Anyone with even passing experience of synthesized speech, text-to-speech, or speech recognition technologies is acutely aware of just how much we still don't understand about rhythm and intonation in human speech. Intonation is extremely important: Changes to an intonation contour -- the characteristic rises and falls of the voice's pitch during an utterance -- can change the meaning conveyed, often dramatically. (Compare "He's reliable." vs. "He's reliable?") Getting intonation wrong, even if everything else is right, makes speech sound at best non-native, at worst even non-human. Fixing this problem is a serious challenge, because intonational distinctions are slippery. Most native speakers cannot say with confidence whether a given utterance does or does not convey "incredulity", or "resigned acceptance", in the same way that they can usually tell whether a given sequence of sounds does or does not mean "cat".
The problem of what intonation means and what exactly carries that meaning is reflected in experimental studies as well; researchers frequently disagree about the interpretation of results. This project aims to overcome this by taking on several particularly difficult examples of American English intonation, using several complementary experimental methods. Comparison of the results from these different approaches will answer not just isolated questions about American English, but more importantly broader questions about how to test competing theories of intonation. The goal is a consensus method for investigating intonation in human language in general. Success in reaching that goal will improve emerging speech technologies and their application across languages, as well as helping linguistic field workers and second-language teachers grappling with questions of how to convey what is meaningful in a given language's intonation system.
