Finding an answer to a question

The huge quantity of available electronic information leads to a growing need for users to have tools able to be precise and selective. These kinds of tools have to provide answers to requests quite rapidly without requiring the user to explore each document, to reformulate her request or to seek for the answer inside documents. From that viewpoint, finding an answer consists not only in finding relevant documents but also in extracting relevant parts. This leads us to express the question-answering problem in terms of an information retrieval problem that can be solved using natural language processing (NLP) approaches. In my talk, I will focus on defining what a "good" answer is, and how a system can find it. A good answer has to give the required piece of information. However, it is not sufficient; it also has both to be presented within its context of interpretation and to be justified in order to give a user means to evaluate if the answer fits her needs and is appropriate. One can view searching an answer to a question as a reformulation problem: according to what is asked, find one of the different linguistic expressions of the answer in all candidate sentences. Within this framework, interlingual question-answering can also be seen as another kind of linguistic variation. The answer phrasing can be considered as an affirmative reformulation of the question, partly or totally, which entails the definition of models that match with sentences containing the answer. According to the different approaches, the kinds of model and the matching criteria greatly differ. It can consist in building a structured representation that makes explicit the semantic relations between the concepts of the question and that is compared to a similar representation of sentences. As this approach requires a syntactic parser and a semantic knowledge base, which are not always available in all the languages, systems often apply a less formal approach based on a similarity measure between a passage and the question and answers are extracted from highest scored passages. Similarity involves different criteria: question terms and their linguistic variations in passages, syntactic proximity, answer type. We will see that, in such an approach, justifications can be envisioned by using text themselves, considered as depositories of semantic knowledge. I will focus on the approach the LIR group of LIMSI has taken for its monolingual and bilingual systems.