Document Type



Master of Science


Computer Science

First Adviser

Baird, Henry S.


An investigation into algorithms for generalized, dynamic sentiment tracking in natural language is reported. Little progress has been made in the automatic analysis of literary fiction. However, accurate recognition and robust modeling of text's emotional content would allow for large scale analysis of the ever increasing number of digitized books in addition to other applications such as sorting, searching, and making book recommendations. But the challenge of human-like reading of these books not only includes computational hurdles but inherent ambiguity. Humans often argue over the correct interpretation of a work of fiction due to subjectivity. To attack this open problem, we resort to shallow statistical methods. We propose a method for tracking the sentiment of the interpersonal relationships of literary characters by leveraging text structure to estimate the direction of sentiment flow. Given structured dialogue (as found in a play), we know who is speaking and can perform standard sentiment analysis on their speech via an emotion lexicon. Furthermore, we can guess at who is listening and direct the sentiment accordingly, producing lists of a character's enemies and allies. Experiments on all of Shakespeare's plays are presented along with discussion of how these methods can be extended to unstructured texts.