February 20-22, 2009
Institute for Research in Cognitive Science
University of Pennsylvania
Philadelphia, PAOrganizers
Abstracts
Austen Clark: University of Connecticut, Department of Philosophy The cognitive status of spatial features in early vision
The perception of the lightness of surfaces has been shown to be affected by information about the spatial configuration of those surfaces and of their conditions of illumination. For example, two surfaces of equal luminance can appear to be of very different lightnesses if one of the two appears to lie in a shadow. How are we to understand the character of the processes that integrate such spatial configuration information to yield the eventual appearance of lightness? This paper makes some simple observations about the vocabulary of appearance used in these contexts, and proposes that the end results can be called "phenomenal" in a traditional sense of that word. Processes whose products are phenomenal are next distinguished from processes characterized in other terms: (a) processes of perceptual grouping; (b) processes of perceptual organization; and (c) attentional (as opposed to preattentive) processes. These four categories are conceptually and empirically distinct. In particular, the paper reviews some recent evidence that appearances as of contours, occlusion, and amodally completed shapes can occur preattentively. Some benefits of detaching our notion of "phenomenal" from any necessary ties to attention or consciousness are briefly discussed.
Respondent: Elisabeth Camp, University of Pennsylvania, Department of Philosophy
Carl Granrud: University of Northern Colorado, Department of Psychological Sciences: Judging the size of a distant object: Strategy use by children and adults
The ability to estimate the size of a distant object improves during childhood. Our research has asked whether this developmental change results from increasing sensitivity to visual cues for distance and size or from increasing use of deliberate size-estimation strategies. We investigated childrens size-estimation abilities in several recent experiments. In each, participants viewed standard discs at near (5-6.1 m) and far (61 m) distances, and chose from a set of nearby discs one that matched the standard. They were also asked to explain their choices. Children who reported that they deliberately inflated their size estimates to compensate for the effects of distance on perceived size were categorized as strategy users. Those who reported no such strategy use were categorized as strategy nonusers. At the far distance, strategy users generally made accurate size estimates or overestimated size, whereas nonusers consistently underestimated size. Self-reported strategy use increased with age between 5 and 11 years; but strategy users made more accurate far-distance size estimates than strategy nonusers regardless of age. In addition, strategy users scored higher than nonusers in tests of verbal and spatial reasoning. Strategy users and nonusers size estimates did not differ at the near distance in any of our experiments, indicating that these two groups did not differ in perceptual abilities, motivation, or task proficiency. Overall, our findings suggest that developmental changes in far-distance size estimation accuracy result from the development of reasoning abilities and deliberate strategy use. In a related series of experiments, we found evidence that adults also use explicit size-estimation strategies to compensate for inaccurate far-distance perception.
Jonathan Cohen: University of California, San Diego, Department of Philosophy, Computation and the ambiguity of perception
There is a class of phenomena that suggests strongly that perception is, in some sense to be explained, ambiguous in what it tells us about the world. In the cases at issue, the perceptual system is capable of responding to a single stimulus -- say, as manifested in the ways in which they sort that stimulus -- in different ways. Indeed, in many cases, subjects can be made to switch at will between these different modes of response. This paper is about that ambiguity, and about how it should be characterized and accounted for within a general theory of perception.
Respondent: Susan Schneider, University of Pennsylvania, Department of Philosophy
Alan Gilchrist: Rutgers University, Newark, Department of Psychology, Objective and subjective sides of visual experience
Every perceptual experience has both an objective and a subjective side. We see object size, independent of distance, but we also see that distant objects project smaller images. This duality was missing from ancient theories. Objects were apprehended directly, whether through rays extending from the eye or by replicas given off by the object. Early modern conceptions, following discovery of nerve transmission and the retinal image, focused on the subjective aspect. Energy made contact with the sensory surface and produced a corresponding sensation. Helmholtz and Hering brought the objective aspect back into the picture by splitting visual experience into two stages, with sensation representing the subjective side and perception, through cognitive processes, representing the objective side. Gestalt theory denied this dualism, rejecting both sensory and cognitive stages. Indeed, evidence against both the doctrine of sensations and the role of cognition in vision is plentiful. Nevertheless, the sensation/perception dualism persists in many forms, including the thriving business in brightness models and the resurgence of past experience theories. But the fact that we have some ability to detect visual angle and luminance does not mean that these are the raw sensations out of which perception of size and lightness is built. Size and color constancy are truly visual, not cognitive. Goldfish show constancy. So instructions to subjects in vision experiments must be carefully constructed to avoid both the proximal mode of perception on the one hand and cognitive judgments on the other. Constancy is never complete, and the precise pattern of constancy failures offers us the best insight into the nature of the visual software.
Respondent: Robyn Oliver, University of Pennsylvania, Department of Psychology
Gary Hatfield: University of Pennsylvania, Department of Philosophy, Perceptual and cognitive factors in spatial perception
In the history of visual theory, there has been an ongoing discussion concerning the structure of phenomenal (in this case, visual) space. Some theorists have held that human visual experience is actually of a two-dimensional sensory core that corresponds to the retinal image. Others have countered that distance and depth are phenomenally immediate. The predominant position among the latter has been that the visual system, under good viewing conditions, achieves full phenomenal constancy: objects appear with the sizes and shapes that they in fact have, at the distance and in the orientation they have. There have been two minority alternatives to this second position about visual space: (a) that visual space is really non-Euclidean, such that parallels on the ground plane (e.g., railroad tracks) appear to converge as they run away in depth, and (b) that the subject replaces an unnoticed sensory core with a visual experience of an object as it appears from close up (e.g., in personal space, or from a normal or preferred viewing position for that object). I want to add to the list of alternatives by suggesting an additional (but previously neglected) description of visual space. Visual space is compressed in a Euclidean 3-D to 3-D projection that allows for railroad tracks to converge while running away in depth and does so without invoking non-Euclidean geometry. I would then account for the ability to give full constancy responses as relying on cognitive factors (as opposed to a phenomenal transformation, as in (b)) that allow perceivers to attain accurate beliefs about objective sizes and shapes and to report such knowledge. I thus propose, as a description of phenomenal visual space, a structure that is distinct from a two-dimensional sensory core but that exhibits converging parallels. And I propose, as a cognitive factor in perception, an ability to judge objective size and shape based on this contracted spatial representation.
Respondent: Carl Granrud, University of Northern Colorado, Department of Psychological Sciences
David Hilbert: University of Illinois, Chicago, Department of Philosophy, Constancy, content and inference
Respondent: Sarah Allred, University of Pennsylvania, Department of Psychology
Don MacLeod: University of California, San Diego, Department of Psychology
Vision is commonly thought of as a two step process in which the acquisition of sense data is followed (in a causal sense) by their interpretation. But mechanistic models that embody this idea have only recently begun to receive consideration. The mechanistic embodiment of interpretive processes should include feedback connections, through which the preconceptions and attentional priorities of the observer can exert a ‘top-down' influence on the neural representation, and such connections are abundant within the brain, extending more sparsely even to the retina.
But an architecture of this type provides no justification for the concept of a 'given', or interpretation-free, core of perceptual experience. Nor is that concept supported by any judgments made by observers with normal vision: phenomenal regression to the real object appears inevitable. Even in the task of selecting a purely 'white' stimulus in an otherwise dark environment, errors arise in a distinctive pattern that is plausibly traceable to fluctuations in the operation of the interpretive or inferential processes associated with color constancy. In this sense the 'eye of the artist' is not available to the normally sighted, although one individual, with sight recently recovered in adulthood, does make photometer-like visual judgments of brightness.
Roughly speaking, the retina signals factors of change in space or time, and the reconstruction of local brightness and color depends on a concatenation of these factors of change, in much the way suggested by Land's early retinex algorithm. Feedback connections could be helpful in implementing such a scheme, but are not essential for it. But the concatenation principle is too simple, and experimental data are more easily accommodated in a less constrained theoretical framework, in which color and brightness are parameters of a centrally constructed model of the environment, and feedback allows the expectations from the perceptual model to be compared with the data provided by the afferent sensory signals.
Respondent: David Brainard, University of Pennsylvania, Department of Psychology
Sarah Allred: University of Pennsylvania, Department of Psychology, Approaching color with Bayesian algorithms
What is the goal of color vision? How ought we to think of color appearance and color constancy? The sensory signal on which visual perception is based is ambiguous with respect to the state of the world. Under one view, the goal of vision is to maintain a stable representation of object properties across changes in the environment. To do this, the visual system must infer what object was most likely to have caused ambiguous sensory signal. One approach to resolving the ambiguity is to rely on statistical regularities in the world. Our visual systems evolved in this world, and thus its statistical regularities are likely to be of functional importance to vision. In this paper, I present a Bayesian approach to solving that estimation problem. In the first section, I describe the Bayesian approach generally. Next I outline three Bayesian models that predict perceived lightness, color constancy and color appearance, respectively. Finally, I discuss some advantages and disadvantages to the Bayesian approach.
Respondent: Jonathan Cohen, University of California, San Diego, Department of Philosophy
Maria Olkkonen: University of Giessen, Department of Psychology, Higher-level influences on color perception and color constancy
Color vision has presumably evolved to improve discrimination and identification of objects. Color constancy, i.e., the stability of color appearance under changing illumination, is an important prerequisite for color-based object identification. Mechanisms of color constancy operate at several levels of visual information processing. Much is known about the sensory mechanisms of color constancy, whereas the role of higher-level mechanisms is far less understood. This paper focuses on the influence of previous experience, specifically memory colors, on color appearance and color constancy. The experimental paradigm was designed so that the data could be explained solely by interactions between different levels of visual information processing, minimizing confounding effects from semantic processing or instructions. Observers' task was to adjust the color of different natural objects displayed on a monitor so that they appeared in their typical colors, and in another run so that they appeared gray. The data showed a perceptual bias towards the typical colors of the objects in cases where the stimuli had all visual features such as shadowing and texture present. The effects were much reduced with recognizable but visually impoverished stimuli. This points strongly to the visual nature of the memory color effect. In a further experiment with man-made instead of natural objects the memory color effect was much weaker, although observers could still name the typical colors of the objects. This suggests a difference between the processing of natural and arbitrary color-form combinations, even when both are learned through experience.
Respondent: Alan Gilchrist, Rutgers University, Department of Psychology
Mark Wagner: Wagner College, Department of Psychology, Sensory and cognitive explanations for a century of size constancy research
The first (and largest) section of this essay reviews size constancy research over the last 90 years and summarizes the conclusions that may be drawn from this research. A meta-analysis of this research based on 118 data sets yields a number of conclusions. For frontally oriented stimuli under full-cue conditions, instructions greatly affect judgments with projective instructions showing strong underconstancy, apparent instructions approximating constancy on the average, objective instructions displaying overconstancy, and perspective instructions showing even stronger overconstancy. For flat stimuli, underconstancy is the rule for all instructions. Reduced cue conditions is also associated with underconstancy. A separate meta-analysis of the affects of age shows a tendency toward underconstancy in very young subjects, constancy for teenagers, and overconstancy for adults. These results are explained in terms of the transformation theory for size judgment.
The second part of this essay discusses attempts to separate size constancy judgments into sensory and cognitive components. I suggest that this distinction is not meaningful unless the sensory component is phenomenally experienced and reportable under some set of experimental conditions. Judgments under these reference conditions would make it possible to describe the affects of cognitive processes on judgments separately from sensory affects. Unfortunately, there is no unambiguous way to determine whether such reference conditions exist, since the only way to operationally define phenomenal experience is in terms of the judgments themselves. An alternative to the sensory/cognitive distinction is a Gestalt approach that sees different instructions as generating different mental organizations.
Respondent: Gary Hatfield, University of Pennsylvania, Department of Philosophy
Albert Yonas: University of Minnesota, Institute of Child Development, Perception of space in infants and adults for action and representation
Albert Yonas and Sherryse Mayo University of Minnesota
Institute of Child Development
One promise of studying perception in very young human infants and other species is to explore the kinds of perceptual processes that are available when intellectual problem solving strategies are not available. If abilities are available early we can rule out the view that perception is initially based on retinal properties and over development the infant acquires the experience and inferential skills needed to infer the distal properties of the world as Helmholtz argued. Studies of responsiveness to optical information for impending collision (looming as Gibson called this event) suggest that the visual system assumes that objects do not change in size if their retinal images symmetrically expand. Rather, very young infants blink at collision events when optical motions provide the relevant information.
While sensitivity to looming is the result of learning over evolutionary history, by 7 months of age infants respond to a pictorial depth cue that depends on experience, familiar size. When an object that varies little in real size is available, infants can learn the relationship between its retinal size and distance. It is also likely that other pictorial depth cues, such as linear perspective, are learned and become effective at about the same time. Given the early age at which these inferential processes are taking place, it seems unlikely that infants are making conscious guesses about the world. Rather, for information that is acquired through the experience of the individual, unconscious inferences occur early in development.
Recent work suggests that when adults are asked to reach for, rather than report, the size or location of the edges of an object, the influence of depth cues is quite different. When binocular depth information is present, adults reach to the actual location in depth of the edges of a trapezoidal window and ignore the illusion of slant. In contrast, they report that the window is slanted in depth. This work supports Goodales notion of a dorsal visual pathway that evolved to direct action and a ventral pathway that underlies recognition and higher cognitive processes.
Respondent: Gary Hatfield, University of Pennsylvania, Department of Philosophy