A Probabilistic Method for Analyzing Japanese Anaphora Integrating Zero Pronoun Detection and Resolution

A Probabilistic Method for Analyzing Japanese Anaphora Integrating Zero Pronoun Detection and Resolution Kazuhiro Seki †, Atsushi Fujii ††, ††† and Tetsuya Ishikawa †† †National Institute of Advanced Industrial Science and Technology 1-1-1, Chuuou Daini Umezono, Tsukuba 305-8568, Japan ††University of Library and Information Science 1-2, Kasuga, Tsukuba, 305-8550, Japan †††CREST, Japan Science & Technology Corporation k.seki@aist.go.jp fujii@ulis.ac.jp ishikawa@ulis.ac.jp Abstract This paper proposes a method to analyze Japanese anaphora, in which zero pronouns (omitted obligatory cases) are used to refer to preceding entities (antecedents). Unlike the case of general coreference resolution, zero pro- nouns have to be detected prior to resolution because they are not expressed in discourse. Our method integrates two probability param- eters to perform zero pronoun detection and resolution in a single framework. The first pa- rameter quantifies the degree to which a given case is a zero pronoun. The second parame- ter quantifies the degree to which a given entity is the antecedent for a detected zero pronoun. To compute these parameters efficiently, we use corpora with/without annotations of anaphoric relations. We show the effectiveness of our method by way of experiments. 1 Introduction Anaphora resolution is crucial in natural lan- guage processing (NLP), specifically, discourse analysis. In the case of English, partially mo- tivated by Message Understanding Conferences (MUCs) (Grishman and Sundheim, 1996), a number of coreference resolution methods have been proposed. In other languages such as Japanese and Spanish, anaphoric expressions are often omit- ted. Ellipses related to obligatory cases are usu- ally termed zero pronouns. Since zero pronouns are not expressed in discourse, they have to be detected prior to identifying their antecedents. Thus, although in English pleonastic pronouns have to be determined whether or not they are anaphoric expressions prior to resolution, the process of analyzing Japanese zero pronouns is different from general coreference resolution in English. For identifying anaphoric relations, existing methods are classified into two fundamental ap- proaches: rule-based and statistical approaches. In rule-based approaches (Grosz et al., 1995; Hobbs, 1978; Mitkov et al., 1998; Nakaiwa and Shirai, 1996; Okumura and Tamura, 1996; Palomar et al., 2001; Walker et al., 1994), anaphoric relations between anaphors and their antecedents are identified by way of hand- crafted rules, which typically rely on syntactic structures, gender/number agreement, and se- lectional restrictions. However, it is difficult to produce rules exhaustively, and rules that are developed for a specific language are not neces- sarily effective for other languages. For exam- ple, gender/number agreement in English can- not be applied to Japanese. Statistical approaches (Aone and Bennett, 1995; Ge et al., 1998; Kim and Ehara, 1995; Soon et al., 2001) use statistical mod- els produced based on corpora annotated with anaphoric relations. However, only a few attempts have been made in corpus-based anaphora resolution for Japanese zero pro- nouns. One of the reasons is that it is costly to produce a sufficient volume of training cor- pora annotated with anaphoric relations. In addition, those above methods focused mainly on identifying antecedents, and few at- tempts have …

Content truncated to 500 words