In this talk I will present ongoing work in finding and using certain
types of rhetorical-semantic relationships in text, such as hypernymy,
cause-effect, and contrast. I will first discuss my work in
definitional QA (i.e. answering "Who/What is X?"), with some examples
from the CUAQ/DefScriber system, which combines multi-document
summarization methods with pattern-based relation finding for the
hypernymy relation. I will then discuss some of the issues involved in
finding these kinds of relations in an automated fashion, focusing on
some techniques I have tried and the results obtained. Lastly, I will
discuss prospects for using these relations both for definitional QA
as well as other NLP tasks.
|