« The Other Entry I'm Adding Things To As The Day Drags On | Main | My Google-Fu fails me »
December 30, 2003
Fuckit, new entry it is
More deserving of one than that previous thing. . .
Ta! Interview with Bob Carpenter:Springboard
``In my first year at Carnegie Mellon Lori Levin was teaching Natural Language Processing, and she wanted something to give people that would do basically what PATR-II (Shieber et.al. 1983) did. So I wrote a simple Prolog implementation of PATR with a marginally useful interface and a manual. I took the Gazdar & Mellish textbook (Gazdar et.al. 1989) and did it the way I thought it should have been done, rather than the way it was done in their book. I worked as a TA for Gazdar and Mellish at the 1987 Stanford institute, where they kept me busy debugging the programs before their book came out. They attributed the unification-based system in their book to me, but the key unification algorithm was actually borrowed from Pereira and Shieber, and my wrapping of that system was modified beyond all recognition. It was the first time I built a system that I intended other people to use, and it was really the springboard in some ways for the implementation of ALE (Attribute Logic Engine).Frankenstein
``After moving to CMU in 1988, I shared an office with Carl Pollard. I really hated HPSG (Pollard et.al, 1987) at that time, because to me it looked like a complete Frankenstein theory. Not because it is built up from all these other grammar formalisms, like LFG and GPSG and CG and all that, but I had a real bias against the fact that it was using different sorts of mechanisms to do everything. You can look at Categorial Grammar, and you can say there is a pure logical reasoning mechanism on which everything is based. But in HPSG that uniformity comes only at the level of the feature structure logic. It is important not to confuse the kind of underlying attribute-value logic formalism with the linguistic theory. In HPSG there is this whole linguistic theory that contains things like the Head Feature Principle and the Quantifier Binding Condition and Principle A of the Binding Theory. That is the linguistic theory. HPSG should not be criticized on the fact that it is based on a Turing-powerful constraint resolution system. You just cannot do in the linguistic theory, everything you can do in the underlying theory. Shieber originally pointed this out in the context of PATR.``Now of course no linguistic theory, GB, HPSG, any of these, have really laid out a meta-theory of what really counts as a grammar. You will never find anybody saying: `A GB grammar is the following thing: one of these versions of this, one of these versions of that.' They talk about it that way, they say it is somehow parameterized, but you never actually see a list of parameters. No one can make a proposal so that someone will stand up in the audience and say: `Ah, but that is not a GB grammar!' Similarly, I could add a bunch of features and devices to HPSG and no one could say: `That is not HPSG'. You can say that it is not in the spirit of HPSG, or not in the spirit of GB, but it is really all just a matter of esthetics. So I think insofar as people really want to make theories of a universal grammar, then they should be honest and lay out the possible grammars. Again, GPSG perhaps came the closest to this goal in spirit, but it didn't really excite many linguists. Once you lay something out concretely, it's just a little too easy to see where it falls down.''
Don't worry, I had to look up some of the acronyms too.
Guess I could have linked the definitions, or included them here at the end.
Yep, could have done that.
Tra la la la la.
Or rather, Ta!:
This is the archive of the late, and sadly missed "Ta!", the Dutch students' journal for Computational Linguistics. The magazine was founded in 1993, and stopped about four years later. The homepage no longer exists, the editors have dispersed all over the world, but the interviews are still there, and we are doing all we can to not let them disappear with us. Here they are - enjoy!
Funny thing is, found this looking for an interview that Ted Briscoe conducted with Gerald Gazdar, and found I liked it better.
I had given up on linguistics by 1985. [. . .] I had come to the conclusion that the only way to get ideas across in syntax was to engage in a continuous marketing exercise. Thus Joan Bresnan had spent many years doing relentless marketing of LFG. But if I'd wanted to do that, I'd have gone into sales or something, not become an academic.
Gazdar is just so damned bitter, and I really don't like that sort of thing in a person.
What?
Posted by Aaron at December 30, 2003 12:07 PM
Trackback Pings
TrackBack URL for this entry:
http://www.uppity-negro.com/cgi-uppity-negro/mt-tb.cgi/1321
Listed below are links to weblogs that reference Fuckit, new entry it is:
Comments
LFG = Low Frequency Generator!
My people have a use for every module of the synthesizer.
Posted by: uppity-shinob at December 31, 2003 11:23 AM