Chinese Zero Pronoun Resolution with Deep Memory Network

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 1309–1318 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Chinese Zero Pronoun Resolution with Deep Memory Network Qingyu Yin, Yu Zhang, Weinan Zhang, Ting Liu∗ Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China {qyyin, yzhang, wnzhang, tliu}@ir.hit.edu.cn Abstract Existing approaches for Chinese zero pro- noun resolution typically utilize only syn- tactical and lexical features while ignoring semantic information. The fundamental reason is that zero pronouns have no de- scriptive information, which brings diffi- culty in explicitly capturing their semantic similarities with antecedents. Meanwhile, representing zero pronouns is challenging since they are merely gaps that convey no actual content. In this paper, we address this issue by building a deep memory net- work that is capable of encoding zero pro- nouns into vector representations with in- formation obtained from their contexts and potential antecedents. Consequently, our resolver takes advantage of semantic in- formation by using these continuous dis- tributed representations. Experiments on the OntoNotes 5.0 dataset show that the proposed memory network could substan- tially outperform the state-of-the-art sys- tems in various experimental settings. 1 Introduction A zero pronoun (ZP) is a gap in a sentence, which refers to an entity that supplies the neces- sary information for interpreting the gap (Zhao and Ng, 2007). A ZP can be either anaphoric if it corefers to one or more preceding noun phrases (antecedents) in the associated text, or non-anaphoric if there are no such noun phrases. Below is an example of ZPs and their antecedents, where “φ” denotes the ZP. [警方] 表示 他们 自杀 的 可能性 很高， 不 过 φ1 也 不 排除 φ2 有 他杀 的 可能。 ∗Email corresponding. ([The police] said that they are more likely to commit suicide, but φ1 could not rule out φ2 the possibility of homicide.) In this example, the ZP “φ1” is an anaphoric ZP that refers to the antecedent “警方/The police” while the ZP “φ2” is non-anaphoric. Unlike overt pronouns, ZPs lack grammatical attributes such as gender and number that have been proven to be essential in pronoun resolution (Chen and Ng, 2014a), which makes ZP resolution a more chal- lenging task than overt pronoun resolution. Automatic Chinese ZP resolution is typically composed of two steps, i.e., anaphoric zero pro- noun (AZP) identification that identifies whether a ZP is anaphoric; and AZP resolution, which deter- mines antecedents for AZPs. For AZP identifica- tion, state-of-the-art resolvers use machine learn- ing algorithms to build AZP classifiers in a super- vised manner (Chen and Ng, 2013, 2016). For AZP resolution, literature approaches include un- supervised methods (Chen and Ng, 2014b, 2015), feature-based supervised models (Zhao and Ng, 2007; Kong and Zhou, 2010), and neural network models (Chen and Ng, 2016). Neural network models for AZP resolution are of growing interest for their capacity to learn task-specific represen- tations without extensive feature engineering and to effectively exploit lexical information for ZPs and their candidate antecedents in a more …

Content truncated to 500 words