Electronic Theses and Dissertations





Date of Award


Document Type

Dissertation (Access Restricted)

Degree Name

Doctor of Philosophy


Computer Science

Committee Chair

King-Ip Lin

Committee Member

Max M Louwerse

Committee Member

Vasile Rus

Committee Member

Scott Fleming


Thematic role labeling has been an area of interest in several domains of natural language processing such as language generation and information retrieval. Extracting thematic roles efficiently and with minimal supervision affects all the above areas. Most of the existing computational methods for thematic role extraction are highly dependent on human annotated corpora, and are driven by the rules generated by supervised processes. In this work I extracted thematic roles using a data-driven approach. I built an unsupervised language generation model inspired from the ADIOS model, to learn recurring substructures from language, and with minimal supervision learned the rules that are needed to identify thematic roles. I tested the consistency of the sub-structures encoding thematic role information over PropBank annotated sentences. Results indicated that sub-structures consistently hold semantic role information, and the method robustly showed that thematic role information could be extracted with minimal supervision.


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.