Negotiation, feedback, and perspective within natural language generation

This thesis is an investigation of how natural language generation can take advantage of the ways that language use goes beyond simple, straightforward transmission of information. The two main contributions of the work are the use of annotations to relate linguistic choice to text planning and the use of perspective to model dependence and effects on attitude and context. Placing annotations on linguistic options allow the text planner to detect and respond to interactions between linguistic choices and the communicative plan driving those choices, without conflating the planning and linguistic levels of decision-making. The explicit modeling of perspective and perspective shifts allows the perspective to influence the generator\u27s choices; in particular, the generator can take into account the ways that particular linguistic choices can affect or alter the subsequent perspective. The use of these techniques has been demonstrated in the IGEN implementation, resulting in a generator that can take advantage of the flexibility of natural language to frame its output in a way that reinforces the underlying goals driving the generation