Idiomatic Constructions in HPSG

In this paper I present an approach to idioms in the HPSG framework. Building on earlier work by Copestake 1994, it employs phrasal types that specify the semantic relationship between the idiomatic words involved. Underspecified Phrasal Semantics (UPS) does not require separate lexical entries for the words occurring in idioms, and treats a wide range of data more successfully than previous alternatives. The approach allows for the variability that some idioms exhibit, while being able to express what is fixed. It provides a solution to a problem discussed in McCawley 1981 which involves idioms occurring distributed over a main clause and a subordinate clause. The available psycholinguistic evidence seems to be consistent with the approach, and it is intuitive to view an idiom like spill the beans as a whole, comprising the three words spill , the, and beans , without giving these words an independent existence outside the idiom. The alternative view of the idiom as a special form of spill that happens to occur only together with the words the and beans does not provide a representation for the idiom as a whole, and requires a mechanism to ensure that the parts of the idiom that are being subcategorized for cannot occur by themselves. The reason that this has been the predominant view is that it was thought to be impossible to deal with the variation data in a phrasal approach. The UPS approach can also deal with semantically decomposable and non-decomposable idioms (Nunberg et al. 1994). Examples of decomposable idioms are pull strings and spill the beans , where spill means