Configurational languages like English and non-configurational languages like Russian use both syntax and prosody to encode discourse functions. This crucial interaction of syntax, phonology, and discourse functions has been recently brought to the forefront of linguistic research (É. Kiss 1995 and references therein).
The division of constituent-structure from functional-structure, as well as the possibility of an information (discourse function)-structure, makes Lexical Functional Grammar well suited for capturing these interactions (King 1993/1995, Choi 1996). This paper explores a problem posed by these interactions, in particular by the association of discourse functions with particular c-structure positions and their f-structure counterparts. In LFG, discourse function information has traditionally been encoded in the f-structure via annotations on the c-structure. In the case of structurally encoded topic and focus of arguments, this approach as worked adequately, especially for syntacticized discourse functions. However, in certain cases involving f-structure heads, the standard annotations result in the incorrect scoping of the discourse functions in that more material is focused or topicalised than intended. As such, I propose that discourse function information be captured in an independept projection, i(nformation)-structure, which is a projection off the c-structure and which is accessible to the s(emantic)-structure, and that the information relevant to the i-structure is the core predicate value without its associated argument structure.
Note: for more accurate figures and diagrams, see the postscript version of the paper.