Event Extraction as Dependency Parsing

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 1626–1635, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Event Extraction as Dependency Parsing David McClosky, Mihai Surdeanu, and Christopher D. Manning Department of Computer Science Stanford University Stanford, CA 94305 {mcclosky,mihais,manning}@stanford.edu Abstract Nested event structures are a common occur- rence in both open domain and domain spe- cific extraction tasks, e.g., a “crime” event can cause a “investigation” event, which can lead to an “arrest” event. However, most cur- rent approaches address event extraction with highly local models that extract each event and argument independently. We propose a simple approach for the extraction of such structures by taking the tree of event-argument relations and using it directly as the representation in a reranking dependency parser. This provides a simple framework that captures global prop- erties of both nested and flat event structures. We explore a rich feature space that models both the events to be parsed and context from the original supporting text. Our approach ob- tains competitive results in the extraction of biomedical events from the BioNLP’09 shared task with a F1 score of 53.5% in development and 48.6% in testing. 1 Introduction Event structures in open domain texts are frequently highly complex and nested: a “crime” event can cause an “investigation” event, which can lead to an “arrest” event (Chambers and Jurafsky, 2009). The same observation holds in specific domains. For ex- ample, the BioNLP’09 shared task (Kim et al., 2009) focuses on the extraction of nested biomolecular events, where, e.g., a REGULATION event causes a TRANSCRIPTION event (see Figure 1a for a detailed example). Despite this observation, many state- of-the-art supervised event extraction models still extract events and event arguments independently, ignoring their underlying structure (Bj¨orne et al., 2009; Miwa et al., 2010b). In this paper, we propose a new approach for su- pervised event extraction where we take the tree of relations and their arguments and use it directly as the representation in a dependency parser (rather than conventional syntactic relations). Our approach is conceptually simple: we first convert the origi- nal representation of events and their arguments to dependency trees by creating dependency arcs be- tween event anchors (phrases that anchor events in the supporting text) and their corresponding argu- ments.1 Note that after conversion, only event an- chors and entities remain. Figure 1 shows a sentence and its converted form from the biomedical do- main with four events: two POSITIVE REGULATION events, anchored by the phrase “acts as a costim- ulatory signal,” and two TRANSCRIPTION events, both anchored on “gene transcription.” All events take either protein entity mentions (PROT) or other events as arguments. The latter is what allows for nested event structures. Existing dependency pars- ing models can be adapted to produce these seman- tic structures instead of syntactic dependencies. We built a global reranking parser model using multiple decoders from MSTParser (McDonald et al., 2005; McDonald et al., 2005b). The main contributions of this paper are …

Content truncated to 500 words