Constraints Based Taxonomic Relation Classification

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 1099–1109, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Constraints based Taxonomic Relation Classification Quang Xuan Do Dan Roth Department of Computer Science University of Illinois at Urbana-Champaign Urbana, IL 61801, USA {quangdo2,danr}@illinois.edu Abstract Determining whether two terms in text have an ancestor relation (e.g. Toyota and car) or a sibling relation (e.g. Toyota and Honda) is an essential component of textual inference in NLP applications such as Question Answer- ing, Summarization, and Recognizing Textual Entailment. Significant work has been done on developing stationary knowledge sources that could potentially support these tasks, but these resources often suffer from low cover- age, noise, and are inflexible when needed to support terms that are not identical to those placed in them, making their use as general purpose background knowledge resources dif- ficult. In this paper, rather than building a sta- tionary hierarchical structure of terms and re- lations, we describe a system that, given two terms, determines the taxonomic relation be- tween them using a machine learning-based approach that makes use of existing resources. Moreover, we develop a global constraint opti- mization inference process and use it to lever- age an existing knowledge base also to enforce relational constraints among terms and thus improve the classifier predictions. Our exper- imental evaluation shows that our approach significantly outperforms other systems built upon existing well-known knowledge sources. 1 Introduction Taxonomic relations that are read off of structured ontological knowledge bases have been shown to play important roles in many computational linguis- tics tasks, such as document clustering (Hotho et al., 2003), navigating text databases (Chakrabarti et al., 1997), Question Answering (QA) (Saxena et al., 2007) and summarization (Vikas et al., 2008). is clear that the recognition of taxonomic relationIt between terms in sentences is essential to support textual inference tasks such as Recognizing Textual Entailment (RTE) (Dagan et al., 2006). For exam- ple, it may be important to know that a blue Toy- ota is neither a red Toyota nor a blue Honda, but that all are cars, and even Japanese cars. Work in Textual Entailment has argued quite convincingly (MacCartney and Manning, 2008; MacCartney and Manning, 2009) that many such textual inferences are largely compositional and depend on the ability to recognize some basic taxonomic relations such as the ancestor or sibling relations between terms. To date, these taxonomic relations can be read off manually generated ontologies such as Wordnet that explicitly represent these, and there has also been some work trying to extend the manually built re- sources using automatic acquisition methods result- ing in structured knowledge bases such as the Ex- tended WordNet (Snow et al., 2006) and the YAGO ontology (Suchanek et al., 2007). However, identifying when these relations hold using fixed stationary hierarchical structures may be impaired by noise in the resource and by uncer- tainty in mapping targeted terms to concepts in the structures. In addition, for knowledge sources de- rived using …

Content truncated to 500 words