The VerbCorner Project: Toward an Empirically-Based Semantic Decomposition of Verbs

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1438–1442, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics The VerbCorner Project: Toward an Empirically-Based Semantic Decomposition of Verbs Joshua K. Hartshorne Department of Brain and Cognitive Sciences Massachusetts Institute of Technology 77 Massachusetts Avenue Cambridge, MA 02139, USA jkhartshorne@gmail.com Claire Bonial, Martha Palmer Department of Linguistics University of Colorado at Boulder Hellems 290, 295 UCB Boulder, CO 80309, USA {CBonial, MPalmer}@colorado.edu Abstract This research describes efforts to use crowd- sourcing to improve the validity of the seman- tic predicates in VerbNet, a lexicon of about 6300 English verbs. The current semantic predicates can be thought of semantic prim- itives, into which the concepts denoted by a verb can be decomposed. For example, the verb spray (of the Spray class), involves the predicates MOTION, NOT, and LOCATION, where the event can be decomposed into an AGENT causing a THEME that was originally not in a particular location to now be in that location. Although VerbNet’s predicates are theoretically well-motivated, systematic em- pirical data is scarce. This paper describes a recently-launched attempt to address this issue with a series of human judgment tasks, posed to subjects in the form of games. 1 Introduction One key application of Natural Language Processing (NLP) is meaning extraction. Of particular impor- tance is propositional meaning: To understand “Jes- sica sprayed paint on the wall,” it is not enough to know who Jessica is, what paint is, and where the wall is, but that, by the end of the event, some quan- tity of paint that was not previously on the wall now is. One must extract not only meanings for individ- ual words but also the relations between them. One option is to learn these relations in a largely bottom-up, data-driven fashion (Chklovski and Pan- tel, 2004; Poon and Domingos, 2009). For instance, Poon and Domingos (2009) first extracts depen- dency trees, converts those into quasi-logical form, recursively induces lambda expressions from them, and uses clustering to derive progressively abstract knowledge. An alternative is to take a human-inspired ap- proach, mapping the linguistic input onto the kinds of representations that linguistic and psychologi- cal research suggests are the representations em- ployed by humans. While the exact characteriza- tion of meaning (and by extension, thought) remains an area of active research in the cognitive sciences (Margolis and Laurence, 1999), decades of research in linguistics and psychology suggests that much of the meaning of a sentence – as well as its syntactic structure – can be accounted for by invoking a small number of highly abstract semantic features (usu- ally represented as predicates), such as causation, agency, basic topological relations, and directed mo- tion (Ambridge et al., 2013; Croft, 2012; Jackend- off, 1990; Levin and Rappaport Hovav, 2005; Peset- sky, 1995; Pinker, 1989). For instance, a given verb can appear in some syntactic frames (Sally broke the vase. Sally broke the vase with the hammer. The vase broke.) and not others …

Content truncated to 500 words