Thesis proposal presentation

                              Zhu Zhang

               Candidate for SI/EECS IIDP (dual Ph.D. degree)

                          Friday, October 1
                            412 West Hall
                            12:30 - 2:30

Title:
        Natural Language Relations: Classification and Application

Abstract:

In this dissertation proposal, we study computational models for
classification and application of natural language relations.
Specifically, two type of relations are explored: cross-document
structural relations (semantic connections between sentences cross
document boundary) and semantic relations between entities (in the
context of information extraction).

The first part of the proposal deals with corpus-based
classification of the relations using machine learning techniques.
Different weakly supervised learning algorithms are presented and
empirically evaluated on real data sets. It is shown that
unlabeled data can be used either to boost classification
performance or to reduce the need for labeled data.

The second part of the proposal investigates the application of both
relations to other NLP problems such as text summarization and
question answering. It is shown that CST relations can be used in
a simple algorithm to enhance the output of extractive summarizer.

Committee:

Dragomir R. Radev - chair
Steve Abney
Satinder Singh
David States