Abstract

A well-written text follows an overall structure, with each sentence following naturally from the ones before and leading into the ones which come afterwards. We call this structure ‘coherence’; without it, a document becomes a confusing series of non sequiturs. Understanding the principles that make a text coherent is an important goal of natural language processing. These principles can be applied to the design of systems that create new documents, like summaries, or evaluate the writing style of existing documents.

In this talk, I focus on entity-based coherence, the structural principles dictating the way a text refers to objects in the world. I will present models that describe the way important entities are introduced to the discourse and subsequently referred to, and that automatically determine which phrases refer to important entities and which do not. I will also discuss several methods for evaluating models of coherence, showing in particular that conversational language poses a challenge for several models that work well on text.

I will also present some new work on the high-level plot structure of novels. Plot is a higher-level structure than coherence, and cannot be modeled purely in terms of sentences; I will present a system that represents novels in terms of characters and their emotions over time. Finally, I will discuss the lessons we can learn from sentence-level models when designing such a high-level system.