Joe Pfeiffer <pfeiffer@cs.nmsu.edu> wrote in message news:<1bg0bbzdwq.fsf@cs.nmsu.edu>...> "S. F. Thomas" <sf.thomas@verizon.net> writes: >> >> See my comment on the use of a degree scale vs the use of a yes/no >> response for purposes of evaluating calibrational propositions in my >> response to Joe Pfeiffer in this thread. To repeat briefly: >> utterances >> come whole, not in degrees, and the calibrational question should >> in my opinion reflect that, if what we are attempting to do is to >> model actual language use. Having said that, I readily admit that our >> modeling can jump from meta-language to meta-meta-language if our >> experiments are designed to measure not subjects' own actual use, but >> rather their estimates of overall population use, which is what I >> think >> a degree response captures. That is interesting too. But I think it >> is >> better to start with the fundamental (the subject's own use), rather >> than with his estimate of behavior for the population as a whole ... >> let the modeler do that, rather than the subject. > > But utterances do come in degrees -- that, to me, is the whole point. > When we say ``the water is pretty hot'' the ``pretty'' is qualifying > the ``hot.''
It is still a whole utterance, hedged to be sure, but it is still a whole utterance, in which the relevant term is the hedged "pretty hot", rather than the plain "hot". These two terms mean different things, presumably, and each could be the subject of its own calibrational experiments.> I'll agree that the questions should try to capture the subject's own > use -- but the question, ``on a scale of 0 to 10, to what extent would > you say Ringo is tall?'' seems like it would better capture it than > ``is John tall? No, you're not allowed to say `sort of,' say yes or > no.''
The point is that you ask yes or no of a large enough number of randomly selected competent speakers of the language. The fraction concurring gives you the "degree" that you are after. The problem I have with eliciting a response based on a scale from 0 to 10 is that the meaning of the question, and the response desired, is ambiguous. Why use a vague, ambiguous concept when a concrete, clear one will do ... and moreover get to the heart of the issue. Terms are fuzzy *because* of the randomness of response that can be expected even in a calibrational setting, hence there is uncertainty in the inference that may be drawn, in a non-calibrational setting, regarding the use of the term. Regards, S. F. Thomas