4th draft, 30/09/2012 Maciej Witek Institute of Philosophy, University of Szczecin, Poland Contextual Facilitation of Colour Recognition: Penetrating Beliefs or Colour-Shape Associations? * According to Jerry Fodor (1983) and Zenon W. Pylyshyn (1999, 2003), the early visual system — whose function is to produce the primal sketches of distal layouts — is cognitively impenetrable. To call it cognitively impenetrable or informationally encapsulated is to maintain that its operations cannot be affected by what the subject believes, expects and desires. That is to say, the computational process that mediates the production of primal sketches has no access to the subject's central database or his background beliefs. Rather, it is an input-driven process whose job is to integrate the transduced signals coming from the subject's senses into what Fodor calls shallow outputs or percepts: information-carrying states that represent their distal objects in terms of their locations, sizes, shapes and colours.1 The only non-sensory information the system can use comes from its private or local database. To argue for the cognitive impenetrabilty of early vision is to reject the popular idea that there is a continuity between perception and cognition or, in other words, that the process of perceptual integration is top-down through and through. It is also to claim that there is a level of perceptual representation which is (a) shallow, (b) phenomenally conscious and (c) cognitively impenetrable in that its phenomenal content or character cannot be affected by the top-down processes. The paradigmatic examples of such representations are provided by the so-called persistent visual illusions, that is, by visual experiences whose phenomenal content remains unchanged despite strong beliefs to the contrary. For example, even though the subjects know that the Müller-Lyer lines are not equal in length, they cannot stop seeing one line as longer than the other. Consider, by analogy, the checker shadow illusion: despite their knowing that there is no difference in shade between the shadowed square and the non-shadowed one, the subjects cannot * The work on this paper was supported by a scholarship from the Bednarowski Trust to the University of Glasgow (8 April — 19th May 2010). 1 Admittedly, Fodor adds that percepts represent their object in terms of what he calls “basic perceptual categories”, e.g. “dog” (Fodor 1983: 94-97). Let us assume, however, that early vision — as far as it is consider as an encapsulated process — involves no categorisation or object recognition at all (see Raftopoulos 2001: 434). That is to say, its proper products are shallow in that their phenomenal character involves only shapes, sizes, locations and colours. th 1 help seeing them as different in shade. Let us assume that the overall phenomenal content of a visual experience — or, for simplicity, the experience as such — comprises two aspects: shallow and deep. The former represents the relevant distal stimulus in terms of its size, shape, location, colour and luminosity, whereas the latter represents it as falling under certain categories, such as “dog”, “rabbit”, “duck”, and so on. The proponent of the modular account of early vision would say, therefore, that only deep aspects of visual experiences are cognitively penetrable; their shallow aspects, by contrast, result from informationally encapsulated, data-driven computations. There are, however, some experimental results that seem to undermine the idea of the cognitive impenetrability of early vision. There have been reported cases of (a) shallow and (b) phenomenally conscious visual representations that seems to be (c') affected by the top-down factors, i.e., by what the subjects believe, expect or desire. 2 Typically, these cases involve two subjects focusing their visual attention on the same distal stimulus but having different visual experiences or, more precisely, producing visual experiences that differ significantly in their shallow aspects. One reaction to such findings is to account for the reported differences in terms of beliefs that penetrate the process of early perceptual integration; that is to say, the subjects produce experiences that differ in their shallow aspects because the subjects have different beliefs about what they see. Of course, the proponents of the modular account of early vision attempt to defuse such counterexamples by arguing that the allegedly penetrating beliefs affect either the (i) preperceptual or (ii) post-perceptual stages of vision, that is, either (i) manipulate the subject's visual attention or (ii) contribute to what we call the deep aspect of experience or even to the reportable content of the subjects' introspective judgements.3 In her “Cognitive Penetration of Colour Experience: Rethinking the Issue in Light of an Indirect Mechanism” Fiona Macpherson (2012) argues that there is one case of the alleged cognitive penetrability of visual experience that cannot be explained away along the lines of the modular theory. What she has in mind is the effect reported by John L. Delk and Samuel Fillenbaum4 that suggests, it seems, that the subject's colour experiences are sensitive to the subject's beliefs about the typical colours of objects. Delk and Fillenbaum: took a sheet of paper of a uniform orangish colour. From it, they cut out shapes of various object. Some of these objects were characteristically red: a heart, a pair of lips, an apple. Some were not characteristically red: an oval, a circle, an ellipse, a horse's head, a bell, a mushroom, a square. One at a time, the cutout shapes were placed in front of a coloured background that could be altered by the subject. The background could be altered from a yellow colour through orange to a red colour by twisting a nob. The subjects were instructed to change the background colour until it 2 3 4 See Bruner and Postman 1949 and Bruner and Goodman 1947. For a discussion of these strategies see Pylyshyn 1999. See Delk and Fillenbaum 1965. 2 was the same colour as the cutout shape in front of it, so that the cutout shape could no longer be distinguished from the background. When the cutout shape of an object that had a characteristically red colour was placed in front of the background, subjects altered the background colour so that is was more of a red colour that the colour that they made the background when the cutout shape was of an object that didn't have a characteristically red colour. When the cutout shape was of an object that didn't have a characteristically red colour the subjects made the background less red and more yellow.5 In short, depending on the shape of the cutouts, the subject see some of them as orangish and others as reddish. According to Macpherson, to explain this observed regularity one has no alternative but to acknowledge that the subject's visual experiences are penetrated by his or her beliefs about the typical colours of objects. In other words, one has to agree that the shallow aspect of visual experience is cognitively penetrable.6 In my view, however, what is responsible for the regularity reported by Delk and Fillenbaum are not the subject's cognitive states but, rather, the internal organization of his or her early vision system. Admittedly, the subject may possess beliefs to the effect that hearts and tomatoes are characteristically red and ovals and bells are not. This fact, nevertheless, has nothing to do with the effect described by Delk and Fillenbaum. Therefore, rather than explaining away the case by arguing that the penetrating beliefs bears on either the pre-perceptual or post-perceptual stages of visual information processing, I claim that the mechanism that mediate colour recognition involves no penetrating beliefs at all. Undoubtedly, the mechanism is sensitive to the shape of the processed object. To acknowledge this, nevertheless, is not to say that it is affected by the subject's beliefs about the typical colours of objects. The mechanism is like priming in that it employs the subject's implicit rather than explicit memory.7 That is to say, it is autonomous and sub-personal rather than cognitively penetrable and personal. In what follows I offer a sketchy model of the autonomous mechanism of colour recognition. I also suggest how the model can be tested empirically. It is worth stressing that with this model in hand not only are we in a position to explain the Delk and Fillenbaum effect away, but we are able to consider it as a positive evidence for the modular account of early vision. Let me start with an analogy. Studies on speech perception show that the reaction time for lexical decisions (i.e., word/non-word decisions) depends on whether the target words are predictable in the sentential contexts in which they occur.8 For example, “rope” is recognized as a word more quickly when it follows the phrase “The guide threw the lady a …” than it follows the 5 Macpherson 2012: 38. In a similar vein, James A. Shirillo (1999) argues that given the results of Delk and Fillenbaum's experiment we have no alternative but to accept the idea of the cognitive penetrability of colour recognition. 7 For a discussion of priming as an implicit mechanism see Bar 2000. 8 See Swinney 1979, Forster 1981, Fodor 1983: 76-82 and Fodor 1986: 130-135. 6 3 phrase “The priest threw the lady a …”. 9 The cloze value of “rope” — that is, the value that reflects its predictability in a given context — is high in the first case whereas it is low in the second. At first sight, the data under discussion seem to support the view that language perception is cognitively penetrable, that is, that the process whose job is to represent linguistic stimuli in terms of their phonetic, lexical and grammatical properties can be affected by semantic or background information. One can maintain, for example, that the process that mediates the subject's lexical decisions is penetrated by his or her beliefs, e.g., by his or her belief to the effect that priests, unlike guides, have to do with ropes. In short, one can conclude that lexical decision is informed by topdown information flow. It turns out, however, that we can escape this conclusion and stick to the view that language perception is a purely bottom-up process. In his Modularity of the Mind Fodor puts forth the hypothesis according to which lexical analysis has access to a local database that he calls the mental lexicon. The mental lexicon can be depicted as a network of lexeme-to-lexeme associations that mediate our lexical access: Suppose the mental lexicon is a sort of connected graph, with lexical items at the nodes and with paths from each item to several others. We can think of accessing an item in the lexicon as, in effect, exciting the corresponding node; and we can assume that one of the consequences of accessing a node is that excitation spreads along the pathways that lead from it. Assume, finally, that when excitation spreads through the portion of lexical network, response thresholds for the excited nodes are correspondingly lowered. Accessing a given lexical item will thus decrease the response times for items to which it is connected.10 Let us assume that the language perception system computes a hierarchy of intermediate representations — phonetic, lexical and syntactic — that are subsequently integrated in a linguistic percept or structural description. The job of the mental lexicon, then, is to facilitate the production of the lexical representation of a verbal stimulus. Let us also take for granted that utterances are processed one word at a time. Therefore, the lexical analyses of the phrase “The guide threw the lady a …” results, among others, in accessing the node GUIDE and, consistently, in exciting a number of its neighbouring lexical nodes. In particular, it results in exciting the node ROPE, which becomes highly accessible or, in other words, is assigned a high cloze value. In short, the subject's lexical decisions are contextually facilitated. Appearances to the contrary, however, the mechanism that mediates this facilitation is associative rather than intelligent; that is to say, it makes use of the associated relations between lexical items — the totality of which makes up the mental lexicon — rather than of the beliefs that form the subject's background knowledge. The structure of the mental lexicon is shaped by the subject's linguistic experience and as such mimics the structure of his or 9 10 See Fodor 1986: 131. Fodor 1983: 80. See also Fodor 1986. 4 her knowledge. It is the former, not the latter, however, that underlies the fast and subpersonal process of lexical analyses. The process, let us add, that is cognitively impenetrable. Let us return to the issue of colour recognition. In my view, the Delk and Fillenbaum effect can be accounted for by adopting a strategy similar to the one successfully used by Fodor. I assume, namely, that the early vision system computes a collection of intermediate information-carrying states that represent a distal object in terms, respectively, of its size, shape, location, colour and luminosity, and, next, integrates these representations into a percept or primary sketch. Following Fodor11, I also assume that there is top-down information flow within the system; for example, the intermediate representation that identifies the processed stimulus by its colour can affect the representation that identifies it by its shape, and vice versa. What mediates these interactions, I think, is the system's local database that can be called the mental pallet. We can think of it as a network or graph whose nodes stands for shapes or colours. More precisely, the mental pallet consists of a number of shape-colour associations that make up its associative structure. Some of the associations are strong, whereas others are weak. For example, the node standing for the leaf-like shape is strongly associated with the node representing green and weakly associated with the nodes that stand for yellow, red and brown, respectively. Note that the associations under discussion reflect regularities to be found in nature. That is to say, most leaves we encounter are green and some leaves are yellow, red or brown. Consider now a subject who is shown a leaf-shaped cutout and recognizes its shape. Assume that one of the consequences of recognizing the shape is that of exciting the relevant shape-node in the system's mental pallet. The excitation spreads along the relevant shape-colour associations. As a result, the colour-nodes that are linked with the excited shape-node become accessible and ready for use by the colour recognition subsystem. Now, let us reconsider an example of the Delk and Fillenbaum effect: a subject who is presented with a heart-shaped orangish cutout sees it as reddish (that is, he or she alters the background in front of which the cutout is placed so that the background is more reddish than orangish). How is it possible? According to the mental pallet hypothesis, the subject's early vision system recognizes the shape of the cutout and, as a result, excites the node HEART-SHAPE of the subject's mental pallet. The excitation spreads along the relevant shape-colour associations. In particular, it makes the node RED highly accessible and ready for use by the colour-recognition subsystem. That is to say, the response threshold for red is significantly lowered and, as a result, the subject sees the cutout as reddish. His or her colour-recognition mechanism is influenced by the information of the typical colour of heart-shaped objects. Note, however, that the information does not take the form of a penetrating belief; rather, it is coded in the form of a relevant shape-colour 11 See Fodor 1983: 76. 5 association that is built in to the architecture of the subject's mental pallet. Therefore, if the mental pallet hypothesis is true, the Delk and Fillenbaum effect has nothing to do with the alleged cognitive penetration of early vision. What it suggests, rather, is that there is top-down information flow within the early vision system; the flow, let us add, that is mediated by the associative structure of the mental pallet. In other words, the colour recognition tasks are contextually facilitated. What underlies the facilitation, however, is an associative rather than intelligent process that can be likened to perceptual rather than conceptual priming (i.e., to those forms of priming that depend on implicit rather than explicit memory). Note that this moral can be easily reconciled with what Delk and Fillenbaum's experiment suggests. They conclude: One is led to the conclusion that past association of color and form does in some way influence perceived color, since that is the one respect in which the figures did clearly differ.12 Note that offering the modular account of the Delk and Fillenbaum effect I assume that there are two parallel or twin databases that store roughly the same information of the typical colours of object. The first database is local. It takes the form of the mental pallet, that is, of a network of the colour-shape associations. The second database is global or central. It takes the form of a body of beliefs about the typical colours of objects. One can ask, therefore, what is the point of having these two twin databases since, it seems, one would suffice? Note, first, that the associative mechanism that uses the mental pallet works faster than its intelligent counterpart. It works faster for two reasons. First, it does not involve the identification of the stimuli it processes in terms of such categories as “heart”, “leaf”, “tomato”, and so on; that is to say, it operates on the shallow aspects of visual experiences, since it associates colours with shapes rather than with objects. Second, it does not face the frame problem, that is, the information that can influence its operations is constrained by the associative architecture of the mental pallet. The intelligent mechanism, by contrast, would have to decide which beliefs about the possible colours of, say, hearts, have to be regarded as bearing on the process of visual integration. That is why it is good to have such an associative and unintelligent system. It works fast and facilitate our performance on the colour recognition tasks (note, for example, that in poor light conditions it is easier to recognize an object's shape rather than its colour). Second, it is worth noting that even though the structure of the mental pallet results from learning by conditioning13, it cannot be modified by our experience in a flash. Normally, it takes time to alter some of the shape-colour association and, as a result, to revise the structure of the 12 13 Delk and Fillenbaum 1965: 293, my emphasis — M.W. It is not precluded, however, that the central part of the pallet is innate. 6 mental pallet. That is why it is good to have the second, intelligent mechanism. It learns faster, accommodating the structure and content of its database to new evidence. Its function is to evaluate and, if it is necessary, to correct the outputs of the early vision system. Let me end with two methodological remarks. Note, first, that if the Delk and Fillenbaum effect is persistent — that is, if it does not disappear when the subject is told that the cutouts are in fact orangish — it can be regarded as supporting rather than undermining the modular account of early vision. Second, the mental pallet hypothesis can be empirically tested. (i) One can check whether some amnesic patients — more precisely, patients that are unable to recollect beliefs about the typical colours of objects — are sensitive to the Delk and Fillenbaum illusion. If they are, then it is reasonable to suppose that the mechanism that is responsible for the Delk and Fillenbaum effect is autonomous, that is, that it performs its operations independently of our central processes. In this respect, it can be likened to implicit rather than explicit memory. (ii) One can also ask the normal subjects to learn a number of propositions about the typical colours of cutouts whose shapes are artificial though easily recognizable. Next, the subjects' explicit memory can be tested, that is, they can be asked to recollect the colours of the cutouts they were informed of. It would be very interesting to check whether the new beliefs so formed are likely to influence the subjects' performance on the Delk and Fillenbaum task. If they are not, then it is reasonable to suppose that the colour recognition system works independently of the subject's explicit memory or, in other words, that it is cognitively impenetrable. References: Bar, Moshe, 2000, “Conscious and Nonconscious Processing of Visual Object Identity”, In: Y. Rosetti and A. Revousuo (Eds.), Dissociations: Interactions between dissociable conscious and nonconscious processing, Amsterdam: John Benjamins, 153-174. Bruner, Jerome S. and Goodman Cecile C., 1947, “Value and need as organizing factors in perception”, Journal of Abnormal and Social Psychology 42, 33-44. Bruner, Jerome S. and Postman Leo, 1949, “Perception, Cognition, and Behavior”, Journal of Personality 18(1), 14-31.” Delk John L. and Fillenbaum, Samuel, 1965, “Differences in Perceived Color as a Function of Characteristic Color”, The American Journal of Psychology 78(2), 290-293. Fodor, Jerry, 1983, The Modularity of Mind, Cambridge, Mass.: MIT Press. Fodor, Jerry, 1986, “Modularity if Mind: Jerry Fodor's Response”, In: Z.W. Pylyshyn and W. Demopoulos (Eds.), Meaning and Cognitive Structure, Norwood, New Jersey: Ablex Publishing Corporation, 129-135. 7 Forster, K.I., 1981, “Priming and the effects of sentence and lexical contexts on naming time: Evidence for autonomous lexical processing”, The Quarterly Journal of Experimental Psychology Section A 33(4), 465-495. Macpherson, Fiona, 2012, “Cognitive Penetration of Colour Experience: Rethinking the Issue in Light of an Indirect Mechanism”, Philosophy and Phenomenological Research 84(1), 24-62. Pylyshyn, Zenon W., 1999, “Is vision continuous with cognition? The case for cognitive impenetrability of visual perception”, Behavioral and Brain Sciences, 22, 341-423. Pylyshyn, Zenon W., 2003, Seeing and Visualizing, Cambridge, Mass.: MIT Press. Raftopoulos, Athanassios, 2001, “Is perception informationally encapsulated? The issue of the theory-ladeness of perception”, Cognitive Scienc, 25, 423-451. Schirillo, James A., 1999, “Color memory penetrates early vision”, Behavioral and Brain Sciences 22, 393. Swinney, David A., 1979, “Lexical Access during Sentence Comprehension: (Re)Consideration of Context Effects”, Journal of Verbal Learning and Verbal Behavior 18, 645-659. 8