Template-Based Information Extraction without the Templates

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 976–986, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Template-Based Information Extraction without the Templates Nathanael Chambers and Dan Jurafsky Department of Computer Science Stanford University {natec,jurafsky}@stanford.edu Abstract Standard algorithms for template-based in- formation extraction (IE) require predefined template schemas, and often labeled data, to learn to extract their slot fillers (e.g., an embassy is the Target of a Bombing tem- plate). This paper describes an approach to template-based IE that removes this require- ment and performs extraction without know- ing the template structure in advance. Our al- gorithm instead learns the template structure automatically from raw text, inducing tem- plate schemas as sets of linked events (e.g., bombings include detonate, set off, and de- stroy events) associated with semantic roles. We also solve the standard IE task, using the induced syntactic patterns to extract role fillers from specific documents. We evaluate on the MUC-4 terrorism dataset and show that we in- duce template structure very similar to hand- created gold structure, and we extract role fillers with an F1 score of .40, approaching the performance of algorithms that require full knowledge of the templates. 1 Introduction A template defines a specific type of event (e.g., a bombing) with a set of semantic roles (or slots) for the typical entities involved in such an event (e.g., perpetrator, target, instrument). In contrast to work in relation discovery that focuses on learning atomic facts (Banko et al., 2007a; Carlson et al., 2010), templates can extract a richer representation of a particular domain. However, unlike relation dis- covery, most template-based IE approaches assume foreknowledge of the domain’s templates. Very little work addresses how to learn the template structure itself. Our goal in this paper is to perform the stan- dard template filling task, but to first automatically induce the templates from an unlabeled corpus. There are many ways to represent events, rang- ing from role-based representations such as frames (Baker et al., 1998) to sequential events in scripts (Schank and Abelson, 1977) and narrative schemas (Chambers and Jurafsky, 2009; Kasch and Oates, 2010). Our approach learns narrative-like knowl- edge in the form of IE templates; we learn sets of related events and semantic roles, as shown in this sample output from our system: Bombing Template {detonate, blow up, plant, explode, defuse, destroy} Perpetrator: Person who detonates, plants, blows up Instrument: Object that is planted, detonated, defused Target: Object that is destroyed, is blown up A semantic role, such as target, is a cluster of syn- tactic functions of the template’s event words (e.g., the objects of detonate and explode). Our goal is to characterize a domain by learning this template structure completely automatically. We learn tem- plates by first clustering event words based on their proximity in a training corpus. We then use a novel approach to role induction that clusters the syntactic functions of these events based on selectional prefer- ences and coreferring arguments. The induced roles …

Content truncated to 500 words