A Concept-Centered Approach to Noun-Compound Interpretation

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 81–88 Manchester, August 2008 A Concept-Centered Approach to Noun-Compound Interpretation Cristina Butnariu School of Computer Science and Informatics University College Dublin Belfield, Dublin 4 Ioana.Butnariu@ucd.ie Tony Veale School of Computer Science and Informatics University College Dublin Belfield, Dublin 4 Tony.Veale@ucd.ie Abstract A noun-compound is a compressed proposition that requires an audience to recover the implicit relationship between two concepts that are expressed as nouns. Listeners recover this relationship by considering the most typical relations afforded by each concept. These relational possibilities are evident at a linguistic level in the syntagmatic patterns that connect nouns to the verbal actions that act upon, or are facilitated by, these nouns. We present a model of noun-compound interpretation that first learns the relational possibilities for individual nouns from corpora, and which then uses these to hypothesize about the most likely relationship that underpins a noun compound. 1 Introduction Noun compounds hide a remarkable depth of conceptual machinery behind a simple syntactic form, Noun-Noun, and thus pose a considerable problem for the computational processing of language (Johnston and Busa, 1996). It is not just that compounds are commonplace in language, or that their interpretation requires a synthesis of lexical, semantic, and pragmatic information sources (Finin, 1980); compounds provide a highly compressed picture of the workings of concept combination, so there are as many ways of interpreting a noun compound as there are ways of combining the underlying concepts (Gagné, 2002). Linguists have thus attempted to © 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported license (http://creativecommons.org/licenses/by-nc- sa/3.0/). Some rights reserved. understand noun-compounds as full propositions in which a phrase with two nouns connected by an explicit relation – usually expressed as a verb and a preposition – is compressed into a pair of nouns (Levi, 1978). Since these noun-pairs must allow an audience to reconstruct the decompressed proposition, there must be some systematic means by which the missing relation can easily be inferred. This framing of the problem as a search for a missing relation suggests two broad strategies for the interpretation of compounds. In the first, the top-down strategy, we assume that there are only so many ways of combining two concepts; by enumerating these ways, we can view the problem of interpretation as a problem of classification, in which compounds are placed into separate classes that each correspond to a single manner of concept connection (Kim and Baldwin, 2006), (Nastase and Szpakowicz, 2003). This strategy explicitly shaped the SemEval task on classifying semantic relations between nominals (Girju et al., 2007) and so is employed by all of the systems that participated in that task. In the second, the bottom-up strategy, we assume that it is futile to try and enumerate the many ways in which concepts can relationally combine, but look instead to large corpora to discover the ways in which different word combinations are explicitly framed by language (Nakov, 2006), (Turney, 2006a). In this paper we describe an …

Content truncated to 500 words