> There is linguistic-specific knowledge (which is learned), but no
> linguistic-specific inference rule (which is innate). The rules alone
> are not enough to produce human-level NLP  performance, though should
> be sufficient to learn the needed knowledge (given proper experience,
> of course).

A subtler point, however, is that the *adaptation of inference control
heuristics* that takes place as a consequence of reasoning about
physical events, may be *useful adaptation* for the purpose of
reasoning about linguistic structures.

This, I think, is the deeper reason that understanding the physical
world is useful as a crutch for later on understanding language.

To be more explicit:
In doing backward or forward chaining inference, a mind needs to use
experience to bias its choice of which inference rules to apply at
which steps.

It's obvious that linguistic reasoning experience is helpful in
biasing rule-choices in future linguistic reasoning instances; and
it's obvious that physical reasoning experience is helpful in biasing
rule-choices in future lphysical reasoning instances.

But it's not obvious that physical reasoning experience is helpful in
biasing rule-choices in future linguistic reasoning instances.  Yet I
suspect this is the case.

-- Ben

-------
To unsubscribe, change your address, or temporarily deactivate your 
subscription, 
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]

Reply via email to