An Empirical View on IQA Follow-up Questions

In a realistic Interactive Question Answering (IQA) situation, one third of the users pose follow-up questions, i.e., go beyond a single question per dialogue. We identify two different perspectives according to which these follow-ups can be described: informational transitions and context dependency. By understanding exactly how informational transitions occur in IQA dialogues, we propose a method to guarantee that focus tree based IQA systems provide wide coverage of follow-up questions that trigger the respective set of informational transitions.