LSA.235 | Why NLP Needs Linguistics: a case study

Annie Zaenen
MW 2:55-4:35
location: 32-144
course web site: http://lsa.dlp.mit.edu/Class/235

The evolution of computational linguistics from symbolic to statistical models has given the false impression that linguistic knowledge has become irrelevant for NLP. In fact nothing could be further from the truth: as soon as one moves away from simple Information Retrieval, one needs annotated corpora to be able to train the stochastic models that are en vogue. The problem is that the development of annotation schemes is often done rather haphazardly, creating the risk that the annotations based on them will in fact lead to incorrectly trained models. In this course we look at what kind of linguistic insights help in developing the right annotation schemes, at the kind of information that is needed to supplement these insights, and the skills that might be useful in that task. We will take coreference resolution as a case study.