THE THESIS DEFENSE FOR THE M.S. DEGREE IN COMPUTER SCIENCE FOR

                                    Michael Whidby

Citation sentences (sentences that cite other papers) play a key role in the summarization of scientific articles. However, a citation-based summarization system that depends on generic natural language processing components, such as parsers or sentence compressors, will perform poorly if those components cannot handle citations correctly.

In this thesis, I examine the effect of citation handling on parsing, sentence compression, and multi-document summarization. There are two types of citations that occur in citation sentences: constituent citations and parenthetical citations. I propose an automatic citation classifier based on training data created through Mechanical Turk tasks. I demonstrate that the use of type-specific citation handling as pre-processing improves the performance of a state-of-the-art generic parser, both for quality of the parse trees and running time. Extrinsic evaluations demonstrate that improving the performance of a parser on citation sentences in turn improves the performance of a sentence compressor, Trimmer (Zajic et al., 2007), and a multi-document summarization system, MASCS, according to several summarization measures.

Examining Committee:

CO-ADVISOR                                    Dr. David Zajic

Committee Member(s):                      Dr. Hal Daume III

EVERYONE IS INVITED TO ATTEND THE PRESENTATION PORTION OF THIS DEFENSE