| Speaker identification is the task of attributing utterances to characters in literary narratives. Although only some of the utterances are explicitly attributed in novels, humans readers are able to determine the speakers of the remaining utterances because of their understanding of the plot. This dissertation proposes a method to automatically identify the speakers using supervised machine learning methods that utilize various text clues and a speaker alternation pattern. In addition, the method incorporates an unsupervised actor-topic model that aims to distinguish speakers depending on the content of their statements. The experimental results show that the method substantially outperforms a baseline method, and is competitive and more general when compared to previous approaches to the problem. |
