Machine Learning Thesis
Domain-independent narrative content generation.
The problem of narrative generation, or how to get computers to write coherently plotted interesting stories, has typically been tackled by using a corpus of human-authored information. This approach simplifies the problem to arranging manually-defined events and characters in a logical way and then generating decent prose. To fully achieve narrative intelligence, we will have to overcome the necessity of using human-authored input at the beginning of the process. This thesis therefore focuses on learning appropriate narrative events and characters from works of literature, as a human author would. These events can then be assembled into a plot graph, which can be traversed based on sequential probability to produce a coherent plot.