The VNC-Tokens Dataset

Idiomatic expressions formed from a verb and a noun in its direct object position are a productive cross-lingual class of multiword expressions, which can be used both idiomatically and as a literal combination. This paper presents the VNC-Tokens dataset, a resource of almost 3000 English verb–noun combination usages annotated as to whether they are literal or idiomatic. Previous research using this dataset is described, and other studies which could be evaluated more extensively using this resource are identified. 1. Verb–Noun Combinations Identifying multiword expressions (MWEs) in text is essential for accurately performing natural language processing tasks (Sag et al., 2002). A broad class of MWEs with distinct semantic and syntactic properties is that of idiomatic expressions. A productive process of idiom creation across languages is to combine a high frequency verb and one or more of its arguments. In particular, many such idioms are formed from the combination of a verb and a noun in the direct object position (Cowie et al., 1983; Nunberg et al., 1994; Fellbaum, 2002), e.g., give the sack, make a face, and see stars. Given the richness and productivity of the class of idiomatic verb–noun combinations (VNCs), we choose to focus on these expressions. It is a commonly held belief that expressions with an idiomatic interpretation are primarily used idiomatically, and that they lose their literal meanings over time. Nonetheless, it is still possible for a potentially-idiomatic combination to be used in a literal sense, as in: She made a face on the snowman using a carrot and two buttons. Contrast the above literal usage with the idiomatic use in: The little girl made a funny face at her mother. Interestingly, in our analysis of 60 VNCs, we found that approximately half of these expressions are attested fairly frequently in their literal sense in the British National Corpus (BNC).1 Clearly, automatic methods are required for distinguishing between idiomatic and literal usages of such expressions, and indeed there have recently been several studies addressing this issue (Birke and Sarkar, 2006; Katz and Giesbrecht, 2006; Cook et al., 2007). In order to conduct further research on VNCs at the token level, and to compare the effectiveness of the varying proposed methods for their treatment, an annotated corpus of VNC usages is required. Section 2 describes our dataset, VNC-Tokens, which consists of almost 3000 English sentences, each containing a VNC usage (token) annotated as to whether it is literal or idiomatic. Sections 3, 4, and 5 respectively describe previous research conducted using VNC-Tokens, other work on idioms which could make use of this dataset, and possible ways in which VNC-Tokens could be extended. We summarize the contributions of the VNC-Tokens resource in Section 6. http://www.natcorp.ox.ac.uk 2. The VNC-Tokens Dataset The following subsections describe the selection of the expressions in VNC-Tokens, how usages of these expressions were found, and the annotation of the tokens.