Complex questions: Let me Google it for you

Many questions submitted to community question answering (CQA) sites such as Yahoo!Answers can be reasonably answered by a simple query to a standard Web search engine. However, the user may not know how to reduce their question, which may be several paragraphs in length, down to the 2-4 terms that will return an answer. The work in this short paper is directed towards the creation of a system that can do this reduction for them. More specifically, in this paper we focus on automatically generating training data for this purpose. Given existing question-answer pairs, we determine the combination of terms appearing in the question that best returns the answer. Given the large collection of question-answer pairs that already exist on CQA sites, we hope to generate large numbers of training examples. Using this training data, we hope to learn to extract short queries from long questions that have no existing answers. Although the results in this workshop paper are preliminary in nature, they demonstrate the potential of this approach.