computational metalinguistics
Mar. 28th, 2005 01:54 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
trocheeseems to me that the goal of computational linguistics should not really be trying to encode what we know about language into computers.
the goal should be encoding how we learn about language. We [linguists] have a terribly unclear picture of what it means to have a good theory -- we talk and talk about minimality, elegance and Ockham's Razor, but have lousy metrics for quantifying the quality of a theory.
My department is choosing among several candidates for a new computational linguistics position.
porter spent his entire talk explaining how he mapped multiple databases onto the same format. There were no linguistics results, and the mapping wasn't automatic. I wasn't even convinced that he had read his slides before presenting them, either.
haze spent his entire talk using interesting methods on an interesting (if simple) problem. But he showed no interest in exploring why his methods worked -- a few directed questions from the engineers in the audience revealed that he had no interest in the methods, not even well enough to understand them. Any member of my lab would be better qualified, even those of us who are pre-Master's.
glass spent her talk exploring a technique that tries to learn how linguists analyze data, using some mocked-up linguistics results. I wasn't convinced by the utility of the problem she was trying to solve, but she followed the approach I believe in:
Unfortunately, I think that the faculty will hire porter. This is not helping my mood today.
the goal should be encoding how we learn about language. We [linguists] have a terribly unclear picture of what it means to have a good theory -- we talk and talk about minimality, elegance and Ockham's Razor, but have lousy metrics for quantifying the quality of a theory.
My department is choosing among several candidates for a new computational linguistics position.
porter spent his entire talk explaining how he mapped multiple databases onto the same format. There were no linguistics results, and the mapping wasn't automatic. I wasn't even convinced that he had read his slides before presenting them, either.
haze spent his entire talk using interesting methods on an interesting (if simple) problem. But he showed no interest in exploring why his methods worked -- a few directed questions from the engineers in the audience revealed that he had no interest in the methods, not even well enough to understand them. Any member of my lab would be better qualified, even those of us who are pre-Master's.
glass spent her talk exploring a technique that tries to learn how linguists analyze data, using some mocked-up linguistics results. I wasn't convinced by the utility of the problem she was trying to solve, but she followed the approach I believe in:
linguists seem to know a good solution when they see it. But they can't pin down how it's measured. Therefore let us use a number of exemplars of good (and bad) solutions and try to infer the metric for "good solution".This solution matches what I want.
Unfortunately, I think that the faculty will hire porter. This is not helping my mood today.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
no subject
Date: 2005-03-28 11:14 pm (UTC)no subject
Date: 2005-03-29 02:43 am (UTC)no subject
Date: 2005-03-29 02:57 am (UTC)no subject
Date: 2005-03-29 02:59 am (UTC)heuristic metalinguistics
Date: 2005-04-01 04:34 am (UTC)Does anyone know of good references on this subject, on-line or off? You can eMail me at kcoulter@magma.ca and thanks in advance if'n you do.
Mr. Kerry Coulter, Ottawa, Canada