Automatic Tripartite Classification of Intransitive Verbs

Copyright 2012 by Nitesh Surtani and Soma Paul 26th Pacific Asia Conference on Language,Information and Computation pages 446–455 Automatic Tripartite Classification of Intransitive Verbs Nitesh Surtani, Soma Paul Language Technologies Research Centre IIIT Hyderabad Hyderabad, Andhra Pradesh-500032 nitesh.surtaniug08@students.iiit.ac.in, soma@iiit.ac.in Abstract In this paper, we introduce a tripartite scheme for the classification of intransitive verbs for Hindi and claim it to be a more suitable model of classification than the classical binary unac- cusative/unergative classification. We develop a multi-class SVM classifier based model for automatic classification of intransitive verbs into proposed tripartite classes. We rank the unaccusative diagnostic tests for Hindi based on their authenticity in attesting an intransi- tive verb under unaccussative class. We show that the use of the ranking score in the fea- ture of the classifier improves the efficiency of the classification model even with a small amount of data. The empirical result illus- trates the fact that judicious use of linguistic knowledge builds a better classification model than the one that is purely statistical. 1 Introduction An automatic classification of verbs that are dis- tinct in terms of their syntactic behavior is a chal- lenging NLP task. Some works have been done for automatic determination of argument structure of verbs (Merlo and Stevenson, 2001) as well as automatic classification of verbs (Lapata and Brew, 1999; Schulte, 2000; Schulte, 2006) follow- ing (Levin, 1993) proposal. However, automatic sub-classification of intransitive verbs has not been attempted majorly till now. Sub-classification of in- transitive verbs has bearing on various NLP tasks such as machine translation, natural language gen- eration, parsing etc. For example, we take here a case from English-Hindi MT system. English uses nominative subject for all kinds of intransitive verbs whereas Hindi uses ergative case marker ‘ne’ on subject when the verb is unergative and in perfect tense whereas unaccusative doesn’t as exemplified in (1a) and (1b) respectively. (1) a. English: Ram ran a lot. Hindi: raam-ne khub dauRaa. Ram-erg very much run-3 pft b. English: The glass broke. Hindi: glaas TuT-aa. Glass break-3 pft Classifying intransitive verbs of (1a) and (1b) into subclasses can result in producing right case mark- ing on the subject in the target language Hindi. In parsing, identifying the subclass of the intransitive verb helps in predicting the position of the subject in the Phrase structure tree. One effort of sub- classification of intransitive verbs is described in Sorace (2000) where intransitive verbs are further automatically classified into unergative and unac- cusative following Perlmutter’s (1978) proposal of Unaccusativity Hypothesis. This paper follows the proposal of Surtani et al. (2011) where it has been argued that a a tripartite classification better clas- sify Hindi intransitive verbs. This paper develops a multi-class SVM classifier based model for the auto- matic classification of intransitive verbs in the tripar- tite classification scheme. We propose in this paper two approaches for developing multi-class classifier: (a) a Language dependent Classifier and (b) a Lan- guage Independent Classifier. 446 The paper is organized into the following subsec- tions. In …

Content truncated to 500 words