Avril, Pat C, and Ed, (01)
General point: I agree with Martin Kay that language understanding
is an "AI complete" problem. It won't be solved until all the other
problems of artificial intelligence have been solved. (02)
Martin made that point over 20 years ago. But I believe that today's
technology is sufficiently advanced that we can address those AI
complete problems. But to do so, we must recognize that they are
indeed *AI complete*. They can't be solved with just one algorithm,
one logic, one ontology, or one methodology.
AS
> If class nominalism finds application, it finds it in NL processing
> rather than in chemistry. (03)
According to the above principle, it's a mistake to partition complex
AI problems as NL vs. formal logic vs. common sense vs. anything else.
You can't predict in advance which tools in the AI tool kit may be
necessary or irrelevant for each of them. (04)
JFS
> I believe that statistics are useful as a supplement to symbolic
> representations. But you can't represent semantics with just
> statistics. (05)
PC
> I agree, but in the context of natural language understanding,
> I think that a useful collaboration between statistics and logic
> would parallel the "Thinking Fast and Slow" model of Kahneman,
> where the (fast) intuitive pattern-matching of language input
> presents candidates as possible interpretations, which then
> need the (slow) logical process to determine whether a preliminary
> interpretation makes sense in the current context. (06)
I mostly agree with that point. But I'd add two qualifications: (07)
1. I don't believe that there is any difference between methods
required for natural language understanding or for commonsense
reasoning. Both of require similar methods and mechanisms. (08)
2. Instead of the fast/slow distinction, I'd prefer the terms
shallow/deep. In general, shallow methods are faster than
deep methods. But the differences in speed depend heavily
on the algorithms and the hardware. For many applications,
good technology can make the deeper methods fast enough to
compete with the shallow methods -- and *solve* problems
that the shallow methods cannot even state. (09)
EB
> I intend to remain silent on the subject of the relationship
> of computational ontologies to philosophical ontology. To me,
> that is “How is a raven like a writing desk?” (010)
I agree that the people who work in those two fields have
different goals. But there are insights and techniques that
each can contribute to the other. (011)
John (012)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J (013)
|