A Twin-Candidate Based Approach for Event Pronoun Resolution using Composite Kernel

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 188–196, Beijing, August 2010 A Twin-Candidate Based Approach for Event Pronoun Resolution us- ing Composite Kernel Chen Bin1 Su Jian2 Tan Chew Lim1 1National University of Singapore 2Institute for Inforcomm Research, A-STAR {chenbin,tancl}@comp.nus.edu.sg sujian@i2r.a-star.edu.sg Abstract Event Anaphora Resolution is an important task for cascaded event template extraction and other NLP study. In this paper, we provide a first systematic study of resolving pronouns to their event verb antecedents for general purpose. First, we explore various positional, lexical and syntactic features useful for the event pronoun resolution. We further explore tree kernel to model structural information embedded in syntactic parses. A composite kernel is then used to combine the above di- verse information. In addition, we employed a twin-candidate based preferences learning model to capture the pair wise candidates’ pre- ference knowledge. Besides we also look into the incorporation of the negative training in- stances with anaphoric pronouns whose ante- cedents are not verbs. Although these negative training instances are not used in previous study on anaphora resolution, our study shows that they are very useful for the final resolu- tion through random sampling strategy. Our experiments demonstrate that it’s meaningful to keep certain training data as development data to help SVM select a more accurate hyper plane which provides significant improvement over the default setting with all training data. 1 Introduction Anaphora resolution, the task of resolving a giv- en text expression to its referred expression in prior texts, is important for intelligent text processing systems. Most previous works on anaphora resolution mainly aims at object ana- phora in which both the anaphor and its antece- dent are mentions of the same real world objects In contrast, an event anaphora as first defined in (Asher, 1993) is an anaphoric reference to an event, fact, and proposition which is representa- tive of eventuality and abstract entity. Consider the following example: This was an all-white, all-Christian community that all the sudden was taken over -- not taken over, that's a very bad choice of words, but [in- vaded]1 by, perhaps different groups. [It]2 began when a Hasidic Jewish family bought one of the town's two meat-packing plants 13 years ago. The anaphor [It]2 in the above example refers back to an event, “all-white and all-Christian city of Postville is diluted by different ethnic groups.” Here, we take the main verb of the event, [in- vaded]1 as the representation of this event and the antecedent for pronoun [It]2. According to (Asher, 1993), antecedents of event pronoun include both gerunds (e.g. de- struction) and inflectional verbs (e.g. destroying). In our study, we focus on the inflectional verb representation, as the gerund representation is studied in the conventional anaphora resolution. For the rest of this paper, “event pronouns” are pronouns whose antecedents are event verbs while “non-event anaphoric pronouns” are those with antecedents other than event verbs. Entity anaphora resolution provides critical links for cascaded event template extraction. It also provides useful information …

Content truncated to 500 words