Semi-Daily Journal Archive

The Blogspot archive of the weblog of J. Bradford DeLong, Professor of Economics and Chair of the PEIS major at U.C. Berkeley, a Research Associate of the National Bureau of Economic Research, and former Deputy Assistant Secretary of the U.S. Treasury.

Wednesday, August 23, 2006

Hoisted from Comments: Scott Martens (Who *Knows* Cog Sci and Linguistics) Denies Me a Cookie

He writes, at http://delong.typepad.com/sdj/2006/08/the_blue_car_is.html#comment-21414336:

"You see, I would argue, the neural circuits were well-engraved: an Acura Integra, the smaller of our cars, the more responsive of our cars, the non-station wagon--the features of the blue car were nearly identical to the features of the red car, so when our brains grasped for a verbal referent they had a good chance of picking the standard phrase we used for the red car. And, of course, neither of us had any trouble understanding what the other meant by the phrase "the red car.""

Bad economist. No cookie for you. Go get Marx' "Theses on Feuerbach" off the shelf. Open to page one:

"The main defect of all hitherto-existing materialism - that of Feuerbach included - is that the Object, actuality, sensuousness, are conceived only in the form of the object, or of contemplation, but not as human sensuous activity, practice, not subjectively."

Do not ask what properties your old red car possessed. Ask: How did the red car fit into your system of praxis? Did the new car fit into the same place?

Then, imagine momentarily that you possessed a transcript of all speech within the Delong household from 1988 (when you bought the red car) to its replacement in 2001. Measure the rate of co-occurance of "red" and "car" in this transcript using any of a number of basic, non-connectionist, techniques. Mutual information is the simplest, but I'm doing my thesis on minimum description length, which is more complicated but has other advantages. Using mutual information, you could compare the entropy of "car" with "red car", and will find that they are at least comparable. "Car" paired with other adjectives will most likely not have comparable entropy.

So, on the one hand we have an object of praxis that has a significant role in your life, and on the other we have a conjunction of words occurring in a pattern comparable to a single word. This is no coincidence, comrade. You could have continued to use the phrase "red car" to indicate the new car. You did not do so because you took some effort not to, either for the sake of an imagined linguistic norm or to ease communication with non-Delongs with whom you might need to discuss your Acura. Most likely the first.

There is no need to invoke features here. The object as practice covers this ground far more easily.

The occurrence of priming in this case - the return to saying "the red car" - is a sign that this type of semiotic creation is a behavioral norm which you have consciously chosen to violate. In fact, you've acted in a manner contrary to the linguistic norm. A "Tudor house" is an exact analogy to the "red car" - you're unlikely to see one in Berkeley that was built when the Tudors were on the throne.

Now, there was no need to invoke connectionism in explaining this outcome. Neurons were, to be sure, involved. But so were hands, feet and carburators. As for modeling, quite non-connectionist information theory was up to the task.

Shah here has it partly right:

"Any real discussion has to has some discussion about the interactions and interfaces between all of that mess, because it's pretty much certain (to me) that the emergent properties of both language and consciousness arises from contact points within all the systems in the brain..."

At least as important to symbol use and production are the contact points between physical cognitive devices within our bodies and our practices and (in Marx' sense of the word) human sensuous activities.

But I have issues with Feldmann here:

"For embodied cognitive science, any computational-level formalism must be effectively reducible to the connectionist level and thus to brain mechanisms."

No, no, no, no! Nothing reduces to brain mechanisms because (thumping the table) BRAINS DON'T SIT IN GLASS JARS!!!!! This used to be one of the key tenets of embodied cognition as a research program.

Have we forgotten Bateson's blind man?

"Suppose I am a blind man, and I use a stick. I go tap, tap, tap. Where do I start? Is my mental system bounded at the handle of the stick? Is it bounded by my skin? Does it start halfway up the stick? Does it start at the tip of the stick? But these are nonsense questions. The stick is a pathway along which transforms of difference are being transmitted. The way to delineate the system is to draw the limiting line in such a way that you do not cut any of these pathways in ways which leave things inexplicable."

This basic failure of "embodied cognitive science" to actually embody its cognition in actually existing bodies is the thing that I most have against it. It may be necessary to have a computational level of description for human cognitive activity - it seems reasonable enough - but there can never be a NEURAL level which is complete, descriptively adequate and consistent with macro observation because (more thumping) COGNITION IS NOT CONTAINED IN BRAINS!

I'm all in favor of studying brains. I support research into connectionist algorithms and alternative computing principles. But I do not imagine that the solution to the problems of linguistics is primarily to be found in those things.

And I think that is one area where Feldmann and I are really on different pages. He slags the Chomskyans for ignoring brain research when it goes against them, which is right on because their whole research program is based on the idea that the brain must contain certain things. But I don't think connectionism or brain research are really going to bring clarity to much in linguistics either.

It's possible that in this respect we only differ in emphasis and vocabulary. I haven't read his book yet.

As for grounding things in metaphor... I was once an advocate of this approach. I'm not anymore. The information theoric explanation I offered for the "red car" also offers an alternative perspective on metaphor. Rather than seeing metaphor as a consequence of neural association, it makes at least as much sense to see neural association as a consequence of metaphoric usage.

Instead of making metaphors central to cognition, we might reread some Saussure. There is a meaningful, practical, opposition between "This situation is good" and "This situation is bad". We not only think different things when we hear the one and not the other, we do different things, and the things we do have different consequences depending on the objective conditions underlying those statements.

For Saussure, the existence of this opposition is central to linguistics and semiotics. Oppositions of the same sort are central to information theory and computing.

However, there is no opposition between "This situation is bad" and "This situation stinks". That fact alone would, for a lexicographer, be evidence that "to be bad" is the definition of "to stink" in some contexts. Now, situations cannot literally stink. If they could, there would be an opposition between "this situation is bad" and "this situation stinks" because they would entail different actions and different consequences.

This lack of opposition makes it possible for "this situation stinks" to take some type of metaphorical meaning. The semantic overlap between "to stink" and "to be bad" - that things that stink are not generally positively regarded - in combination with the practical fact that what we often wish to communicate about situations is whether they are positive or negative from some perspective, makes this metaphor comprehensible. Regular use, in turn, makes it conventional.

This explanation seems more likely to me because the kinds of metaphors he is invoking are not actually used all that systematically. Even though "bad" can be systematically expressed as a bad smell, the English language does not systematically express good as a good smell. You can't say "this situation smells like roses".

Again, the mechanisms involved here do include some brain activity, although it is possible to model them using quite non-connectionist explanations, but more importantly they involve practical knowledge of the conditions under which language is used. They turn on activities that go on outside the brain.

Here, I think the cognitivist program in the US and I have to go our separate ways. They tend to see neurobiological modeling and metaphor as being at the root of cognition and semiotics. I tend to see praxis and opposition as central instead. There is some compatibility here, but I think in the end will all be for naught.

It bothers me to see linguistics so easily set adrift from its fundamental Saussurian roots, and it bothers me to see so much of what academic Marxism did right - placing a huge huge emphasis on practice - forgotten as if it never existed.

0 Comments:

Post a Comment

<< Home