Long-Answer Question Answering and Rhetorical-Semantic Relations
--
Over the past decade, Question Answering (QA) has generated
considerable interest and participation in the fields of Natural
Language Processing and Information Retrieval.  Conferences such as
TREC, CLEF and DUC have examined various aspects of the QA task in the
academic community. In the commercial world, major search engines from
Google, Microsoft and Yahoo have integrated basic QA capabilities into
their core web search.

These efforts have focused largely on so-called ``factoid'' questions
seeking a single fact, such as the birthdate of an individual or the
capital city of a country. Yet in the past few years, there has been
growing recognition of a broad class of "long-answer" questions
which cannot be satisfactorily answered in this framework, such as
those seeking a definition, explanation, or other descriptive
information in response. In this thesis, we consider the problem of
answering such questions, with particular focus on the contribution to
be made by integrating rhetorical and semantic models.

We present DefScriber, a system for answering definitional ("What is
X?"), biographical ("Who is X?") and other long-answer questions
using a hybrid of goal- and data-driven methods. Our goal-driven, or
top-down, approach is motivated by a set of "definitional
predicates" which capture information types commonly useful in
definitions; our data-driven, or bottom-up, approach uses dynamic
analysis of input data to guide answer content. In several
evaluations, we demonstrate that DefScriber outperforms competitive
summarization techniques, and ranks among the top long-answer QA
systems being developed by others.

Motivated by our experience with definitional predicates in
DefScriber, we pursue a set of experiments which automatically acquire
broad-coverage lexical models of "rhetorical-semantic relations"
(RSRs) such as Cause and Contrast. Building on the framework of Marcu
and Echihabi (2002), we implement techniques to
improve the quality of these models using syntactic filtering and
topic segmentation, and present evaluation results showing that these
methods can improve the accuracy of relation classification.

Lastly, we implement two approaches for applying the knowledge in our
RSR models to enhance the performance and scope of DefScriber. First,
we integrate RSR models into the answer-building process in
DefScriber, finding incremental improvements with respect to the
content and ordering of responses. Second, we use our RSR models to
help identify relevant answer material for an exploratory class of
"relation-focused" questions which seek explanatory or comparative
responses. We demonstrate that in the case of explanation questions,
using RSRs can lead to significantly more relevant responses.