept–making features.

The promiscuity of object files is supposed to evident when infants are able to compare a representation of an object–set stored in memory to one in the current attentional register. Compare a hypothetical experimental task: you are shown two color tiles, and asked, after the first is put away, if the second is brighter. You are able to make such comparisons. Does the fact that a representation of the first ends up in a memory register suggest that your representation of the first color is conceptual—or, conceptual to a certain extent? I think not. More likely, you have, on the one hand, percepts of two color tiles, and on the other hand, memories of them—representations in a short–term memory register. And yet, it was the capacity of an object representation to remain in a memory register that earned it the charge of informational promiscuity. Thus, being just that promiscuous—as much as Carey and Xu have shown object representations to be—is not a good mark of concept–hood, as it is a feature possessed by many perceptual representations.

Now consider the claim that infant object representations "articulate physical knowledge" that is not accounted for by Indexing. It is clearly true that the strict Indexing picture says nothing about infants' knowledge of solidity or contact mechanics, nor about their apparent ability to learn generalizations about objects. But why presume that the computational role which gives rise to their expectations about solidity and contact–mechanics is a conceptual role? There's nothing to preclude there being various other automatic, unthinking processes, of roughly the same modular sort as Indexing, that accounts for this additional data. Infant object representations are plausibly involved in a richer set of expectations than Indexing accounts for, but the fact that they are does not suggest that those representations are conceptual to that extent.

In sum, Carey and Xu have not described a role for object files that's specifically conceptual, rather than merely functional or computational; that's why there's nothing preventing a perceptual representation from playing the same role. No clear mark of concept–hood has emerged here.

6. A Clearer View of the Landscape

Having considered two leading accounts of infant object cognition currently on the market, we face an obvious question: What's the true story about object apprehension? Everyone in the debate holds that this capacity falls somewhere between the two poles of thought and perception—but the features that define these poles have not been laid out. I shall list some features plausibly belonging to thought and conceptual representation, and consider which of these are possessed by object apprehension, as best we can tell. The results I get will then guide my own synthesis of an account. That synthesis will also be guided by the general principle that we should get away with as little full–blown thought as possible in psychological explanation: a given cognitive ability should be explained with the least sophisticated and powerful mechanism that is adequate to the task, for the familiar reason of theoretical parsimony [cf. Pylyshyn1999].

6.1 Some Marks of Thought

As psychologists generally use the term, ‘concept’ denotes a building block of thought—a mental representation of a category, from which thoughts are constructed. The features on the list below are very widely accepted as marks of conceptual processes and representations. For this reason, and also for reasons of space, I shall defend each one only briefly. Also, none of them is meant to be a necessary or sufficient condition for conceptual representation or for thought, taken individually; neither is the list intended to be entirely comprehensive.[4] Still, they are useful diagnostic criteria.

a. Categorization. Concepts are the mental representations of categories; thinking involves the placement, by some mental means or another, of objects into categories.

This is just what many psychologists are talking about when they talk about concepts.

b. Concepts are inference-licensing. For instance, CHAIR licenses inferences to CAN BE SAT UPON, so the thought that that's a chair licenses the inference to that can be sat upon; and similarly for that’s a dog and that has fur, etc.

Again, a concept is a mental representation of a category, and by mentally categorizing, we license inductive inferences [see, for instance, Smith 1995].

c. Amodal or cross–modal representation. Perception tends to involve modally specific representations (visual, auditory, etc.) while thought tends to involve representations that are modally unspecific—or at least, processes in which representations belonging to different modalities may engage with each other.

It will be objected that some early-perceptual representations are in fact cross-modal: witness, for example, mirror neurons which integrate visual and proprioceptive information [Rizzolati et al 2001], and the computation of location by the superior colliculus from inter-aural time differences plus visual information [McMurray 2002]. But compare the degree of independence from modal commitments possessed by the representations generated by such neurons with that possessed by, for example, the concept GOOD. This thought-symptom, like the others, is a matter of degree.[5]

d. Independence. While perceptual representations are fleeting, conceptual ones carry a perceiver beyond her immediately perceivable surroundings; they possess a certain independence from the here and now—in surviving occlusion, for instance [Spelke 1988], and in being deployable at will.

It might be objected that survival through occlusion is found even in very low-level visual representations, e.g. those that survive the momentary blindness (~20 ms) that we experience during a saccade.[6] This is indeed notable, but we can distinguish different degrees of independence corresponding to different lengths of survival. Regarding the second sense, compare APPLE to a perceptual representation of a particular apple. The former can be called up at will in the course of deliberative thought processes (Might that red thing be an apple? Did the victim consume a poisoned apple?), while the latter cannot.[7]

e. Fodorian Centrality. Central processes, which are responsible for the formation of beliefs, are domain–general, relatively slow, non–mandatory, and informationally unencapsulated [Fodor 1983], i.e. they have access, in principle, to everything the organism knows.

Fodorian modules, on the other hand, are domain–specific and fast; their operation is mandatory or automatic; they are informationally encapsulated, having access to only a small subset of the organism's knowledge; and their computations are inaccessible to central processes.[8] Fodor has argued with vigor that central processes have a key role in our cognitive architecture [Fodor 1983, 2000].

The following objection might come from “massive modularist” quarters. There are no processes that are central in Fodor’s sense; we can account for the domain-generality, slowness, etc. of our thinking with recourse only to modules, if we allow that there are “conceptual” or “central-process” modules. Thus, object apprehension’s lacking this alleged ‘mark’ would be no indication of its non-thought-likeness. But this objection has rather limited force. Sensible proponents of modular thought hold that to account for all the phenomena of our higher flights of thinking—the slowness of deliberation in chess, the domain-crossing leaps of scientific inference—we must advert to something beyond the individual functioning of conceptual modules [Sperber 1994, Carruthers 2003].[9] Assuming this moderate view, a process fitting the profile of Fodorian modularity may be classed a species of thought, but its thoughtfulness would be of a lower grade than those higher flights.

6.2 Object Apprehension and the Marks of Thought

Which features on the list above are possessed by the mechanisms of object apprehension? Regarding Categorization, we encounter the difficulty that although concepts are very generally taken to be implicated in categorization, "categorization" can be understood in a number of ways. Some categorization tasks in which concepts have been implicated are "theory-based", such as those in which an altered, raccoon-looking skunk is classed as a skunk rather than a raccoon. "Similarity-based" categorization, on the other hand—judging whether a visually presented item belongs to a specified category, for instance—is much faster and less reflective [Smith 1995]. Might the latter sort of task resemble processes involved in infant object apprehension?

According to a popular and plausible story, a visual item is classed as a dog or not via a process of matching against a prototype, stored in memory, and built up by various experiences of doghood. If the item possesses a criterial number of doggy features, then the concept DOG is tokened. Consider, on the other hand, that the processes involved in infant object apprehension are generally supposed to have a substantial innate component. While it’s plausible that we have built in some general-purpose mechanisms for constructing prototypes, the content of DOG itself is not plausibly built in; and there's no evidence that anything with prototype structure, built up in memory, is implicated in young infants' object apprehension.[10] Thus, with respect to the roles of learning, memory, and prototype structure, such “categorization” as is involved in infant object apprehension differs from that associated with concepts like DOG.[11]

There is a very broad similarity between classing an item as a dog and apprehending an object. In both cases, we see the following sequence of events. Some mental mechanism is engaged in response to certain cues (continuous motion, cohesive boundary; fur, 4 legs). Once this happens, certain expectations are formed; certain inferences, of a sort, are licensed (one object rather than two; barks, wags tail). Thus we can count object apprehension as a species of categorization, broadly construed. But if it is so construed, the visual system’s generation of rigid object-percepts (§3.2) also counts as a species of categorization. Thus ‘categorization’ so broadly construed has limited utility as a mark of thought, though it may mark a landmark on the road from perception to thought.

Regarding the licensing of inferences, we can again read this feature broadly enough that it applies to object apprehension, but as in the previous case we are then left without a real symptom of thought. There is some sense of “inference” that we may reasonably apply to processes of object apprehension: we might say that from continuous motion, the presence of one object is inferred; and from separate motion, felt under a cloth, the presence of two objects is inferred; etc. However, perception has often been regarded as inferential in a sense. For example, Helmholtz held that we unconsciously infer the size of objects from the size of the retinal image, plus depth cues, together with a rule we've learned associating the two [Gleitman 1987]; and on the “constructivist” view of perception popular in cognitive science, we put together many small physical signals, together with “assumptions” made by our perceptual systems, such as the rigidity constraint, to arrive at a perceptual “conclusion” [Van Leeuwen 1998]. Clearly, it would be too liberal to regard this kind of inference as a symptom of thought. And why not suppose that such inferences as occur in object apprehension are of this general sort?

As for cross-modality, we have on the one hand the data from Streri and Spelke that infants’ object representations combine visual and haptic information (§2.1); and on the other hand the findings that some early-perceptual representations are cross-modal (§5.1). I contrasted the independence from any particular modality that such representations enjoy with that possessed by the concept GOOD. The case of object representations resembles that of the former representations more than the latter. Here again we distinguish a broad reading of this ‘mark’, viz. cross-modal to some degree or other, which does apply to object apprehension but is too broad to be diagnostic.

I distinguished two sorts of independence from immediately perceptible surroundings: survival through occlusion, and deployability at will. Object representations certainly exhibit the first to a degree, and their survival over tens of seconds sets them apart from, for instance, the early-visual representations that survive for about 20 ms during a saccade. As for the second, there is no evidence that infant object representations are deployable at will.[12]

Finally, object apprehension fits the profile of modularity better than that of Fodor-centrality. Consider domain specificity. Fodor argues that an "eccentric stimulus domain"—one whose elements exhibit idiosyncratic properties, like those of sentences or human faces—is likely to be processed by a cognitive mechanism that is highly specialized [Fodor 1983]. The Spelke–principles define quite a specific domain. Regarding mandatoriness, we have no choice about whether we parse the world into objects: we cannot "turn off" object perception, and choose instead to see the visible world as a mosaic of surfaces. And object perception appears to be fast and effortless.

Are the inferences involved in object perception informationally encapsulated? There’s no reason to think that everything known to an infant is potentially relevant to her inferring, for example, that two surfaces belong to a single object. It seems, rather, that only very specific kinds of cues—"common fate" or adjacency, for example—are relevant to forming this expectation. Thus, we can explain the expectations that infants have regarding objects without adverting to any unencapsulated inferences. Unencapsulated inferences are slower and more computationally expensive than encapuslated ones, involving the trawling of large bodies of information [Fodor 1983]. By the general principle of parsimony endorsed above, then, we should assume that object apprehension is encapsulated.

One might object here that the cross-modal character of infant object representations suggests that there is a central stage on which representations from different modalities may mingle; so the match with Fodorian modularity is not as good as I make out. But there are other ways to account for this transfer of information—ways that are likely to be employed in the cases noted earlier of cross-modal perceptual representation. Regarding visuo-haptic object representation, there could be a hardwired pathway from the haptic modules involved in processing the data of separate motion to the visual modules involved in object apprehension, such that evidence of separate motion in particular, from that particular source, is 'admissible' (processible) in making contact–mechanical judgments about objects. We have no reason to believe that evidence of separate motion coming from any source would be admissible, nor that evidence about arbitrary features of objects is admissible; thus no central stage is called for.

While the mechanisms of object apprehension do look to be modular rather than Fodor-central, it is important to keep in mind that this does not itself imply that they are not mechanisms of thought, if the proponents of modular thought are right. However, on at least one prominent picture of modular thought, thought-modules are distinguished from perceptual ones by taking conceptual inputs [Sperber 1994]—and there is no reason to think that the mechanisms of object apprehension do that.

In sum, object apprehension is not looking very thought–like: with respect to my list of thought-features it has nearly struck out. It emerged that it was necessary to refine several of these ‘marks’ in order to distinguish a genuine symptom, as opposed to a feature belonging to any number of perceptual representations. Further refinements would undoubtedly make a better list.

A final note regarding the commonalities between object apprehension and thought: there is some considerable evidence that various non–human primates, including non–great apes such as cotton top tamarins and rhesus monkeys, have many of the same object–tracking abilities as young infants [Hauser et al 2000, Sulkowski and Hauser 2001]. If the mechanisms of object apprehension were continuous with those of thought, then this would be surprising, as the primates in question are very different from us with respect to their capacity for thought.

7. Object Indexing Plus

The foregoing results suggest that some version of Indexing, rather than a view tending more toward the conceptual end of the spectrum, is the best game in town. OI has the advantage over VC of not ascribing to object indexes any spurious conceptual features. But Indexing is at best only part of the story of object apprehension.[13] For one thing, this mechanism furnishes no account of the cross-modal character of object representations. These representations cannot be exhaustively accounted for just in terms of a visual mechanism.

Indexing stories have been proposed about only about a narrow range of the object cognition data. Some other studies seem amenable to an Indexing explanation, but some, like the ones involving cross–modal transfer, do not.[14] There's no reason to expect that the mechanisms underlying the latter will be any more thought–like than Indexing—taken individually, at least. But Indexing plus a haptic component of object representation yields something that is more thought–like in toto than Indexing, in that it is freer of modal commitments.

The most plausible picture, then, is one of a patchwork of mechanisms, including Indexing, each one fitting the same generally modular profile of Indexing, cooperating in such a way as to yield a total object–apprehending apparatus that is more thought–like than its individual parts, but still a far cry from Fodorian central processing. Importantly, the proposed patchwork does not allow the straightforward identification of infant object representations with object indexes, even for young infants—pace both OI and VC. Infant object representations result from the functioning of various mechanisms, from whose design it falls out that objects obey the Spelke principles. To account for studies suggesting that young infants have knowledge of contact–mechanics and solidity not accounted for by Indexing, Carey and Xu concluded that indexes themselves have a more complicated role than is envisioned by OI, and that this role is conceptual. I propose instead that distinct, coordinated mechanisms are responsible for some of this contact–mechanical knowledge. Distinct mechanisms are needed anyway, in order to account for cross–modal transfer.

References

Baillargeon, R., 1995: "A Model of Physical Reasoning in Infancy". In C. Rovee–Collier and L. Lipsitt, eds., Advances in Infancy Research (Vol. 9, 305–371). Norwood, NJ: Ablex.

Baillargeon, R., and Graber, M., 1987: "Where is the Rabbit? 5.5–month–old Infants' Representation of the Height of a Hidden Object". Cognitive Development 2, 375–392.

Bogartz., R., Shinskey, J., and Speaker, C., 1997: "Interpreting Infant Looking: The Event Set x Event Set Design". Developmental Psychology 33.

Bresnan, J., and Kaplan, R., 1982: Introduction: Grammars as Mental Representations of Language. In J. Bresnan, ed., The Mental Representation of Language. MIT Press.

Carey, S., and Xu, F.. 2001: "Infants' Knowledge of Objects: Beyond Object Files and Object Tracking". Cognition 80, 179–213.

Carruthers, P., 2003: "The Cognitive Functions of Language". Forthcoming in Behavioral and Brain Sciences.

Chomsky, N., 1965: Aspects of the Theory of Syntax. MIT Press.

__________, 1980: Rules and Representations. Columbia U. Press.

Cummins, R., 1989: Meaning and Mental Representation. MIT Press.

Dennett, D., 1981: "A Cure for the Common Code", reprinted in Brainstorms: Philosophical Essays on Mind and Psychology. MIT Press.

Dienes, Z., and Perner, J., 1999: "A Theory of Implicit and Explicit Knowledge". Behavioral and Brain Sciences 22 (5).

Evans, G., 1982: The Varieties of Reference. Clarendon Press.

Feigenson, L., Carey, S., and Hauser, M.: "Infants' Spontaneous Ordinal Choices". In press.

Field, H., 1978: "Mental Representation". In ed. Ned Block, Readings in Philosophy of Psychology. Harvard University Press.

Fodor, J., 2000: The Mind Doesn’t Work that Way: The Scope and Limits of Computational Psychology. MIT press.

——1985: "Fodor's Guide to Mental Representation". Reprinted in ed. Alvin Goldman, Readings in Philosophy and Cognitive Science (MIT Press) 271–296.

——1983: Modularity of Mind. MIT Press.

——1975: The Language of Thought. Harvard University Press.

Gleitman, H., 1987: Basic Psychology. Norton.

Hadley, Robert, 1995: "The 'Explicit–Implicit Distinction". Minds and Machines 5, 219–242.

Hauser, M.D., Carey, S., and Hauser, L.B., 2000: "Spontaneous Number Representation in Semi–Free–Ranging Rhesus Monkeys. Proceedings of the Royal Society of London: Biological Sciences 267, 829–833.

Hirschfeld, L.A., and Gelman, S.A., eds., 1994: Mapping the Mind: Domain Specificity in Cognition and Culture. Cambridge U. Press.

Kirsh, D., 1990: "When is Information Explicitly Represented?" In P. Hanson, ed., Information, Language, and Cognition. Vancouver: University of British Columbia Press.

Leslie, A., 1994: "ToMM, ToBY, and Agency: Core Architecture and Domain Specificity". In Hirschfeld and Gelman.

Leslie, A. M., Xu, F. Tremoulet, P. D., and Scholl, B. J., 1998: "Indexing and the Object Concept: Developing 'What' and 'Where' Systems". Trends in Cognitive Sciences, Vol. 2, No. 1.

Margolis, E., and Lawrence, S., eds., 1999: Concepts: Core Readings. MIT Press.

Marr, D., 1982: Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. Freeman.

Matthews, R.J., 1991: "Psychological Reality of Grammars." In A. Kasher, ed., The Chomskyan Turn. Blackwell Publishers.

Pylyshyn, Z.W., 1999: "Is Vision Continuous with Cognition? The Case for Cognitive Impenetrability of Visual Perception". Behavioral and Brain Sciences, Vol. 22, No. 3, 341–423.

————1994: "Some Primitive Mechanisms of Visuospatial Attention". Cognition 50, 364–384.

Smith, E., 1995: “Concepts and Categorization”. In E. Smith and D. Osherson, eds., An Invitation to Cognitive Science, Vol. 3: Thinking. MIT Press.

Scholl, B. J., and Leslie, A. M., 1999: "Explaining the Infant's Object Concept: Beyond the Perception/Cognition Dichotomy". In E. Lepore and Z. Pylyshyn, eds., What is Cognitive Science? Blackwell Publishers.

Spelke, E.S., 1988: "Where Perceiving Ends and Thinking Begins: The Apprehension of Objects in Infancy". In A. Yonas, Ed., Perceptual Development in Infancy. Erlbaum Press.

Spelke, E., and Van de Walle, G., 1993: "Perceiving and reasoning about objects: Insights from infants". In N. Eilan, R. McCarthy, and B. Brewer, eds., Spatial Representation: Problems in Philosophy and Psychology. Oxford U. Press.

Spelke, E., Kestenbaum, R., Simons, D. J., and Wein, D., 1995: "Spatiotemporal Continuity, Smoothness of Motion and Object Identity in Infancy". British Journal of Developmental Psychology 13.

Sperber, D., 1994: "The Modularity of Thought and the Epidemiology of Representations". In Hirschfeld and Gelman.

Stich, S., 1978: "Beliefs and Subdoxastic States." Philosophy of Science 45.

Streri, A., and Spelke, E.S., 1988: "Haptic Perception of Objects in Infancy". Cognitive Psychology 20, 1–23.

Sulkowski, G., and Hauser, M., 2001: "Can Rhesus Monkeys Spontaneously Subtract?" Cognition 79, 239–262.

Van Leeuwen, C., 1998: “Perception”. In W. Bechtel and G. Graham, eds., A Companion to Cognitive Science. Blackwell.

[1] 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.

[2] They also acknowledge a way in which object files are non–or pre–conceptual: they are not symbols for specific kinds (dog, cup...), so applying them does not involve placing objects into kinds.

[3] 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.

[4] 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].

[5] 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.

[6] Bob McMurray is the author of this objection.

[7] We can, of course, call up a memory of a perception of an apple, but that is different than calling up the perceptual representation itself.

[8] Some features of Fodor-modules have been omitted here—for instance, they are associated with a fixed neural architecture and they develop on a fixed schedule.

[9] I have in mind here Sperber’s “metarepresentational module”, and the modular integration accomplished by the language faculty on Carruthers’ view.

[10] I see two other plausible differences. One is with respect to automaticity—judgments of category membership can be very fast and effortless, but they can also be slowed down, as in tricky borderline cases. The second concerns basicness [Rosch 1976]: if OBJECT were built on the same model as DOG, you wouldn't expect it to appear so early in development, given that basic-level concepts tend to be the first acquired, among concepts for which there are prototypes.

[12] On the other hand, there is some evidence that these representations form the core of the adult’s object concept, and it might be suggested that the latter enjoys the same sort of deployability as APPLE or DOG. But interestingly, this does not appear to be the case. SPELKE-OBJECT is more like, say, C-COMMAND than DOG: it has taken scientific labor to uncover these representations as playing an important role in our cognition; they are not accessible nor deployable in the way that DOG is.

[13] Scholl and Leslie sort of acknowledge this, but not quite as forcefully as I think is appropriate. In personal correspondence, Brian Scholl has amplified this acknowledgment.

[14] In the first category we find demonstrations that young infants use common motion, but not static featural information, as a clue to object-hood [Kellman and Spelke 1983]. 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 and Graber 1987].