David, (01)
What I presented in those two pages was a short summary of a project
that continued in one form or another for about 30 years. I did not
work on the TQA project, but I frequently talked with the people who
did and with the people who were working on other NLP projects at IBM. (02)
JFS
>> TQA was called "natural language", but I would call it a CNL, since
>> it was completely specified by a formal grammar and a mapping to
>> first-order logic. The users loved it, but it required a PhD linguist
>> at IBM Yorktown to update and extend the ontology. (03)
DE
> Sounds eminently practical (NOT!).
>
> How long did the PhD tolerate this task?
>
> After the PhD linguist moved on to more interesting tasks, how long did
> TQA survive? (04)
There were several PhD linguists who were working on TQA for many years.
Warren Plath, who had worked on machine translation in the 1960s, was
the project manager. The parser, written by Stan Petrick for his PhD
dissertation at MIT in 1965, was based on a version of Chomsky's
transformational grammar. In the early 1990s, when IBM was having
financial problems, Stan took the early retirement option from IBM
and became the chairman of the linguistics dept. at the U. of Wyoming. (05)
Fred Damerau, whom I cited in the paper, was well aware that
knowledge acquisition was critical if NLP was ever going to be
practical. He designed some tools that made it easier to update
the ontology. But there were three related issues: (06)
1. Updating the database with new relations and fields as new
applications were developed or extended. (07)
2. Updating the vocabulary with new words and annotating them
with syntactic and semantic features. Occasionally, the grammar
had to be extended to make the expressions more idiomatic.
(The base grammar was very general, but it was important to
support the forms that people most commonly used.) (08)
3. Updating the ontology for representing, relating, and reasoning
about the semantics of both the database and the vocabulary. (09)
Back in the 1980s, everybody realized that these three things were
related, and it would be very helpful if the tools for doing all
three were compatible. Developing such tools was a hot topic in
both the DB and AI fields. (010)
In the early 1990s, many of us who subscribe to Ontolog Forum were
working on projects that addressed these issues. WordNet is very
helpful for these tasks. We were hopeful that the Semantic Web
might help integrate them, but OWL went off in a direction that
is useless for any of them. (011)
The only useful parts of OWL are the ones that have been absorbed
in Schema.org. The obsession with decidability makes the notation
more cumbersome without any help for mapping NL <=> logic or
logic <=> database (relational or otherwise). (012)
For our work at VivoMind, we have found the many lexical resources
available on the WWW to be valuable for automated and semiautomated
extraction of ontologies directly from documents about a subject.
WordNet, Roget's Thesaurus, and many other resources are useful.
Standardized nomenclatures, including Schema.org, are important.
But we have found OWL to be of no value whatsoever. (013)
But we're not dogmatic. If any client has an OWL ontology (very
unlikely), we can happily map it to Prolog or other *undecidable*
languages and improve the performance by orders of magnitude. (014)
John (015)
_________________________________________________________________
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 (016)
|