Using Semantic Roles to Improve Question Answering

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 12–21, Prague, June 2007. c⃝2007 Association for Computational Linguistics Using Semantic Roles to Improve Question Answering Dan Shen Spoken Language Systems Saarland University Saarbruecken, Germany dan@lsv.uni-saarland.de Mirella Lapata School of Informatics University of Edinburgh Edinburgh, UK mlap@inf.ed.ac.uk Abstract Shallow semantic parsing, the automatic identification and labeling of sentential con- stituents, has recently received much atten- tion. Our work examines whether seman- tic role information is beneficial to question answering. We introduce a general frame- work for answer extraction which exploits semantic role annotations in the FrameNet paradigm. We view semantic role assign- ment as an optimization problem in a bipar- tite graph and answer extraction as an in- stance of graph matching. Experimental re- sults on the TREC datasets demonstrate im- provements over state-of-the-art models. 1 Introduction Recent years have witnessed significant progress in developing methods for the automatic identification and labeling of semantic roles conveyed by senten- tial constituents.1 The success of these methods, of- ten referred to collectively as shallow semantic pars- ing (Gildea and Jurafsky, 2002), is largely due to the availability of resources like FrameNet (Fillmore et al., 2003) and PropBank (Palmer et al., 2005), which document the surface realization of semantic roles in real world corpora. More concretely, in the FrameNet paradigm, the meaning of predicates (usually verbs, nouns, or ad- jectives) is conveyed by frames, schematic repre- sentations of situations. Semantic roles (or frame 1The approaches are too numerous to list; we refer the inter- ested reader to Carreras and M`arquez (2005) for an overview. elements) are defined for each frame and corre- spond to salient entities present in the evoked situ- ation. Predicates with similar semantics instantiate the same frame and are attested with the same roles. The FrameNet database lists the surface syntactic realizations of semantic roles, and provides anno- tated example sentences from the British National Corpus. For example, the frame Commerce Sell has three core semantic roles, namely Buyer, Goods, and Seller — each expressed by an indirect object, a di- rect object, and a subject (see sentences (1a)–(1c)). It can also be attested with non-core (peripheral) roles (e.g., Means, Manner, see (1d) and (1e)) that are more generic and can be instantiated in sev- eral frames, besides Commerce Sell. The verbs sell, vend, and retail can evoke this frame, but also the nouns sale and vendor. (1) a. [Lee]Seller sold a textbook [to Abby]Buyer. b. [Kim]Seller sold [the sweater]Goods. c. [My company]Seller has sold [more than three million copies]Goods. d. [Abby]Seller sold [the car]Goods [for cash]Means. e. [He]Seller [reluctanctly]Manner sold [his rock]Goods. By abstracting over surface syntactic configura- tions, semantic roles offer an important first step to- wards deeper text understanding and hold promise for a range of applications requiring broad cover- age semantic processing. Question answering (QA) is often cited as an obvious beneficiary of semantic 12 role labeling (Gildea and Jurafsky, 2002; Palmer et al., 2005; Narayanan and …

Content truncated to 500 words