On 8/19/06, Ben Goertzel <[EMAIL PROTECTED]> wrote:
> > The problem of context may be avoided by using an unambiguous language (for
> > internal representation). Context-dependent words are a feature of natural
> > language (NL) only. It arises when an NL word maps to multiple concepts in
> > the knowledge representation. It is necessary to separate NL processing
> > (ambiguous) with the internal KR (unambiguous).
>
> Well, no.
>
> It is true that NL has unnecessary ambiguities, but it is NOT true
> that a good knowledge representation will remove ALL ambiguities.
> Complete removal of ambiguity is an unachievable ideal, and trying to
> approach it too closely results in massive increase in the complexity
> of representations. Some degree of ambiguity is often useful to
> achieve representational compactness.
>
> And, context is not just about words having different meanings ---
> it's about concepts having different meanings in different situations
> ... which need to be determined **on the fly** by a contextualizing
> process, and can't be pre-figured up-front by decomposing every
> concept into a set of context-specific subconcepts...
> > internal representation). Context-dependent words are a feature of natural
> > language (NL) only. It arises when an NL word maps to multiple concepts in
> > the knowledge representation. It is necessary to separate NL processing
> > (ambiguous) with the internal KR (unambiguous).
>
> Well, no.
>
> It is true that NL has unnecessary ambiguities, but it is NOT true
> that a good knowledge representation will remove ALL ambiguities.
> Complete removal of ambiguity is an unachievable ideal, and trying to
> approach it too closely results in massive increase in the complexity
> of representations. Some degree of ambiguity is often useful to
> achieve representational compactness.
>
> And, context is not just about words having different meanings ---
> it's about concepts having different meanings in different situations
> ... which need to be determined **on the fly** by a contextualizing
> process, and can't be pre-figured up-front by decomposing every
> concept into a set of context-specific subconcepts...
Let's take the examples:
- "apple pie" means a pie with apple fillings.
- "pizza pie" is a pie that is like pizza (not a pie with pizza fillings).
- "apple juice" means a juice made from apples.
In #1 & #3, I don't see how you can generate the target concepts merely by mixing "apple" with "pie/juice" on the fly. In all these cases, the target concepts are actually complex concepts that are first defined in episodic memory (eg we know that there is this thing which is "a pie with apple fillings", and this thing is mapped to the phrase "apple pie" by the natural language processor).
So, complex concepts are formed in episodic memory first, using an internal representation that is unambiguous, because they are formed from and grounded by experience (ie sensory processing).
YKY
To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
