The Taos
Discussion
Monday, September 12, 2005
Communication from John Sowa
Following is a note I sent to CG (Cognitive Graphs) list as part of a discussion about the evolution of language. In particular, I recommend the interview with Terrence Deacon (some excerpts included below).
I heartily recommend Paul Bloom's book, "How Children Learn The Meaning Of Words". It’s quite fascinating and very readable. The whole process is astonishing and complex. One feature that seems to exist even in very young children is called in the book, "theory of mind", meaning that the child has notions about how others (especially adults) must be thinking to use a word in a particular way. Your paragraph above brought this to mind. If we have to understand, and then model, any large part of these processes that seem to go on in children, I think it will be a long time before we get machines that handle language with good facility.
Terry Deacon, who recently gave the following interview:
http://www.childrenofthecode.org/interviews/deacon.htm
The Co-Evolution of Language and the Brain
By training and research specialization, Deacon is a neurophysiologist, but he has also studied a great deal of linguistics and anthropology (and, by the way, Peirce's semiotics). I have excerpted some of his remarks below.
"A Pinker + Jackendoff article suggests that techniques from artificial intelligence can serve to build computer models with which we can shed light on the mystery of language origins and evolution." Too bad it is just the *opposite* approach that we need.
I agree that the "opposite approach" would be exceedingly valuable *if* it could tell us anything at all about how to design our computer systems, but the most we have learned is that all current models are hopelessly inadequate and some are more inadequate than others.
And for what it's worth, most of the better models have come from people with some knowledge of Artificial Intelligence (AI) and computer science. Look at the interview with Terry Deacon, who has done quite a bit of study of AI and NLP (Natural Language Processing).
The Pinker + Jackendoff article is probably not one of the best sources of insight because it is part of an enraged bickering between Hauser, Chomsky, and Fitch on one side and Steven Pinker and Ray Jackendoff on the other and it is full of details in support of the argumentation rather than analysis of the possible structures at work.
But what I find most encouraging about the article is precisely the fact that the *linguists* Pinker + Jackendoff have demolished Chomsky's longstanding arguments, which have for many years distracted attention from more promising approaches. It's especially important that Jackendoff coauthored that article, since he has been one of Chomsky's strongest supporters for many, many years. If Jackendoff criticizes Chomsky, other linguists who hope for tenure will be willing to follow.
But in any case, Jackendoff is one of the linguists I consider somewhat misguided, although less so than his mentor. Splitting Jackendoff from Chomsky is progress, albeit rather slow.
I find Tomasello hypothesis much more illuminating: If the main difference between "ape's thought" and ours is that we identify with conspecifics and try to model their intentions, then this feature introduces the need for a generalizable "meta" capability, thinking about the thinking. This feature leads us to think about our own thinking. This brings about Chomsky's "recursion" without the need to hypothesize a built-in intricate structure to start with.
Deacon emphasized the use of symbols, which is the central theme of his book, _The Symbolic Species_ (which I highly recommend).
We need to understand processes not "nature"...
Could it be that "nature" and "process" are interdependent concepts? Could you identify a "process" without invoking constituents of that process in interaction? And isn't such an identification in fact projecting a concept of the "nature" of that process, as well as the "nature" of those constituents? To invoke a "nature" is only to indicate an intent to reveal the "point" of the term in a discourse.
I cannot believe that one can begin to understand the processes without at least some understanding of the nature -- and vice-versa. The two are inseparable.
John Sowa