Labeling Gaps Between Words: Recognizing Overlapping Mentions with Mention Separators

Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2608–2618 Copenhagen, Denmark, September 7–11, 2017. c⃝2017 Association for Computational Linguistics Labeling Gaps Between Words: Recognizing Overlapping Mentions with Mention Separators Aldrian Obaja Muis and Wei Lu Singapore University of Technology and Design {aldrian_muis,luwei}@sutd.edu.sg Abstract In this paper, we propose a new model that is capable of recognizing overlapping mentions. We introduce a novel notion of mention separators that can be effectively used to capture how mentions overlap with one another. On top of a novel multigraph representation that we introduce, we show that efficient and exact inference can still be performed. We present some theoret- ical analysis on the differences between our model and a recently proposed model for recognizing overlapping mentions, and discuss the possible implications of the differences. Through extensive empirical analysis on standard datasets, we demon- strate the effectiveness of our approach. 1 Introduction Named entity recognition (NER), or in general the task of recognizing entity mentions1 in a text, has been a research topic for many years (Mc- Callum and Li, 2003; Nadeau and Sekine, 2007; Ratinov and Roth, 2009; Ling and Weld, 2012). However, as noted by Finkel and Manning (2009), many previous works ignored overlapping men- tions, although they are quite common. Figure 1 illustrates some examples of overlapping men- tions adapted from existing datasets. For example, the location mention Pennsylvania appears within the mention of type organization a Pennsylvania radio station. In practice, overlapping mentions have been found in many existing datasets across different domains (Doddington et al., 2004; Kim et al., 2003; Suominen et al., 2013). Developing algorithms that can effectively and efficiently ex- tract overlapping mentions can be crucial for the 1As noted in (Florian et al., 2004), mention recognition is more general than NER, where a mention can be either named, nominal, or pronominal. At issue is the liability of a Pennsylvania ::::::::: GPE radio station :::::::::::::::::::: ORG under the federal wiretap statute. CAT expression directed by the IL2 :: DNA regulatory region ::::::::::::::: DNA or by a multimer of the NF-AT :::: PROT -binding site ::::::::::::: DNA was lower. Figure 1: Examples of overlapping mentions. performance of many downstream tasks such as relation extraction (Mintz et al., 2009; Gupta and Andrassy, 2016), event extraction (Lu and Roth, 2012; Li et al., 2013; Nguyen et al., 2016), coref- erence resolution (Chang et al., 2013; Lu et al., 2016), question answering (Mollá et al., 2007), and equation parsing (Roy et al., 2016). Overlapping mention recognition is non-trivial, as existing methods that model mention recogni- tion as a sequence prediction problem – e.g., us- ing linear-chain conditional random fields (CRF) (Lafferty et al., 2001) – have difficulties in han- dling overlapping mentions (Alex et al., 2007). Finkel and Manning (2009) proposed to use a tree- based constituency parsing model to handle nested entities.2 Due to the tree structured representa- tion used, the resulting algorithm has a time com- plexity that is cubic in n for its inference proce- …

Content truncated to 500 words