Learning how to answer questions using trivia games
Gideon S. Mann · 2002
In this paper we examine a sentence comprehension task: given a question, and an extended sentence known to answer that question, the goal is to extract the short answer to the question. As an initial solution, a novel robust statistical model is presented which combines the semantics of the expected answer with the expected context within which the answer will be found. Two distinct trivia game databases, with no additional annotation, are used to train and test the model.