Both the reseach community and commercial enterprises have set out to address anew some of the ancient conundrums of question answering and to answer new challenges posed by the growth of the internet and its usership. Successes in the question answering track in TREC and elsewhere have filled the sails of research and industry projects: novel combinations of NLP and IR have made it possible to provide in many cases a concise response to short, factual questions. However, the application of corpus techniques to the analysis of large query sets provides some sobering data points about the actual expectations and interests of users of such systems. Initial results provide food for thought and humbling insights into the linguistic and computational complexity of the tasks ahead. It is likely that appropriate responses to many types of questions will remain beyond the abilities of automatic solutions for some time to come.