ABSTRACT
The purpose of this project was to provide a solution to the problem set out in the introduction, namely how can we better model when speakers choose to use noncanonical syntactic forms. Previous literature provided only necessary conditions which must be fulfilled if the use of the non-canonicals are to be felicitous. This simple model of condition-form pairing, where there is a one-to-one mapping from conditions to form, is unable to account for a variety of problematic cases: (1) cases where the discourse conditions are present but the form is absent, (2) cases where the discourse conditions are present, but the form is still infelicitous, and (3) cases where the discourse conditions hold, and the non-canonical is not merely acceptable, but in fact crucial to the meaning and coherence of the discourse in that particular context. In addition, this simple model is of little use in an NLG system because of its inherent capacity for overgeneration, such that the non-canonical form is generated whenever the necessary conditions hold.
