Ji Wu asked:

I am now dealing with machine translation. As you know, the traditional method processing natural
language based on Chomsky grammar theory is a deterministic view. Now probability view gains
more interest and has more success stories. So my questions are:

(1) Can natural language be computed deterministically?

(2) Can probability explain all the aspects of natural language?

===========

1) Not at this point. Ultimately... well, that depends on your viewpoint on language, of course. What
Chomsky did, among other things, was to show that a formalistic, AI-inspired approach to language
analysis was at least more fruitful than any other to that point, and to validate, to a reasonable extent,
what might be termed "mental constructs" (e.g., see his article on Skinner). I don't actually think much
of the statistical approach to language, because I see it as both a) an approximation to metaphorical
(in the broad sense of that term) usage, b) because I don't see the neural basis for it (despite what
might be termed the "averaging" of inputs over ensembles of neurons), and c) because what kind of
grammar do you get purely from statistics? On the other hand, there's probably someamount of
Bayesian processes going on... why not? But I'm very strongly influenced by the West Coast cognitive
linguistic movement, which denies the Chomskian viewpoint. I'd like to see a bottom-up approach...
and of course so would many others... and that latter, pdf-inspired approach does, I think, rule out a
purely statistical one. So in answer to your question... I don't believe, for several reasons, that this will
happen.

2) No, given the above reasoning.

Steven Ravett Brown