Date Thesis Awarded


Access Type

Honors Thesis -- Access Restricted On-Campus Only

Degree Name

Bachelors of Science (BS)


Computer Science


Xipeng Shen

Committee Members

Ann Reed

Haining Wang


This paper investigates two areas of coreference resolution --specialization and cataphoricity. In doing so I attempt to build upon an existing state-of-the-art system to achieve greater performance. Coreference systems utilizing specialization of models and specialization of features have both been previously proposed, but no investigation has been made as to their relative effectiveness or possible interrelationship. In this paper I demonstrate that most readily constructible specialization models are equivalent. The existence of cataphoric mentions is largely ignored in coreference resolution. In this paper I introduce several proposals for countering the performance losses due to cataphora. In particular, I propose a method of cataphoricity classification that largely counters these losses. I present results for several potential methods of using this classifier to create performance gains --particularly joint determination using integer linear programming. These methods are demonstrated to be ineffective, providing guidelines for future work.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.


Thesis is part of Honors ETD pilot project, 2008-2013. Migrated from Dspace in 2016.

On-Campus Access Only