Marvelous reply! I enjoyed reading it, and so can hardly let the conversation slide by failing to respond. I fear I am too sophmoric in my reply, but I try.
2008/8/21 Wolfgang Teubert <w.teubert at bham.ac.uk>:
> Dear All,
> I am very grateful to Linas for his comments on our mental concepts of chairs and tables. He knows that it is 'something more' than what has been said in the discourse. I am not so sure. There have been many acts of what I call primordial speech situations in which chairs and tables have been pointed out to me. For me, such things are props that make up an extension of the speech situation. The act of ostentation make them part of the communication. If I had a dog it wouldn't know what a chair is. It would not even have noticed it had been part of such a speech situation. I recall discussions with people (not with dogs, though) in which we attempted, for instance, to negotiate if barstools with a back rest should be referred to as stools ("Could you bring two more stools over, please?"). Does the result of such speech situations and negotiations mould my mental concept? I think it does. Does each of us have the same (perhaps innate) concept of a chair (as Fodor sees it)? Rather unlikely. Does everybody have their individual concepts? If yes, how do we understand each other, except by assuming that meaning is in the discourse and not in our heads?
I really like the notion of language as being negotiated, as it does anchor many difficult linguistic phenomena. So, yes, "chairs" don't exist for dogs, because they do not participate in human language. But this does beg the question: might not some dogs have a mental concept of something, which, if they could only talk, they might call a "chair"? There are any number of dog-as-hero stories, where the dog brings the right item, performs the right action, for a given situation, which would lead one to conclude that dogs can make appropriate conceptual abstractions. Even more amazing is the recent research on tool-use in crows. We have no hope of teaching human language to a crow, and yet, one would have an extremely difficult time of arguing that crows do not possess the abstract notion of what we would call "a tool".
As a guiding principle for linguistics, it seems very correct to state that nothing exists outside of the corpus. Yet, one must not err into a debate as to whether qualia exist (qualia being, by definition, private, personal, incommunicable, "ineffable"). So: "my foot hurts" -- you cannot ever truly feel my pain. Language, the corpus, communicates the general notions of "foot" and "hurts", so that, one day, when your foot hurts, you'll be able to reflect, and think "this part of my body, we call it 'a foot', and this foot of mine, its emanating a sensation which we've negotiated a term for: it is 'pain". I'd argue that qualia exist.
I'm of the opinion that the sensation itself is ineffable -- indescribable, although real. Now, as to "the catch" in rowing: -- is it like a foot that hurts, unknowable till you actually experience it? My original intent was to say "yes, of course it is" -- authors and coaches have attempted to wrap words around it, with varying success, but there remains a body-sensation that remains apart from those words (and motivates the creation of those words).
There are factions in cognitive linguistics which argue that, ultimately, one cannot create a "talking machine" (AI) that correctly manipulates concepts linguistically, without also endowing that machine with senses. So, yes, corpus linguistics and cognitive linguistics are irreconcilable, at this level. But these same factions also make a claim that there are ways of uniting "rules", "predicates", "logics" (first-order logic, non-axiomatic logic) with probabilities (Markov chains) in such a way that these probabilistic logic networks can be used to learn and capture lexical items (by working from a corpus) *as well as*, with no alteration of the theory, learn and capture sensorimotor sensations. And so, at this deeper level, the claim is that cognitive and corpus linguistics can be reconciled, although one has to be abstract in doing so.
(To be specific: someone pointed to Susan Atkin's "Lexical Implication Rules" -- the goal is to have a machine automatically learn and deduce these; the presumption is that proabilistic logic can be used to learn these from a corpus, and that, once learned, these can be employed in some useful fashion in some other part of the system. No one has done so yet, and is unlikely to do so soon, but there is the general sense that this kind of stuff is not out of reach. Only time will tell what the reach will be.)