A Local Detection Approach for Named Entity Recognition and Mention Detection

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1237–1247 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1114 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1237–1247 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1114 A Local Detection Approach for Named Entity Recognition and Mention Detection Mingbin Xu, Hui Jiang, Sedtawut Watcharawittayakul Department of Electrical Engineering and Computer Science Lassonde School of Engineering, York University 4700 Keele Street, Toronto, Ontario, Canada {xmb, hj, watchara}@eecs.yorku.ca Abstract In this paper, we study a novel approach for named entity recognition (NER) and mention detection (MD) in natural lan- guage processing. Instead of treating NER as a sequence labeling problem, we propose a new local detection approach, which relies on the recent fixed-size ordi- nally forgetting encoding (FOFE) method to fully encode each sentence fragment and its left/right contexts into a fixed- size representation. Subsequently, a sim- ple feedforward neural network (FFNN) is learned to either reject or predict en- tity label for each individual text frag- ment. The proposed method has been eval- uated in several popular NER and MD tasks, including CoNLL 2003 NER task and TAC-KBP2015 and TAC-KBP2016 Tri-lingual Entity Discovery and Linking (EDL) tasks. Our method has yielded pretty strong performance in all of these examined tasks. This local detection ap- proach has shown many advantages over the traditional sequence labeling methods. 1 Introduction Natural language processing (NLP) plays an im- portant role in artificial intelligence, which has been extensively studied for many decades. Con- ventional NLP techniques include the rule-based symbolic approaches widely used about two decades ago, and the more recent statistical ap- proaches relying on feature engineering and sta- tistical models. In the recent years, deep learning approach has achieved huge successes in many ap- plications, ranging from speech recognition to im- age classification. It is drawing increasing atten- tion in the NLP community. In this paper, we are interested in a fundamen- tal problem in NLP, namely named entity recogni- tion (NER) and mention detection (MD). NER and MD are very challenging tasks in NLP, laying the foundation of almost every NLP application. NER and MD are tasks of identifying entities (named and/or nominal) from raw text, and classifying the detected entities into one of the pre-defined cate- gories such as person (PER), organization (ORG), location (LOC), etc. Some tasks focus on named entities only, while the others also detect nominal mentions. Moreover, nested mentions may need to be extracted too. For example, [Sue]P ER and her [brother]P ER N studied in [University of [Toronto]LOC]ORG. where Toronto is a LOC entity, embedded in an- other longer ORG entity University of Toronto. Similar to many other NLP problems, NER and MD is formulated as a sequence labeling prob- lem, where a tag is sequentially assigned to each word in the input sentence. It has been extensively studied in the NLP community (Borthwick et al., 1998). The …

Content truncated to 500 words