Multiword expressions and the lexicon

The term ‘multiword expression’ (MWE) refers to a diverse group of linguistic phenomena connected by the fact that they do not fit neatly into the word-phrase dichotomy: like phrases, they appear to be made up of multiple words; but, like words, they have idiosyncrasies (of meaning, form, or both) which must be learned. Examples include idioms, like spill the beans, ‘reveal the secrets’; prepositional verbs, like rely on; or light verb constructions, like have a break. Such expressions are of interest to linguistic theory because they straddle the boundary between the productive and idiosyncratic components of language, i.e. between what is (or can be) computed and what must be memorised. The thesis begins by investigating a number of properties of MWEs which any theory must give an account of: for example, they can be idiomatic in a variety of ways, such as by having unpredictable meanings, containing words which do not appear outside of the expression (run amok), or having a syntactic structure not attested elsewhere (by and large); they are also variably flexible, sometimes allowing modification (they spilled the financial beans), passivisation (the beans were spilled), or extraction (the beans that they spilled); and they have a specific psycholinguistic profile: idioms are processed faster than compositional phrases. A theory of MWEs has to give an account of how the more idiomatic properties square with the more productive ones. This thesis aims to do just that. It is argued that there are essentially four families of theory one might entertain: one which treats MWEs as big words, one which treats them as phrases which are composed of special versions of the words they contain, one which obtains a literal parse of the phrase first and then translates the literal meaning into the idiomatic one, and one which treats them as stored phrases which possess internal structure. The first approach treats MWEs as wholly word-like (i.e. as atomic units), and the next two treat them as wholly phrase-like (i.e. as composed of smaller independent units). It is only the fourth which takes the tension inherent in MWEs seriously, recognising that they are units in their own right, but also have internal structure. I argue, therefore, that this is the kind of theory we should prefer. The proposal I develop hinges formally on the use of a Tree-Adjoining Grammar (TAG), a tree-rewriting formalism which allows us to store (internally structured) phrases as elementary objects. This is integrated into the architecture of Lexical Functional Grammar (LFG), replacing the usual context-free grammar used there. By treating TAG elementary trees as descriptions, we are able to reduce LFG lexical entries to nothing but a set of constraints (on all levels of linguistic structure). The resulting framework has a number of attractive properties: it inherits from LFG the ability to separate configurational from functional information in the syntax, and gains from TAG a natural capacity to represent MWEs, and the ability to ‘lexicalise’ the grammar (associate every elementary structure with a lexical item), which I argue is a desirable property for a lexicalist theory like LFG. Because of this last result, I call the theory Lexicalised LFG.