Neural Modeling of Multi-Predicate Interactions for Japanese Predicate Argument Structure Analysis

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1591–1600 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1146 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1591–1600 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1146 Neural Modeling of Multi-Predicate Interactions for Japanese Predicate Argument Structure Analysis Hiroki Ouchi1,2 Hiroyuki Shindo1,2 Yuji Matsumoto1,2 1 Nara Institute of Science and Technology 2 RIKEN Center for Advanced Intelligence Project (AIP) { ouchi.hiroki.nt6, shindo, matsu }@is.naist.jp Abstract The performance of Japanese predicate ar- gument structure (PAS) analysis has im- proved in recent years thanks to the joint modeling of interactions between multi- ple predicates. However, this approach re- lies heavily on syntactic information pre- dicted by parsers, and suffers from error propagation. To remedy this problem, we introduce a model that uses grid-type re- current neural networks. The proposed model automatically induces features sen- sitive to multi-predicate interactions from the word sequence information of a sen- tence. Experiments on the NAIST Text Corpus demonstrate that without syntactic information, our model outperforms previ- ous syntax-dependent models. 1 Introduction Predicate argument structure (PAS) analysis is a basic semantic analysis task, in which systems are required to identify the semantic units of a sen- tence, such as who did what to whom. In pro- drop languages such as Japanese, Chinese and Italian, arguments are often omitted in text, and such argument omission is regarded as one of the most problematic issues facing PAS analy- sis (Iida and Poesio, 2011; Sasano and Kurohashi, 2011; Hangyo et al., 2013). In response to the argument omission prob- lem, in Japanese PAS analysis, a joint model of the interactions between multiple predicates has been gaining popularity and achieved the state-of- the-art results (Ouchi et al., 2015; Shibata et al., 2016). This approach is based on the linguistic in- tuition that the predicates in a sentence are seman- tically related to each other, and capturing this re- lation can be useful for PAS analysis. In the exam- Figure 1: Example of Japanese PAS. The upper edges denote dependency relations, and the lower edges denote case arguments. “NOM” and “ACC” denote the nominative and accusative arguments, respectively. “ϕi” is a zero pronoun, referring to the antecedent “男 i (mani)”. ple sentence in Figure 1, the word “男 i (mani)” is the accusative argument of the predicate “逮捕し た (arrested)” and is shared by the other predicate “逃走した (escaped)” as its nominative argument. Considering the semantic relation between “逮捕 した (arrested)” and “逃走した (escaped)”, we in- tuitively know that the person arrested by someone is likely to be the escaper. That is, information about one predicate-argument relation could help to identify another predicate-argument relation. However, to model such multi-predicate inter- actions, the joint approach in the previous stud- ies relies heavily on syntactic information, such as part-of-speech (POS) tags and dependency re- lations predicted by POS taggers and syntactic parsers. Consequently, it suffers …

Content truncated to 500 words