Seminar on Computational Learning and Adaptation

  Generative and Discriminative Models in NLP: A Survey

Kristina Toutanova
Computer Science Department
Stanford University

Recently there have been a number of theoretical and empirical studies comparing generative and discriminative approaches to classification. The problem has also been studied by several researches in the context of Natural Language Processing problems. In this talk I survey general Machine Learning results in this area as well as the results of several empirical studies for NLP problems. To illustrate the characteristics of the two approaches to classification I compare generative and discriminative models for word-sense disambiguation, part-of-speech tagging, and syntactic parsing.

Return to the seminar schedule