Large Margin Methods for Part of Speech Tagging

Part of speech tagging, an important component of speech recognition systems, is a sequence labeling problem which involves inferring a state sequence from an observation sequence, where the state sequence encodes a labeling, annotation or segmentation of an observation sequence. In this paper we give an overview of discriminative methods developed for this problem. Special emphasis is put on large margin methods by generalizing multiclass Support Vector Machines and AdaBoost to the case of label sequences. Experimental evaluation on Part of Speech Tagging demonstrates the advantages of these models over classical approaches like Hidden Markov Models and their competitiveness with methods like Conditional Random Fields.