A distributional account of the semantics of multiword expressions
Afsaneh Fazly, Suzanne Stevenson · 2008
The lexical status of multiword expressions (MWEs), such as make a decision and shoot the breeze, has long been a matter of debate. Although MWEs behave much like phrases on the surface, it has been argued that they should be treated like words because their components together form a single unit of meaning. However, MWEs are not a homogeneous lexical category, but rather can have distinct semantic and syntactic properties. For example, the overall meaning of an MWE may vary in how much it diverges from the combined contribution of its constituent parts, with make a decision, e.g., having a strong relation to decide, while shoot the breeze is entirely idiomatic. In order to understand whether and how MWEs should be represented in a (computational) lexicon, it is necessary to look into the relationship between the underlying semantic properties of these expressions and their surface behaviour. We examine several properties of MWEs pertaining to their semantic idiosyncrasy, and relate them to the distributional behaviour of MWEs in their actual usages. Accordingly, we propose statistical measures for quantifying each property, which we then use for separating different types of MWEs that require different treatment within a lexicon. 1