>Thank you both.
>
[quoted text clipped - 43 lines]
>they need someone to wash the dishes at CL/NLP? I could do
>that....Need a job!!!

Signature
Peter Duncanson, UK
(in alt.usage.english)
On Sep 28, 4:34 am, "Peter Duncanson (BrE)" <m...@peterduncanson.net>
wrote:
> >Thank you both.
>
[quoted text clipped - 61 lines]
> Peter Duncanson, UK
> (in alt.usage.english)
Yup. CL/NLP (for some reason they can't settle on
which one to use, and they both encompass the
same intellectual territory) is what allows, for example,
automatic translation, or text-to-speech, or speech-to-text,
or pinyin-to-characters, or data mining.
Et very profitable cetera.
Most of it is done with statistics, and the statistics
are generated from large (ideally multigigaword)
text or speech corpora (pl. of "corpus"), of which
there are now many, with more appearing every day.
These include ALL uses of words, including the
unusual, the unclear, and the ill-chosen. And
parsers are useless unless they provide ALL
the possible parses for a potentially ambiguous
sentence. So, at least in English, where most
words can be at least verb, adjective, or noun
(transitive, intransitive, descriptive, proper,
or common), this means that virtually every
sentence is automatically multiply ambiguous.
Therefore disambiguation is a big big headache
for CL/NLP. And any way to automatically
disambiguate is going to be worth money.
That's what I meant above.
References:
1) http://www.umich.edu/~jlawler/routledge/book-7.doc
(2 chapters on CL/NLP from my book
"Using Computers in Linguistics")
2) http://www.umich.edu/~jlawler/abney95c.pdf
(a famous paper by my colleague Steve Abney
on the problems of ambiguity in CL/NLP)
-John Lawler
Don't anthropomorphize computers. They hate that.