Answering questions from text and knowledge base
Abstract: Question answering systems provide answers to questions that are expressed in natural language, so that a user does not have to care about how the information is represented and how it will be retrieved. We are interested in questions like "Who is Bill Clinton's daughter married to?", i.e. factual open domain questions.
Answers can be found in texts or in knowledge bases, like DBPedia for example. Each kind of resource requires a different kind of process, but both require to analyze the question in order to bring it closer to the form in which answers are expressed, and that must take into account the inherent variability of natural language. Thus we will present the problems that have to be modeled in this context and how they are currently solved in the state of the art for both textual and knowledge base search.
To go further, it would be interesting to conceive hybrid systems where the capabilities of each system can benefit the other, leading to overall performance enhancement, either by enlarging the information covered or the possibility to retrieve it. We will also present the few approaches that exist and how we envision this problem.