Fast Domain Adaptation for Part of Speech Tagging for Dialogues

Proceedings of Recent Advances in Natural Language Processing, pages 41–48, Hissar, Bulgaria, 12-14 September 2011. Fast Domain Adaptation for Part of Speech Tagging for Dialogues Sandra K¨ubler, Eric Baucom Indiana University {skuebler,eabaucom}@indiana.edu Abstract Part of speech tagging accuracy deterio- rates severely when a tagger is used out of domain. We investigate a fast method for domain adaptation, which provides ad- ditional in-domain training data from an unannotated data set by applying POS tag- gers with different biases to the unanno- tated data set and then choosing the set of sentences on which the taggers agree. We show that we improve the accuracy of a trigram tagger, TnT, from 85.77% to 86.10%. In order to improve perfor- mance on unknown words, we investigate using active learning for learning ambi- guity classes of domain specific words, yielding an accuracy of 89.15% for TnT. 1 Introduction Part of speech (POS) tagging for English is often considered a solved problem. There are well es- tablished approaches such as Markov model tri- gram taggers (Brants, 2000), maximum entropy taggers (Ratnaparkhi, 1996), or Support Vector Machine based taggers (Gim´enez and M`arquez, 2004), and accuracy reaches approximately 97%. However, most experiments in POS tagging for English have concentrated on data from the Penn Treebank (Marcus et al., 1993). If POS taggers trained on the Penn Treebank are used to tag data from other domains, accuracy deteriorates signif- icantly. Blitzer et al. (2006) apply structural cor- respondence learning for learning pivot features to increase accuracy in the target domain. However, their approach is restricted to discriminative ap- proaches to POS tagging. In this paper, we investigate a simple and fast method for domain adaptation that is usable with any POS tagger: selecting reliably tagged in- domain data to add to the training set. This method has been successful for domain adaptation for de- pendency parsing (Chen et al., 2008). We use a corpus of dialogues collected in a collaborative task as target domain, thus introducing the chal- lenges of processing spontaneous speech: hesita- tions, corrections, false starts, and contractions. We assume that this domain is more challenging than a target domain of biomedical texts, which is often used for domain adaptation research. Spon- taneous speech dialogues do not only differ in ter- minology, but also in the types of sentences. Di- alogues, for example, contain a higher percentage of questions and imperatives than formal written language, such as news or scientific writings. Our domain adaptation experiments concentrate on adding in-domain training data based on an en- semble of POS taggers. The experiments show that extending the training set generally improves POS tagging accuracy. However, it cannot provide information on the ambiguity classes for words that do not appear in the source domain. For this reason, we integrate an active learning strategy for adding ambiguity classes for words that are iden- tified automatically as unlikely to be tagged cor- rectly. The remainder of the paper is structured as fol- lows: In section 2, we provide an overview of do- …

Content truncated to 500 words