Date of Award

Spring 1-1-2014

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

James H. Martin

Second Advisor

Martha S. Palmer

Third Advisor

Wayne H. Ward

Abstract

Unsupervised training has recently been successfully used to enhance the performance of neural networks. To understand the advantage provided by the structure of unsupervised pre trained models, a network theory based analysis of word representation similarities was performed, revealing the structure discovered by unsupervised models trained on a large english language corpus. A Part of Speech Tagger and two versions of Semantic Role Labelers were defined and tested to explore architectural configurations and training strategies. In order to thoroughly test various Neural Network Natural Language Models, a highly configurable software implementation was developed.

Share

COinS