|
Hi, I'm doing a graduate program in linguistics and wondered if there are any valid (research) careers for linguists in NLP? I attended some basic programming courses during my BA and I'm doing computational linguistic courses (focused on Jurafskys Speech...) as a part of my MA program. I can handle the programming-coursework but I lack formal training in math and informatics. Reading research papers it seems like machine learning / statistics are much more important than linguistic knowledge in current NLP research - am I missing something? Thank you very much! |
|
Yes, but only up to a point. Statistical NLP as practiced today started from a clean break with the heuristic-heavy approaches that preceded it. In this process, a lot of very useful and interesting ideas were thrown out together with the methodology that applied them, and this includes a lot of the more linguistic sides to NLP. However, this trend has been reversing recently. For example:
According to some high-profile people in the area more linguistically heavy NLP is the way to go, so if what is currently being done in a subfield of natural language processing seems naive to you, spend a bit of time looking at its structure and mistakes and see if you can improve it with your linguistic knowledge (and if you feel like you can, there's a lot of state of the art and nearly state of the art software out there for you to modify and test, plus computationally-inclined people in your lab/group can certainly co-work with you on interesting projects). 1
I agree with your answer. I'd add the advice to him to build on his linguistics and programming knowledge by getting a PhD in an NLP group where he can learn the math and statistics background as well, in a few classes and from his PhD advisor, e.g. at Columbia, Berkeley, Stanford, JHU, Edinburgh, ISI/USC, CMU and many other places.
(Oct 12 '10 at 14:13)
Frank
|
|
The works cited by Alexandre above can't really be called "linguistic" (maybe except for Raghunathan et.al which is borderline and Hal's which I just do not know). But there are more linguistic-oriented stuff going around (though maybe harder to find). Your name sounds European. If so, you are in luck -- much more linguistics stuff in Europe. Check out for example the people at Prague, and the computational linguistics group at Saarland university (there are many other places -- UVA to name one). For the US, check out the works of Martha Palmer (Colorado), Jason Baldridge (Texas), and the guys in Indiana (Markus Dickinson and Sandra Kubler). Also worth checking are Lilian Lee (Cornell) and the Columbia group. If you are into psycholinguistics, you may be interested in the work of Roger Levy (UCSD). There are others, of course, these were just the first to come to mind. One area which I is particularly suited for linguists in NLP these days is resource creation. Defining the interesting tasks. Coming up with good guidelines. Annotating the data. |
|
There's many computational linguistics researchers who started out in linguistics, and for that matter plenty of computational linguistics practitioners with a linguistics background as well. The basics of linguistics, often old-fashioned descriptive linguistics, tend to be the most useful, with the latest linguistic theory much less so. If already you're getting an MA in linguistics, then you'd ideally do a PhD in computer science, informatics, applied statistics, or (rare) in computational linguistics itself. Ideally somewhere that's strong in statistically oriented NLP, with good depth in machine learning. |
Thank you very much for your help!
The papers were very interesting, especially those by Hal, Haghighi and Klein and by Raghunathan et al. They may not be linguistic (research done by and for linguists?), but their approaches appeal to me, as they seem to be driven by linguistic intuition/research.
You're right, I'm German, currently in Jena - but getting a Ph.D. abroad is definitely an interesting idea.