Question Answering (QA) provides more precise responses to users'
queries than Information Retrieval, which provides users a list
of and links to, relevance-ranked documents in response to a
query. Our approach to QA takes advantage of the NLP capabilities
we have developed at CNLP on both questions and the answer-providing
sources. During query processing, the system converts the question
into a logical query representation used for first stage access
into the document collection, as well as expanding the query
to its semantic equivalents, and determining the focus of the
query. Answer finding in sources combines two different approaches
- keyword & NLP-based inferencing, after which answer triangulation
takes place to select the most likely answer, given the system's
detailed understanding of the user's question.
We participated in last year's TREC Conference, http://TREC.nist.gov
where our system was tested using 693 general interest factual
questions posed against almost a million source documents.
Our system performed well, ranking amongst the top echelon
of the 28 participating systems.
QA continues as an active area of research at CNLP and many
of our funded projects focus on QA. We will be adding even
more sophisticated abilities in the coming months and expect
to see even better QA results.
|
