Throughout modern linguistic theory, it has been assumed almost without exception that sentences are best represented in terms of certain types of abstract graphs called trees. The choice of trees as underlying sentence structure captures a number of intuitions about the building blocks of sentence structure, what are called constituents: they are contiguous; each constituent has a unique structural position; every independent sentence comprises a single constituent; etc. A grammatical theory must have some way of selecting those trees that represent well-formed sentences. Consequently, it must make use of some primitive vocabulary for talking about properties of linguistically relevant tree structures. The question of what enters into this primitive vocabulary has been relatively neglected. While the domination relation has been widely assumed as a primitive, this project proposes that c-command, a highly grammatically relevant structural relation, is a better choice and should instead play the the central role of characterizing well-formedness. The proposed research will address the following issues: (1) What is the precise nature of c-command? What is the range of structural objects that it allows us to define? (2) Is c-command expressive enough to define the constructs of grammatical theory, while not being overly permissive as to allow the definability of linguistically irrelevant structures? (3) To what degree can the use of primitive c-command elucidate the otherwise puzzling dichotomy between segments and categories that has been introduced in recent work to distinguish different ways of relating constituents together? (4) In what respect are views of grammar that build structural representations using sequences of derivational steps simplified in the context of primitive c-command, in particular Chomsky's minimalist proposals as well as Tree Adjoining Grammars, a system which has achieved significant results both in theoretical and computational li nguistics? (5) What are the implications of c-command for models of human sentence processing? In particular, to what degree can the assumption that the parser manipulates c-command relations account for the relative acceptability of certain local ambiguities while explaining the existence of processing difficulty for others? This project also aims to show that the use of c-command in sentence processing can unify two competing strands of work in the development of sentence processing models based on semantic dependencies and on structural configurations, maintaining the advantages of each.
