Avril, (01)
Every ontology has some built-in philosophical assumptions.
To quote Peirce, (02)
CSP
> Find a scientific man who proposes to get along without
> any metaphysics... and you have found one whose doctrines
> are thoroughly vitiated by the crude and uncriticized
> metaphysics with which they are packed. (03)
AS
> Does anyone know how the contemporary linguistic philosophy
> contributes to automated natural language understanding? (04)
Two points: (05)
1. It contributed a huge amount to the Carnap-Kripke-Chomsky-Montague
strand of formal linguistics. (06)
2. But Einstein criticized Russell's "Angst vor der Metaphysik" as
a dead end for science. That criticism also applies to point #1.
See http://www.jfsowa.com/pubs/worlds.pdf (07)
During the 1980s and '90s, many people tried to apply those theories
to practical NLP. But none of them were practical. Unfortunately,
that failure caused the NLP pendulum to swing to statistical methods
that reject logic-based methods completely -- or almost completely. (08)
One example (among many) is Bob Moore. He wrote his dissertation
and many papers about logic-based methods. But he switched to the
statistical methods when he joined Microsoft Research around 2001,
and he's continuing those methods at Google Research. (09)
I believe that statistics are useful as a supplement to symbolic
representations. But you can't represent semantics with just
statistics. For my recommended approach, see the following
article and the papers cited at the end: (010)
http://www.jfsowa.com/pubs/eg2cg.pdf (011)
AS
> When we order the computer to list all red objects, the computer lists
> all members of the class red. This is useful in the linguistic sense (012)
No, it's not. (013)
AS
> but over-propagating the class approach is not practical, as Putnam
> testifies: “Let us, then, keep our properties, while not in any way
> despising the useful work performed for us by our classes!” (014)
Putnam's article is a good source of some useful criticisms: (015)
Putnam, Hilary (1970) On properties, reprinted in Putnam (1975),
_Philosophical Papers_, vol 1, _Mathematics: Matter and Method_,
Chicago: University Press, pp. 305-322. (016)
He begins by citing some of his good buddies, but he continues with
what I believe is a devastating criticism of their assumptions. (017)
HP, op. cit. p. 305
> It has been maintained by such philosophers as Quine and Goodman that
> purely 'extensional' language suffices for all the purposes of properly
> formalized scientific discourse... (018)
Putnam continues with a discussion of physical properties. But he later
discusses "psychological" properties: (019)
HP, p. 313
> I am inclined to hold the view that psychological properties would be
> reduced not to physical properties in the usual sense..., but to
> _functional states_... [For example,] the property of being a finite
> automaton with a certain machine table... (020)
As the example illustrates, Putnam's "psychological properties" are
more computer-like than human-like. But if the extensional approach
is inadequate for a computer-like psychology, it would be even worse
for supporting human-like psychology or language based on it. (021)
The following slides say a bit more about some of the issues: (022)
http://www.jfsowa.com/talks/vague.pdf (023)
The concluding slide is copied below. See also Slides 10 and 11
about reasoning with images. This topic is related to the eg2cg.pdf
article cited above. (024)
John
________________________________________________________________ (025)
CONCLUSIONS (026)
The mapping from language to the world uses all the capabilities
of human intelligence and experience. (027)
The model-theoretic semantics for logic is too rigid:
● A finite set of symbols with fixed definitions.
● Two-valued denotations {T, F}.
● A formal algorithm for computing the denotations. (028)
A fixed set of word senses can be useful for a specialized task. (029)
But no fixed set of senses defined by a fixed ontology can
support the flexibility of human language and reasoning. (030)
More generally, no discrete set of ontological categories can
adequately represent a continuously variable world. (031)
Dynamic methods are needed to extend, revise, and supplement
the logic and ontology. (032)
_________________________________________________________________
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 (033)
|