Learning to Model Multilingual Unrestricted Coreference in OntoNotes

Proceedings of the Joint Conference on EMNLP and CoNLL: Shared Task, pages 129–135, Jeju Island, Korea, July 13, 2012. c⃝2012 Association for Computational Linguistics Learning to Model Multilingual Unrestricted Coreference in OntoNotes Baoli LI Department of Computer Science Henan University of Technology 1 Lotus Street, High&New Technology Industrial Development Zone, Zhengzhou, Henan, China, 450001 csblli@gmail.com Abstract Coreference resolution, which aims at correctly linking meaningful expressions in text, is a much challenging problem in Natural Language Processing community. This paper describes the multilingual coreference modeling system of Web Information Processing Group, Henan University of Technology, China, for the CoNLL-2012 shared task (closed track). The system takes a supervised learning strategy, and consists of two cascaded components: one for detecting mentions, and the other for clustering mentions. To make the system applicable for multiple languages, generic syntactic and semantic features are used to model coreference in text. The system obtained combined official score 41.88 over three languages (Arabic, Chinese, and English) and ranked 7th among the 15 systems in the closed track. 1 Introduction Coreference resolution, which aims at correctly linking meaningful expressions in text, has become a central research problem in natural language processing community with the advent of various supporting resources (e.g. corpora and different kinds of knowledge bases). OntoNotes (Pradhan et al. 2007), compared to MUC (Chinchor, 2001; Chinchor and Sundheim, 2003) and ACE (Doddington et al. 2000) corpora, is a large-scale, multilingual corpus for general anaphoric coreference that covers entities and events not limited to noun phrases or a limited set of entity types. It greatly stimulates the research on this challenging problem – Coreference Resolution. Moreover, resources like WordNet (Miller, 1995) and the advancement of different kinds of syntactic and semantic analysis technologies, make it possible to do in-depth research on this topic, which is demanded in most of natural language processing applications, such as information extraction, machine translation, question answering, summarization, and so on. Our group is exploring how to extract information from grain/cereal related Chinese text for business intelligence. This shared task provides a good platform for advancing our research on IE related topics. We experiment with a machine learning strategy to model multilingual coreference for the CoNLL-2012 shared task (Pradhan et al. 2012). Two steps are taken to detect coreference in text: mention detection and mention clustering. We consider mentions that correspond to a word or an internal node in a syntactic tree and ignore the rest mentions, as we think a mention should be a valid meaningful unit of a sentence. Maximal entropy algorithm is used to model what a mention is and how two mentions link to each other. Generic features are designed to facilitate these modeling. 129 Our official submission obtained combined official score 41.88 over three languages (Arabic, Chinese, and English), which ranked the system 7th among 15 systems participating the closed track. Our system performs poor on the Arabic data, and has relatively high precision but low recall. The rest of this paper is organized as follows. Section 2 …

Content truncated to 500 words