It is widely accepted at this stage of investigation into lexical semantics that some form of lexical decomposition is necessary to capture generalisations about the relationship between syntactic form and intended meaning (Jackendoff 1983, 1990; Dowty 1979; Pinker 1989; Pustejovsky 1991, pustejovsky:95a; Davis 1995; inter alia). It is equally well accepted that it is impossible to decompose meaning into necessary and sufficient conditions for identification of the entities words correspond to, for it will always be possible to discover some additional element of meaning which is needed to distinguish between two arbitrarily closely related words (Pulman 1983; Medin 1989; Keil 1989).
The position which Jackendoff advocates with respect to semantic decomposition is that word meanings must have internal structure, due to the creativity of language use (i.e. the ability of speakers of a language to understand and create an indefinitely large number of sentences with which they have no prior experience) and the regularities which accompany that use, and that part of this structure is the specification of necessary conditions for the application of a word. He then goes on to attempt to identify these necessary conditions. The decomposition of word meaning into smaller semantic elements allows specification of a generative, compositional system which constrains the way such elements can be related and thereby constrains the ways in which sentences can be constructed (to prevent semantically anomalous sentences), while not attempting to predict a priori what structures will actually be created and used. It is clearly desirable from a computational perspective to develop a system with such generative properties (this point is strongly argued by Pustejovsky 1991, pustejovsky:95a), and thus an analysis of the framework which Jackendoff proposes is warranted. In particular, since his framework largely derives from the analysis of linguistic alternations, it is likely to provide precisely the grain of decomposition necessary for developing the linguistic representation Copestake so persuasively advocates in the quote at the start of this chapter. This follows from the fact that linguistic alternations, such as the locative alternation exemplified by ji28, have subtle semantic distinctions (Levin 1993, Levin and Rappapport Hovav 1995) which determine the syntactic position of the arguments of the verb, and can be used to predict when a verb will not participate in an alternation (Davis 1995, Markantonatou and Sadler 1995). These semantic distinctions therefore directly interact with syntactic form and must be considered a part of lexical knowledge rather than world knowledge about the verb forms.
Bill loaded books onto the truck. Bill loaded the truck with books.