Classifying Ellipsis in Dialogue: A Machine Learning Approach

Classifying Ellipsis in Dialogue: A Machine Learning Approach Raquel FERN´ANDEZ, Jonathan GINZBURG and Shalom LAPPIN Department of Computer Science King’s College London Strand, London WC2R 2LS, UK {raquel,ginzburg,lappin}@dcs.kcl.ac.uk Abstract This paper presents a machine learning approach to bare sluice disambiguation in dialogue. We ex- tract a set of heuristic principles from a corpus-based sample and formulate them as probabilistic Horn clauses. We then use the predicates of such clauses to create a set of domain independent features to an- notate an input dataset, and run two different ma- chine learning algorithms: SLIPPER, a rule-based learning algorithm, and TiMBL, a memory-based system. Both learners perform well, yielding simi- lar success rates of approx 90%. The results show that the features in terms of which we formulate our heuristic principles have significant predictive power, and that rules that closely resemble our Horn clauses can be learnt automatically from these features. 1 Introduction The phenomenon of sluicing—bare wh-phrases that exhibit a sentential meaning—constitutes an empirically important construction which has been understudied from both theoretical and computational perspectives. Most theoret- ical analyses (e.g. (Ross, 1969; Chung et al., 1995)), focus on embedded sluices considered out of any dialogue context. They rarely look at direct sluices—sluices used in queries to re- quest further elucidation of quantified parame- ters (e.g. (1a)). With a few isolated exceptions, these analyses also ignore a class of uses we refer to (following (Ginzburg and Sag, 2001) (G&S)) as reprise sluices. These are used to request clarification of reference of a constituent in a partially understood utterance, as in (1b). (1) a. Cassie: I know someone who’s a good kisser. Catherine: Who? [KP4, 512]1 b. Sue: You were getting a real panic then. Angela: When? [KB6, 1888] Our corpus investigation shows that the com- bined set of direct and reprise sluices constitutes 1This notation indicates the British National Corpus file (KP4) and the sluice sentence number (512). more than 75% of all sluices in the British Na- tional Corpus (BNC). In fact, they make up ap- prox. 33% of all wh-queries in the BNC. In previous work (Fern´andez et al., to ap- pear), we implemented G&S’s analysis of di- rect sluices as part of an interpretation module in a dialogue system. In this paper we apply machine learning techniques to extract rules for sluice classification in dialogue. In Section 2 we present our corpus study of classifying sluices into dialogue types and dis- cuss the methodology we used in this study. Section 3 analyses the distribution patterns we identify and considers possible explanations for these patterns. In Section 4 we identify a num- ber of heuristic principles for classifying each sluice dialogue type and formulate these prin- ciples as probability weighted Horn clauses. In Section 5, we then use the predicates of these clauses as features to annotate our corpus sam- ples of sluices, and run two machine learning algorithms on these data sets. The first ma- chine learner used, SLIPPER, extracts opti- mised rules for identifying sluice dialogue types that …

Content truncated to 500 words