I wonder how they'll treat the race to ring-in, which in the human game begins just after Trebek finishes reading the question. (Apparently there's an in-studio light indicating to contestants when ringing-in is possible.)
The program might want to make its ring-in decision based on a snap decision of whether the question is likely to be amenable to its analysis, even if the exact answer isn't yet know -- human contestants do this, and then take a second or two to cogitate before giving their actual answer.
The computer has the opposite problem than humans have. It has all the information available instantly but it can't understand the question. Parsing a question can be very quick, the problem is being sure that it was understood correctly. If it was, the answer will be known instantly and the computer should ring. If it wasn't, it shouldn't.
My guess is that the computer will ring either when it knows the answer beyond a certain confidence threshold (faster than a human being can hit a button) or not at all.
I suggest there may be ways to predict "this is the kind of question I can usually parse and answer" long before a confident parse-and-settled-answer could complete. Thus it could make sense to ring in before any potential answers have been formulated or ranked.
I agree. As an example, if it is using traditional natural language parsing techniques, then it should know whether there are less than ten syntax trees or less than one thousand syntax trees by ring-in time. In this case, the former can be determined with a much higher degree of confidence. Even short phrases can have a multitude of possibilities, the canonical example being, 'Time flies like an arrow.' Of course, I have no idea how they're actually approaching this problem, but the point is, heuristics can be used in the process of question-answering to indicate enough of a confidence level for ring-in.
The program might want to make its ring-in decision based on a snap decision of whether the question is likely to be amenable to its analysis, even if the exact answer isn't yet know -- human contestants do this, and then take a second or two to cogitate before giving their actual answer.