Presuppositions and Synchronous Tree-Adjoining Grammar

Abstract

Technology continues to evolve at a rapid rate, and with it the desire for more human-like interactions with the resulting systems. An obvious area for improvement is in natural language comprehension. Communicating by means of natural language requires more than just producing sounds that mimic language: complex interactions go beyond sounds to the meaning behind the utterance. Thus, a sophisticated system understands not only what the user has said, but also what was not said. Presuppositions are one such inference where the material is implicit rather than explicit. How systems detect the presence of, and respond to, presuppositions can vastly improve or detract from the interaction. However, in order to respond appropriately to presuppositions, systems must be able to represent them in such a way that efficient computation over the information can be performed. This thesis will propose an extension to synchronous tree-adjoining grammar to incorporate the notion of presupposition as presented by Karttunen (1973). The thesis will begin by defining and exploring the concept of a presupposition. It will then detail the Karttunen (1973) theory of plugs, holes and filters to explain presupposition projection. Before presenting an introduction to synchronous tree-adjoining grammar, the thesis will briefly discuss pre-existing attempts to model presuppositions for computation, thereby illustrating the importance and the difficulty of the problem at hand. Having introduced synchronous tree-adjoining grammar, the thesis will then constructively guide the reader through the extension to synchronous tree-adjoining grammar that implements the plug, hole and filter theory. Finally, the thesis will discuss potential additions to and applications of the presented extension.