Argument Structure and Referent Systems

We describe here a theory of semantic composition, which integrates referent systems into argument structure. The main idea is that argument structure is the key component in linguistic signs, and that it provides the interface between syntax and semantics. To make this idea work for language, referent systems as defined in Vermeulen (1995) have been used. Referent systems (RS) are based on the idea that the identity of variables across different representations is irrelevant (see also Fine (2007)). Thus, whether or not two variables must be taken to be coreferential in the formation of a syntactic constituent must be effected by encoded linguistic cues, be they word order or a particular word form. This basic insight started the present investigation. Many adaptations had to be made in order to make the basic idea compatible with the requirements of natural language. In RSs the natural merge operation in semantics makes all variables occurring free in the respective representations disjoint. Thus, in contrast to an idea promoted in Zeevat (1989), where variables shared across representations are taken to be pointing to the same object, here this is treated as an unwarranted assumption. In natural languages, however, there are clearly defined circumstances in which certain variables need to be identified during the merge operation, for example when a head is merged with its complement. The way this is done in RSs is that this fact is explicitly encoded by means of a shared name. This name is taken from a specified set of names. This idea has been the starting point of the current theory. This theory takes it that argument structures are lists of argument identification statements (AISs), that need to be discharged one by one in order to reach the phrase level. A discharge is obtained by matching a single variable from the argument structure in both the functor and its argument. This match is subject to several conditions: (i) the variables occur in particular AISs that are visible during match, (ii) the variables have matching names, and (iii) no morphological conditions are violated. In the present paper we shall simplify the actual theory in order to concentrate on its main engine, the argument structure. In particular, we shall not say much about the morphology. The theory is fully implemented, see the section on implementation for source code.