Notes On Kosslyn "Mental Imagery"

(NB: To date in this course, we have been talking fairly generically about how information processing systems manipulate complex tokens. We have not spent a great deal of time describing the specific properties of tokens (and of the rules for manipulating them) that may exist in the human information processor. The first half of Kosslyn's chapter is designed to do this. Mental images, as far as course content is concerned, can plausibly be viewed as candidates for complex tokens. Our ability to inspect and transform these tokens are examples of aplying specific rules to mental images. Thus mental imagery is one possible component of a classical account of human cognition.

The second half of Kosslyn's chapter is designed to illustrate a deeper, and more challenging, issue: Is there any reason to believe that mental imagery is part of the functional architecture? Kosslyn made a very big name for himself by carrying out a number of clever experiments to make the claim that images are indeed primitive. As a result, it is no surprise that this is the conclusion drawn in the chapter. Beware of this though -- the imagery debate is one of the most complex in cognitive science, and many questions have been raised about Kosslyn's work and about his conclusions. If you are interested in getting a detailed look at both sides of this debate, then see me and I can point you to additional readings.

Purposes And Problems

Many people solve spaital tasks by "seeing" with their "mind's eye" -- they use mental imagery. But what is mental imagery? Imagery bears a strong resemblance to perception, and thus has shared purposes. (1) Imagery can be used to recognize properties of visual objects. (2) Imagery can be used to retrieve spatial information from memory, particularly information that is difficult to access using other means. (3) Imagery can also be used to anticipate spatial information.

If a computer program was to be created that mimiced imagery, it would have to solve for different problems: (1) generate an image, (2) inspect the image, (3) retain the imagery in memory long enough for it to be inspected, and (4) transform the image to solve spatial problems.

The Human Processing System

How does the human brain solve these four subproblems?

At least two different subsystems are involved in image generation. The first builds images on the basis of distinct parts that are activated individually. As a result, it takes longer to create an image of a many-parts ojbect than of a few-parts object. The second subsystem positions parts in the image. These two subsystems are analogous to the distinction between two separate processing streams in the human brain, one for processing object appearances, the other for processing object locations.

"Objects seem to be `inspected' in imagery just as they are in actual perception: once a representation is formed, it apparently is treated the same way regardless of whether it arose from the sense or from memory." Psychophysical evidence -- such as modality specific interference -- supports this claim. So does neuropsychological data. For example, visual neglect produced by brain damage is duplicated in similar neglect of visual images.

We can retain relatively little information in an image at once, where the critical measure of information is "number of chunks." This should be expected if imagery utilizes the same anatomical pathways as vision, because one of the major functional goals of vision is to throw as much unneccesary information away as is possible.

Images can be rotated, expanded, shrunk, and folded. (NOTE: with respect to course content, these are some of the formal operations that can be applied to images when images are viewed as complex tokens.)

Individual Differences

There are many different aspects of imagery, consistent with the claim that it is subsuemed by many subsystems, which in turn are likely related to many different brain areas. As a result, imagery is not an "all or none" trait that can be used to characterize individual differences (e.g., good imagers vs. bad imagers) in any meaningful way.

What Is A Mental Image?

Much effort has been expended in recent years to understand what a mental image is. (NB: for course content, this issue reads: how is mental imagery embodied in the functional architecture?) "No one denies that people experience `seeing with the mind's eye', but there is controversy over what this experience reveals about how the brain actually stores the relevant information." There are two proposals -- propositional representations vs. depictive representations.

Propositions are a language-like code that involves putting together strings of concepts and relations between them: ON(telephone,desk) represents the notion that "the telephone is one the desk." In this type of code, meaning is assigned to symbols arbitrarily, it is an abstract code, and it may have a true/false semantics.

Depictive representations differ from propositional representations in almost every respect. Their syntax is a spatial arrangement of points and empty space. Their semantics comes from resemblance to the things that they represent. Depictive space can be defined functionally; one need not propsoe the existence of a physical "picture in the head."

The Imagery Debate

Key issue: Are deptictive properties part of the functional architecture, or are they merely epiphenomenal? (NB: A less emotionally charged way of putting this qeustion is to ask whether images exist at the bottom level of a functional analysis, or whether they are constructed out of simpler functional components.) Cognitive psychology has conducted numerous experiments to get at this issue.

"By their very nature, depictions embody space (recall that `distance' is an intrinsic part of the representation). Thus, if depictive representaions underlie the experience of `having an image', then the spatial nature of the representaiton should affect how images are processed." In contrast, space is not an intrinsic property of propositional images. So, to get at the functional architecture issue, one can run experiments to find out if spattial properties of mental images are obligatory (suggesting their primitive nature) or not (suggesting that they arise from simpler, nonspatial things.)

The intrinsic spatial nature of images was tested by Kosslyn and his colleagues via scanning experiments. Does scanning time increase as a fucntion of increases in image distance? The answer to this question is yes, according to a number of Kossln's results, the most famous of which involve the scanning of an imagined map.

The issue that next arose concerning this research was whether these results were due to demand characteristics of the experiments. Did experimental instrucitons provide tacit knowledge to subjects which produced the relationship between time and distance? If it did, then imagery is cognitively penetrable, and is not part of the functional architecture. Several experiments have attempted to test this directly, with mixed results. Propositional supporters find that changes in instruction change subject performance. Depictive supporters find that these changes do not occur.

For Kosslyn, the critical information for resolving the imagery debate came from neuroscience. Three key facts are important: (1) Some visual areas of the brain are topographically organized. (2) Connections between visual areas are bidirectional -- information in one of these areas might be from bottom-up, or from top-down. (3) Visual memories are stored in a part of the brain that is *not* topographically organized. "These facts are consistent with the notion that visual memories are stored in an abstract (propositional?) format and that an image is formed in order to make accessible information about the local geometry of a shape." From this perspective, when images are created, they will literally be "pictures in the head."

PET studies have been used to confirm this theory. "We found that parts of the brain known to be topographically organized in humans are active during visual mental imagery, even when subjects close their eyes. But is this merely epiphenomenal? Apparently not -- brain damage to topographically organized visual areas also impairs mental imagery. So, "we not only have behavioral evidence that distance is an intrinsic part of image representations, but we also have neuroloigcal evidence shwoing that topographically organized parts of the brain play a key role in imagery." Kosslyn'g conclusion -- depictive representations underlie imagery -- but not solely so!

Images And Brain

"The experiments summarized here illustrate how behaivoral data can be combined with neurological results, allowing us to distinguish among alternative theories of mental representation."

Pearl Street | "An Invitation To Cognitive Science" Home Page | Dawson Home Page |