Chris, (01)
On Wed, Oct 1, 2008 at 11:40 PM, Christopher Menzel <cmenzel@xxxxxxxx> wrote:
>
> You have yet to identify a single problem. Can you articulate one? (02)
Knowledge soup. (03)
>> I referenced once before here Giuseppe Vitiello's lecture on a basis
>> for concepts in the dynamics of many-body systems. Perhaps it bears
>> repeating:
>>
>> http://www.archive.org/details/Redwood_Center_2007_01_23_Giuseppe_Vitiello
>
> That you think this work has any direct bearing on AI/OE suggests to
> me that perhaps you are under a misimpression of the subject matter of
> these disciplines. (04)
On what basis do you conflate AI and OE? (05)
>> VSA's are especially good. They seem to present (compositional)
>> representation as a vector product in much the way I've been
>> advocating.
>
> You do recognize that VSAs are a move in the connectionist AI
> paradigm, which of course has many problems and challenges of its own,
> right? And that connectionism itself, while a promising approach,
> hasn't exactly delivered the goods either, right? And that it's not
> clearly even a competitor to the logicist paradigm? You might find
> the paper "Is the connectionist-logicist clash one of AI's wonderful
> red herrings?" by Selmer Bringsjord worth reading in this regard:
>
> http://kryten.mm.rpi.edu/connectionist_logicist_clash.pdf (06)
If you looked carefully you would see that Bringsjord's thesis fits
very well with what I have been saying. (07)
Wasn't a reconciliation of the connectionist and logicist paradigms
the central theme of Simon Levy's presentation in particular, of the
references I gave? Here's the link to him again: (08)
E.g. Simon Levy presents VSA's in the context of "The Need for New
Representational Principles"
(http://www.chaoticlanguage.com/node/37.) (09)
Paper: (010)
http://www.cs.wlu.edu/~levy/pubs/agi_2008_levy_gayler.pdf (011)
And presentation: (012)
http://video.google.com/videoplay?docid=-6666777138089848257 (013)
>> From my point of view all we need to do is mix the idea of a vector
>> product giving a compositional character (VSA's), with example/usage-
>> based ideas already current in linguistics.
>
> Sounds good. Care to sketch how that might be done? Just a hint? (014)
Sure. (015)
People commonly use vectors of contexts to represent word meaning.
Unfortunately so far these have all modeled compositionality poorly.
Compositionality is where the logicist paradigm has excelled. The
suggestion is we could get a good model of word compositionality,
while keeping the flexibility of connectionist/distributed
representation, by implementing the combination of these vectors of
word contexts as a vector/outer product (c.f. the traditional
connectionist inner product.) (016)
It is not so hard to do. I built a system to break natural language
sentences into meaningful parts on this basis. (017)
If anyone wants to test my implementation they can contact me and I
will see if I can get them a working version. (018)
Of course it is the complexity aspect of such a model which makes it
interesting to do this (it can project out more classes than examples,
so a different class for each context/combination.) That it has not
been tried up to now is largely because, even in the connectionist/PDP
community, they have failed to see this complexity issue (c.f. Chris
Anderson's no Theory of Everything etc.) (019)
-Rob (020)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: 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 Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (021)
|