Exploiting Syntactico-Semantic Structures for Relation Extraction

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 551–560, Portland, Oregon, June 19-24, 2011. c⃝2011 Association for Computational Linguistics Exploiting Syntactico-Semantic Structures for Relation Extraction Yee Seng Chan and Dan Roth University of Illinois at Urbana-Champaign {chanys,danr}@illinois.edu Abstract In this paper, we observe that there exists a second dimension to the relation extraction (RE) problem that is orthogonal to the relation type dimension. We show that most of these second dimensional structures are relatively constrained and not difficult to identify. We propose a novel algorithmic approach to RE that starts by first identifying these structures and then, within these, identifying the seman- tic type of the relation. In the real RE problem where relation arguments need to be identi- fied, exploiting these structures also allows re- ducing pipelined propagated errors. We show that this RE framework provides significant improvement in RE performance. 1 Introduction Relation extraction (RE) has been defined as the task of identifying a given set of semantic binary rela- tions in text. For instance, given the span of text “... the Seattle zoo . . . ”, one would like to extract the relation that “the Seattle zoo” is located-at “Seattle”. RE has been frequently studied over the last few years as a supervised learning task, learning from spans of text that are annotated with a set of seman- tic relations of interest. However, most approaches to RE have assumed that the relations’ arguments are given as input (Chan and Roth, 2010; Jiang and Zhai, 2007; Jiang, 2009; Zhou et al., 2005), and therefore offer only a partial solution to the problem. Conceptually, this is a rather simple approach as all spans of texts are treated uniformly and are be- ing mapped to one of several relation types of in- terest. However, these approaches to RE require a large amount of manually annotated training data to achieve good performance, making it difficult to ex- pand the set of target relations. Moreover, as we show, these approaches become brittle when the re- lations’ arguments are not given but rather need to be identified in the data too. In this paper we build on the observation that there exists a second dimension to the relation extraction problem that is orthogonal to the relation type di- mension: all relation types are expressed in one of several constrained syntactico-semantic structures. As we show, identifying where the text span is on the syntactico-semantic structure dimension first, can be leveraged in the RE process to yield improved per- formance. Moreover, working in the second dimen- sion provides robustness to the real RE problem, that of identifying arguments along with the relations be- tween them. For example, in “the Seattle zoo”, the entity men- tion “Seattle” modifies the noun “zoo”. Thus, the two mentions “Seattle” and “the Seattle zoo”, are involved in what we later call a premodifier rela- tion, one of several syntactico-semantic structures we identify in Section 3. We highlight that all relation types can be ex- …

Content truncated to 500 words