bnevin@BBNCCH.ARPA (Bruce Nevin) (03/12/86)
To elaborate on points made by Doug Ice and Andy Walker, sentences are typically disambiguated in English with appropriate intonation. There are tricks of punctuation to capture most of the tricks of intonation, and though third-level or deeper nestings are awkward for punctuation, they are also awkward for intonation. There is a perverse kind of `rule of the game' in linguistics that one should read ambiguous examples with flat intonation so as not to force the audience interpretation one way or another. Seems to me this is absurd. Unless the aim is to put them in the hapless position of a machine being given the written sentence with poor or inadequate punctuation. Arguing on the other side, when readers find the appropriate intonation for a poorly punctuated sentence they rely on the redundancy that pervades language. Since machines are expected to cope with all sorts of ill-formed input, poor punctuation being the least of it, we must provide means for them to do the same. (In fact, most readers do a poor job of finding the appropriate intonations when reading text . . . probably because they become so narrowly focussed on the word-by-word and sentence-by-sentence decoding task that they cut themselves off from the possibilities of discourse structure, nonverbal communication, and knowledge-base-type pretext and context, which their imaginations churn out for them on a `parallel' track, if they only pay attention. Could there be a clue here why machines are having trouble?) Bruce Nevin bn@bbncch.arpa