A Probabilistic Model for Associative Anaphora Resolution

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1455–1464, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP A Probabilistic Model for Associative Anaphora Resolution Ryohei Sasano and Sadao Kurohashi Graduate School of Informatics, Kyoto University, Yoshida-honmachi, Sakyo-ku, Kyoto {sasano,kuro}@i.kyoto-u.ac.jp Abstract This paper proposes a probabilistic model for associative anaphora resolution in Japanese. Associative anaphora is a type of bridging anaphora, in which the anaphor and its antecedent are not coref- erent. Our model regards associative anaphora as a kind of zero anaphora and resolves it in the same manner as zero anaphora resolution using automatically acquired lexical knowledge. Experimen- tal results show that our model resolves associative anaphora with good perfor- mance and the performance is improved by resolving it simultaneously with zero anaphora. 1 Introduction The correct interpretation of anaphora is vital for natural language understanding. Bridging anaphora (Clark, 1975) represents a special part of the general problem of anaphora resolution, which has been studied and discussed for various lan- guages and domains (Hahn et al., 1996; Murata et al., 1999; Poesio et al., 2004; Gasperin and Vieira, 2004; Gasperin and Briscoe, 2008). Usually bridging anaphora considers two types:1 associative anaphors are noun phrases (NPs) that have an antecedent that is necessary to their interpretation but the relation between the anaphor and its antecedent is different from iden- tity; and indirect anaphors are those that have an identity relation with their antecedents but the anaphor and its antecedent have different head 1The terminology that we use here is introduced by Hawkins (1978), which is also used in (Vieira et al., 2006). nouns. In this paper, we focus on associative anaphora in Japanese. Associative anaphora resolution is decomposed into two steps: acquiring lexical knowledge for as- sociative anaphora resolution, and resolving asso- ciative anaphora using the acquired knowledge. Grammatical salience plays a lesser role for resolving anaphors with full lexical heads, than for pronominal anaphora (Strube and Hahn, 1999; Modjeska, 2002). Furthermore, since associative anaphors and their antecedents usually have differ- ent head nouns, string matching technique cannot be applied. Therefore, a large and diverse amount of lexical knowledge is essential to understand as- sociative anaphora. For example, to recognize the meronymic relation between “a house” and “the roof” in (1), such knowledge as “a roof” is a part of a building or vehicle is required. To recognize the attributive relation between “Prius” and “the price” in (2), such knowledge as “price” is a price of some goods or service is required. (1) There was a house. The roof was white. (2) Toyota launched the hybrid car Prius in 1997. The price was 21.5 million yen. To acquire such lexical knowledge, various studies have been carried out. Early studies used hand-crafted lexical knowledge such as Word- Net (Strube and Hahn, 1999; Vieira and Poe- sio, 2000; Meyer and Dale, 2002), but obtained poor or mediocre results. Hence, Poesio et al. (2002) proposed to exploit “Nh of Nm” phrases in large corpora to resolve associative anaphora …

Content truncated to 500 words