Great expectations: unsupervised inference of suspense, surprise and salience in storytelling
Stories interest us not because they are a sequence of mundane and predictable events but because they have drama and tension. Crucial to creating dramatic and exciting stories are surprise and suspense. Likewise, certain events are key to the plot and more important than others. Importance is referred to as salience. Inferring suspense, surprise and salience are highly challenging for computational systems. It is difficult because all these elements require a strong comprehension of the characters and their motivations, places, changes over time, and the cause/effect of complex interactions. Recently advances in machine learning (often called deep learning) have substantially improved in many language-related tasks, including story comprehension and story writing. Most of these systems rely on supervision; that is, huge numbers of people need to tag large quantities of data to tell the system what to teach these systems. An example would be tagging which events are suspenseful. It is highly inflexible and costly. Instead, the thesis trains a series of deep learning models via only reading stories, a self-supervised (or unsupervised) system. Narrative theory methods (rules and procedures) are applied to the knowledge built into the deep learning models to directly infer salience, surprise, and salience in stories. Extensions add memory and external knowledge from story plots and from Wikipedia to infer salience on novels such as Great Expectations and plays such as Macbeth. Other work adapts the models as a planning system for generating new stories. The thesis finds that applying the narrative theory to deep learning models can align with the typical reader. In follow up work, the insights could help improve computer models for tasks such as automatic story writing, assistance for writing, summarising or editing stories. Moreover, the approach of applying narrative theory to the inherent qualities built in a system that learns itself (self-supervised) from reading from books, watching videos, listening to audio is much cheaper and more adaptable to other domains and tasks. Progress is swift in improving self-supervised systems. As such, the thesis's relevance is that applying domain expertise with these systems may be a more productive approach in many areas of interest for applying machine learning.