Proxytype theory and/or Objectivist epistemology II

[dan2100]

Continuing with my overview of Jesse J. Prinz’s Furnishing the Mind: Concepts and Their Perceptual Basis, last time I went over his six "desiderata" -- the criteria he thinks any viable theory of concepts has to meet. (See http://www.objectivistliving.com/forums/index.php?showtopic=8342 ) Following the order in the book, after going over the desiderata, he evaluates some major rival theories against them. I'm leaving much out here and hope that the details can be filled in later in responses.

Imagism

The first rival he considers at length is "imagism" (not to be confused with the poetry movement), which he traces back to ancient times, though elaborate defenses of this view probably weren’t found until Berkeley and Hume. (What about Ancient and Medieval philosophers? It seems both imagism and definitionism (see below) have their roots in Ancient philosophy and representatives among Medieval philosophers. Also, John Locke is often classed as an imagist, but, as Prinz notes, he did not have a view of abstracting ideas from the details of any particular object.) Imagism holds that "concepts are derived from perceptual states" and are mental images – recollections or combinations of previous perceptions or impressions. (p26)

How does this fare against Prinz’s criteria? He believes it scores strongly on acquisition – namely that "concepts are acquired by storing perceptual states in memory" and "imaginative combinations" of these. (p27) This also goes for explaining "phylogenic acquisition" – or how such capabilities evolved in the first place. One criterion down, five to go!

He also believes imagism gives a good "account of cognitive content" – one could form different images or concepts from the same referent. (p27) This strikes me as another win for imagism.

He thinks it also does fair against the categorization criterion. Obviously, it explains some types of identification as "concepts are qualitative similar to percepts." To use his example, a perceived dog will be experienced as similar to and "can be directly compared to the DOG concept…" (p27) But see below on how this creates a problem.

This works for category production too because one can "read information off" concepts as images. The dog concept image would have fur, a tail, and bark properties. (Or would it? Different people might produce different dog concepts -- maybe yours might lack a tail or mine might lack bark. What if I'm a deaf person? Might my dog concept, if it's derived from my "image" of dogs,lack a bark property?) This might well explain typicality as typical members of a category are more like the concept as an image. (p28)

Prinz notes that imagism fails for the scope criterion. It’s hard to see how there be mental images for "nonperceptible entities or objects" such as "virtue, truth, [and] prime number." (p28) (Even so, one might think of how people who use logical operations often do use some form of imagery as an aid. Think of the use of Venn diagrams. See, e.g., Eric Schechter's Classical and Nonclassical Logics: An Introduction to the Mathematics of Propositions, especially chapters 3 and 4.)

He also presents a cogent case for imagism failing – at least in its traditional forms – against the intentional content, compositionality, and publicity criteria. On the last, it’s hard to see how people experiencing different objects – say, different dogs – and, therefore, having different mental images – your image of a dog might be like a golden retriever, mine like a bulldog – would share the same concept.

With compositionality, some combinations – Hume’s "golden mountain" (see Hume's An Inquiry Concerning Human Understanding) – fit snugly into the imagist account, but others – Prinz offers "carnivorous plant" – do not. In his example, it’s hard to figure what the image for a "carnivorous plant" concept should be. Needless to say, concepts for virtue and other nonperceptibles seem more daunting. Does "virtuous polity" conjure up an image? How about "transitive relation?"

Decomposition appears to work against imagism as well. He notes that once one has an image for a carnivorous plant, it’s hard to see how this can be broken apart or applied to other plants or carnivores – much less to other carnivorous plants. (p32)

Intentionality offers at least one hurdle: resemblance. Prinz points out that the dog concept should refer to dogs and only dogs, but its image resembles both dogs and wolves. Depending on how full-blooded this image is, it might resemble even more: stuffed animal toys, paintings of dogs, or films of dogs. (p30)

Overall, despite his sympathy for imagism - aspects of it are adopted into his proxytype theory - Prinz finds it wanting and so rejects it.

Definitionism

The second rival he considers at length is definitionism: "concepts are definitions." (p32) He locates its origins in ancient times with Plato’s view of Forms. He traces its adherents to more recent times in the views of Frege and Christopher Peacocke. (pp33-6)

He believes definitionism does quite well with the intentional content, compositionality, cognitive content, and publicity criteria. (p37-8) However, he notes it has "serious flaws" with the acquisition, categorization, and scope criteria. On the last, he uses Wittgenstein’s analysis of the concept of "game" and the notion of family resemblances. (Cf. Wittgenstein's Philosophical Investigations §66-8)

Kelley (The Art of Reasoning (1988)) believes Wittgenstein gives up too easily when searching for common, defining feature for games. However, this doesn’t rescue definitionism, but only shows that Wittgenstein’s criticism is not strong. For Kelley and for Rand, a major problem with definitionism is that it confuses part with whole – specifically confusing the definition of a concept with the concept. This bleeds into the whole theory, including acquisition. Rand notes that the definition of a concept usually comes after acquiring or recognizing it. (Cf. Introduction to Objectivist Epistemology 2/e, especially chapter 5.)

Prinz believes acquisition is a bigger hurdle, as definitionism seems to presuppose either some innate "seed" concepts or some non-conceptual starting points. The latter, he notes, are often met by alloying definitionism with imagism. (pp34-5) This mix is not without its problems, but I won't touch on them at this time.

His attack on definitionism’s failure to meet the categorization requirement seems, likewise, not strong. That base level concept not be of the high level categories could be due to context, though this shows concepts could not be just definitions. (pp42-4)

Overall, definitionism, like imagism, seems to pass some criteria, but fail others in some significant ways -- meaning, as you might guess, Prinz rejects it.

Prototype Theory

Next, he considers prototype theory. This theory views concepts as based on "typical features" – "ones that are diagnostic, statistically frequent, or salient" – and "unlike defining features, they are often contingent…" (p52) He points out that there "are a number of ways to make prototypes more precise" – "a prototype might be a point in a multi-dimensional space, a mental image of some real or idealized category instance, or a set of feature representations…" (p53) To illustrate the last, he uses a "set of binary features" to represent the concept "bird": "has wing, flies, has feathers, has a beak, eats worms."

The strengths of prototype theory should be obvious: it "explain typicality effects and basic level categories." (Typicality effects are things like how a more typical member of a category is readily identified -- e.g., a rose bush is more readily classed as a flowering plant than is grass.) It also "boast a plausible account of concept acquisition. Casual observation of category members is rarely sufficient for discovering a unifying definition, but it is sufficient for abstracting a prototype." (p56) He also believes it scores well on scope and cognitive content criteria, though I won't touch on these at this time.

Even so, he lists serious problems with it on the criteria of intentional content, compositionality, and publicity. He also notes that the high score on the categorization criterion is not without its problems. The problem with intentional content is that the more typical examples are more referred to or even fall under the concept to a "greater degree:" "… the bird prototype refers to sparrows more than it does to ostriches." (p59) (One might wonder if, as Wittgenstein is too quick to knock down definitions for the concept of game, Prinz is making up an easy to trash version of the bird prototype. Maybe the prototype could change so that having feathers becomes more central than flying or other features. This is perhaps the weakness of Prinz’s overall account: he does not have a robust account of concept change. Perhaps concept change could be a seventh criteria...)

He attempts that attempts to get around this problem, such as assigning a threshold to reference complicate the theory and seem dubious at best. For instance, to use another of his examples, "eels are fish, but they are utterly dissimilar to the FISH [sic] prototype" and "being above the threshold for the fish prototype is not necessary for being a fish." Add to this, "eels are very much like snakes in appearance and likely to exceed the threshold for the SNAKE prototype." (p60)

This seems to hold only if one is maintaining a very tight link between this prototype and a specific sense modality and even just a narrow range for that. Eels look like snakes, but they don’t feel like snakes. (They probably don’t taste like snakes either.) They also have a different bone structure, DNA, etc. They are not obligate air breathers. Also, just in terms of the visual approach, they might look like snakes at the gross level but are not covered in scales like snakes, do not act like snakes, or live under the same conditions as snakes. Even a casual and purely visual observer might note that. Prinz might not be testing the best version of prototype theory against his criteria.

Prototype theory’s failure with regard to compositionality is also notable. This is because typical properties or typical instances of compounded concepts tend to be atypical of their component concepts. His telling example is "pet fish" – they "typically live in bowls, but neither pets nor fish typically live in bowls…" (p62)

Similar to imagism, the publicity criterion is also tough on prototypes: it’s hard to see how, in many cases, two individuals would form and share the same prototypes. Prinz uses dogs to illustrate this, but let’s consider pets again. Someone exposed to more exotic pets is likely to have a very different "pet" prototype than someone exposed only to the "traditional" pets for Americans of cat, dog, goldfish, and parakeet. The latter is unlikely to include cricket, monkey, iguana, or chinchilla in her repertoire of pets – and unlikely to class them together. Back to dogs, Prinz points out that Larry Barsalou found that "typicality judgments vary considerably." This varies not only between people, but for the same person at different times. (I suspect the reader who thought of "pets" as dogs, cats, and goldfish, before reflecting on this, has a slightly wider prototype after reading this. I’ve noticed such a tendency in myself: exposure to new examples – even just from reading about them – seems to broaden and refine one’s concepts.)

Categorization, which seems a big plus for prototype theory, has a problem with regard to changes. Prinz cites Frank Keil's 1989 book Concepts, Kinds, and Conceptual Development regarding transforming a prototypical raccoon into a prototypical skunk via painting a white stripe down its back. (One can more easily think of the Pepé Le Pew cartoon where Pepe, the skunk, falls madly in love with a black cat that accidentally gets a white stripe painted down its back. I certainly don’t think – though Pepé Le Pew does – that the cat is now less a cat, more skunk. Pu does not see the transformation of the cat, so his belief is based only on the final appearance. This seems to point to Keil’s example not being good test. Cf. Keil pp159-82.) This example, though somewhat artificial, illustrates something deeper is happening with categorization – or people would find the painted raccoon more like a skunk and misclassify it often. (Again, I feel this might not be the best account of a prototype. If the prototype is limited to purely (visual?) surface appearances, then prototype theory faces a serious challenge when the raccoon is painted. However, if the prototype is expanded a bit to include other features of the raccoon, this doesn’t appear to be a good test at all – it’s easily dealt with by noting that raccoons can be painted to look like skunks. But it seems that snakes, octopodes, and tudor style homes cannot be painted to look like skunks.)

Prinz also points to how abstract prototypes might cause problems. He conjures up a prototype of "tropical fish" based on experiencing "a species… in which adults have blue scales and juveniles have yellow fins, but very few have both." The yellow-finned, blue-scaled prototype is unlikely to be close to any real world instance – the "typical" instance would be either but not both. (p64)

Again, the refrain is: prototype theory passes some criteria and fails others, so it fails as a general theory of concepts -- just like imagism and definitionism.

I hope to cover Prinz's evaluations of exemplar theory and "theory" theory also known as the "knowledge approach" next time. (On the latter, see Gregory L. Murphy's aptly named The Big Book of Concepts pp60ff.)

Edited March 3, 2010 by Dan Ust