Unsupervised Recognition of Literal and Non-Literal Use of Idiomatic Expressions

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 754–762, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics Unsupervised Recognition of Literal and Non-Literal Use of Idiomatic Expressions Caroline Sporleder and Linlin Li Saarland University Postfach 15 11 50 66041 Saarbr¨ucken, Germany {csporled,linlin}@coli.uni-saarland.de Abstract We propose an unsupervised method for distinguishing literal and non-literal us- ages of idiomatic expressions. Our method determines how well a literal inter- pretation is linked to the overall cohesive structure of the discourse. If strong links can be found, the expression is classified as literal, otherwise as idiomatic. We show that this method can help to tell apart lit- eral and non-literal usages, even for id- ioms which occur in canonical form. 1 Introduction Texts frequently contain expressions whose mean- ing is not strictly literal, such as metaphors or id- ioms. Non-literal expressions pose a major chal- lenge to natural language processing as they often exhibit lexical and syntactic idiosyncrasies. For example, idioms can violate selectional restric- tions (as in push one’s luck under the assumption that only concrete things can normally be pushed), disobey typical subcategorisation constraints (e.g., in line without a determiner before line), or change the default assignments of semantic roles to syn- tactic categories (e.g., in break sth with X the ar- gument X would typically be an instrument but for the idiom break the ice it is more likely to fill a patient role, as in break the ice with Russia). To avoid erroneous analyses, a natural language processing system should recognise if an expres- sion is used non-literally. While there has been a lot of work on recognising idioms (see Section 2), most previous approaches have focused on a type- based classification, dividing expressions into “id- iom” or “not an idiom” irrespective of their actual use in a discourse context. However, while some expressions, such as by and large, always have a non-compositional, idiomatic meaning, many id- ioms, such as break the ice or spill the beans, share their linguistic form with perfectly literal expres- sions (see examples (1) and (2), respectively). For some expressions, such as drop the ball, the lit- eral usage can even dominate in some domains. Hence, whether a potentially ambiguous expres- sion has literal or non-literal meaning has to be inferred from the discourse context. (1) Dad had to break the ice on the chicken troughs so that they could get water. (2) Somehow I always end up spilling the beans all over the floor and looking foolish when the clerk comes to sweep them up. Type-based idiom classification thus only ad- dresses part of the problem. While it can au- tomatically compile lists of potentially idiomatic expressions, it does not say anything about the idiomaticity of an expression in a particular context. In this paper, we propose a novel, cohesion-based approach for detecting non-literal usages (token-based idiom classification). Our approach is unsupervised and similar in spirit to Hirst and St-Onge’s (1998) method for detecting …

Content truncated to 500 words