Chapter 34 Demand Characteristics?: The Second Phase of the Debate
The second phase of the debate began about eight years after the first, when Pylyshyn elaborated his views (Pylyshyn 1981).
This phase of the debate focused on the data collected earlier. Whereas the proponents of depictive representation claimed that the data reflected the processing of depictive representations, the propositionalists now focused on possible methodological problems with the experiments. Two such problems were raised: “experimenter expectancy effects" and "task demands."Intons-Peterson (1983) performed an experiment in which she compared scanning images to scanning physically present displays. Half of the experimenters were told that the image scanning should be faster and half were told that the perceptual scanning should be faster. She found that the experimenters' expectations influenced the results: when experimenters expected faster perceptual scanning, the subjects produced this result; when they expected faster image scanning, there was no difference in overall times. Thus, the experimenters were somehow leading the subjects to respond as the experimenters expected.
Jolicoeur and Kosslyn (1985) decided to test the idea that the increases in times with increasing distance scanned reflect the subjects' responding to experimenter expectancy effects. We performed a series of experiments using Intons-Peterson's methodology. For example, we told one experimenter that we expected a U-shaped function, with the most time being required to scan the shortest and longest distances. The reason for this prediction, we explained, was that the four closest objects "group" into a single chunk—because of the Gestalt laws of similarity and proximity—and so they are "cluttered" together, making it difficult to scan among them.
And the longest distances require more time than the medium ones because more scanning is involved.The results from this experiment were identical to those found previously: times increased linearly with increasing distance. In additional experiments Jolicoeur and Kosslyn varied experimenter expectancy in different ways, none of which affected scan times. Indeed, these experimenters failed to replicate Intons-Peterson's original finding. What could be going on here? Many details of such experiments can differ from laboratory to laboratory (for instance, making sure subjects always keep their fingers on the response buttons), and these details could be critical for obtaining experimenter expectancy effects. The important point is that, whatever caused the experimenter expectancy effect in Intons-Peterson's study, it was not present in the procedures used in the initial studies of image scanning. Thus, these results cannot be explained away as simply reflecting how well subjects can satisfy the expectations of the experimenter.
Taldng an alternative tack, Pylyshyn (1981) claimed that the very instruction to scan an image induces subjects to pretend to scan an actual object—which leads them to take more time to respond when they think they would have taken more time to scan across a visible object. The way the subjects estimate how long to wait (unconsciously) would involve propositional processing of some sort.
This potential concern was ruled out by image-scanning experiments that eliminated all references to imagery in the instructions. Finke and Pinker (1982, 1983; see also Pinker, Choate, and Finke 1984) showed subjects a set of random dots on a card, removed the card, and presented an arrow. The question was, if the arrow were superimposed over the card containing the dots, would it point directly at a dot? Subjects reported using imagery to perform this task, and Finke and Pinker found that the response times increased linearly with increasing distance from the arrow to a dot.
Furthermore, the rate of increase in time with distance was almost identical to what we had found in our earlier experiments. Because no imagery instructions were used, let alone mention of scanning an image, a task-demands explanation seems highly implausible.Goldston, Hinrichs, and Richman (1985) actually went so far as to tell the subjects the predictions, which is never done in typical psychological experiments. Even when subjects were told that the experimenter expected longer times with shorter distances, they still displayed increased times with distance scanned. Telling subjects different predictions did affect the degree of the increase with distance, but this result is not surprising: given the purposes of imagery, one had better be able to control imaged events! What is impressive is that even when subjects were, if anything, trying for the opposite result, they still took longer to scan across longer distances.
Finally, Denis and Carfantan (1985) described the basic scanning experiment to naive subjects and asked them to predict the outcome. Although these subjects were good at predicting many of the effects of imagery (for example, that it will help one to memorize information), they were very poor at predicting the results of scanning experiments and the like. If subjects are using knowledge about perception and physics to "fake" the data in the experiments, it is puzzling that they evince no such knowledge in this situation.
Bibliography
Anderson, J., and Bower, G. 1973. Human Associative Memory. New York: V. H. Winston and Sons. Denis, M. and M. Carfantan. 1985. People's knowledge about images. Cognition 20:49-60.
Finke, R- A., and S. Pinker. 1982. Spontaneous imagery scanning in mental extrapolation. Journal of Experimental Psychology : Human Latming and Memory 8:142-147.
Finke, R. A., and S. Pinker. 1983. Directional scanning of remembered visual patterns. Journal of Experimental Psychology: Learning, Memory, and Cognition 9:398-410.
Goldstoa D. B., J. V. Hinridis, and C. L Richman. 1985. Subject's expectations, individual variability, and the scanning of mental images. Memory and Cognition 13:365-370.
Intons-Petersoa M. J. 1983. Imagery paradigms: How vulnerable are they to experimenters' expectations? Journal of Experimental Psychology: Human Perception and Performance 9:394-412.
Jolicoeur, P., and S. M. Kosslyn. 1985. Is time to scan images due to demand Characterisitics? Memory and Cognition 13:320—332.
Kosslya S. M. 1973. Scanning visual images: Some structural implications. Perception and Psychophysics 14: 90-94.
Kosslya S. M., BalL T. M., and B. J. Reiser. 1978. Visual images preserve metric spatial information: Evidence from studies of image scanning. Journal of Experimental Psychology : Human Perception and Performance 4:47-60.
Lea, G. 1975. Chronometric analysis of the method of lod. Journal of Experimental Psychology: Human Perception and Performance 2:95-104.
Pinker, S. 1980. Mental imagery and the third dimension. Journal of Experimental Psychology : General 109: 354—371.
Pinker, S., P. A. Choate, and R. A. Finke. 1984. Mental extrapolation in patterns constructed from memory. Memory and Cognition 12:207-218.
Pylyshya Z. 1973. What the mind's eye tells the mind's brain: A critique of mental imagery. Psychological Bulletin 80:1-24.
Pylyshya Z. 1981. The imagery debate: Analogue media versus tadt knowledge. Psychological Review 87: 16-45.
Introduction
ConnecHonism is a loosely organized research program involving researchers in computer science, psychology, and in some cases neurobiology. The research program has received considerable attenHon both in academia and the popular press, and is sometimes touted as a radical breakthrough in our understanding of the human mind. On the other hand, there are those who argue that ConnecHonism is nothing more than "high tech" Lockean assodaHonism. In fact, the truth probably lies somewhere between these posiHons.
Careful study of assodaHonist and ConnecHonist writings reveals not only marked differences, but a number of fundamental SimilariHes as well.The basic idea underlying assodaHonism certainly is not new (it can be found in Aristotle, according to some). We begin with Thomas Hobbes, who is interested in giving an account of our train of thinking. The secHon contains a famous passage in which Hobbes shows how the discussion of a dvil war could be causally related to someone asking the price of a Roman penny. The idea of the war triggers a sequence of related or conneded ideas, resulting in the seemingly anomalous quesHon.
John Locke develops the assodaHonist doctrine somewhat, arguing that some ideas come to be assodated by natural connections holding between them while other ideas come to be assodated through custom (educaHon, interests, etc.). Locke also argues that assodaHon can account for certain kinds of pathological thinking. For example, if one has a bad experience in a particular room, one might be unable to enter the room again without thinking of the experience. This is because the ideas of the room and the experience will have become inextricably assodated. David Hume proposes certain addiHonal prindples that govern the assodaHon of ideas: resemblance, conHguity, and cause and effect.
Willam James gives a helpful survey of work in assodaHonist psychology and addresses two very important issues: the quesHon of whether any general assodaHve prindple might underlie the proposed assodaHonist laws, and the quesHon of whether neural mechanisms underlie assodaHonist psychology. James thus anHdpates those contemporary philosophers who take ConnecHonism to be grounded in neural mechanisms.
We begin the contemporary debate with an introduction (by James McClelland, David Rumelhart, and Geoffrey Hinton) to a version of ConnecHonism known as parallel distributed processing (PDP). While these writers do not make explidt reference to the early assodaHonist psychologists, it is dear that they share certain fundamental views.
In PDP models of memory, for example, properties might be assodated with mutually exdtatory units (processors). So, if a unit representing Rene Descartes were activated, there might be a corresponding exdtaHon of a unit representing the property of being a philosopher, or the property of being French. The connection strengths between units within the network are set by training the network with a general learning algorithm that may be considered a descendent of the prindples first enund- ated by Locke, Hume, and subsequent assodaHonists.246 Part IV Introduction
The PDP perspective stands in marked contrast to what is sometimes called the classical theory of computation, in which computation consists of formal operations on complex syntactic objects. For example, on the classical view the inference from the sentence P&Q to the sentence P is executed by a formal mechanism sensitive only to the syntactic form of PfcQ. Jerry Fodor and Zenon Pylyshyn take strong exception to the PDP paradigm, suggesting that there are several reasons for preferring the classical theory. They argue that PDP models, by eschewing structure-sensitive processes, give up the ability to account for a number of phenomena including (i) the productivity of human linguistic processes (i.e., the ability to create and comprehend sentences of unbounded length like "This is the cat that ate the rat that lived in the house that...."),
(ii) Systematidty (understanding "Jack likes Jill" entails understanding 4Jill likes Jack"),
(iii) compositionality (the meaning of a sentence is a function of the meaning of its parts), and (iv) inferential coherence (inferences from, e.g., P&Q to P).
Paul Smolensky is unconvinced that these arguments pose a problem for connec- tionism. Smolensky notes that the kinds of problems raised by Fodor and Pylyshyn do not argue against Connectionist treatments of "soft" mental processes but merely its ability to handle "hard" processes such as logical inference as well. Smolensky concedes that there are structure-sensitive processes, speculating that they need not be handled in a classical model but could be accounted for by supposing that "the mind is a statistics-sensitive engine operating on structure-sensitive numerical representations."
Gaims and counterclaims regarding PDP systems abound today, but Seymour Papert offers some deflationary remarks. He notes that the mathematical properties of PDP networks have yet to be explored and suggests that even for very simple ancestors of these networks, the actual properties are difficult to determine, and once determined, often unexpected. The abilities of full-blown Connectionist systems (as opposed to toy implementations) are simply unknown.
Further Reading
Some classical works in assodationist psychology:
Hartley, D. 1749. Observations on Man, His Frame, His Duty, and His Expectations. London. Hebb, C 0.1949. The Organization of Behavior. New Yorie Jdin Wiley.
Mandler, G., and Mandler, J., eds. 1964. Thinking: From Association to Gestalt. New Yorie John Wiley. MiD, J. 1829. Analysis of the Phenomena of the Human Mind. London.
Some further reading on Connectionism:
BechteL W4 and A. Abrahamson. 1990. Connectiomsm and the Mind: An Introduction to Parallel Processing Networks. Oxford: Basil Blackwell.
Fodor, J., and B. McLaughlin. 1990. "Gxinectionism and the Problem of Systematicity: Why Smolensky's Solution Doesn't Work." Cognition 35,183-204.
Minsky, M4 and S. Papert. 1986. Perceptrons. Expanded edition. Gunbridge, MA MTT Press.
Rumelhart, D4 J. McClelland, and the PDP Research Group. 1986. Parallel Distributed Processing, voL 1. Cambridge, MA* MTT Press.
McClelland, J4 D. Rumelhart, and the PDP Researdi Group. 1986. Parallel Distributed Processing, voL 2. Cambridge, MA MTT Press.