Some Exotic Discourse Markers of Spoken Dialog

Some Exotic Discourse Markers of Spoken Dialog Nigel Ward 1 University of Tokyo Abstract: In Japanese and English conver- sation corpora we have found an interesting discourse marker: a 110 millisecond region of low pitch region, which can cue back-channel feedback from the hearer. We have also found that many back-channel responses appear to be more explainable if considered not as words, but as complex sounds which directly convey dis- course functions with sotmd symbolism. c) coming after at least 700ms of speech, d) providing you have not output back-channel feedback within the preceding 1.0 seconds, e) after 350ms wait you should produce back-channel feedback. Our model for English is the same except for some parameters: pitch level (clause a) 26th-percentile, re- covery time (clause d) 800ms, and delay (clause e) 700ms. 1 Introduction Back-channel feedback, also called "listener re- sponses'~ is, to a first approximation, those responses produced by one participant which do not interfere with utterances by the other participant. In Ameri- can English yeah, mm and uh huh are typical back- channel feedback. In Japanese un is most typical. This paper summarizes some findings on back- channels and their cues, as primitive examples of dis- course markers in spoken dialog. Definitions, details, discussion, and references appear elsewhere (Ward 1996; ~Vard 1998; Ward & Tsukahara to appear: ~vVard submitted; Ward & Tsukahara 1998). The findings arose from our corpora of unrestricted two- person conversations in Japanese and in American English, and might not apply to ta,sk-directed dis- courses, small talk, multi-party conversations, and conversations with audiences. 2 A Cue for Back-channel Feedback In Japanese and in English, the prosody of the speaker's utterances can mark times when the lis- tener is welcome to produce back-channel feedback. One specific cue is a region of low pitch. Behavior of Japanese listeners can be modeled with the following rule: Upon detection of a) a region of pitch less than the 28th-percentile pitch level and b) continuing for at least 110ms, I nigel~@s~npo.t.u-tokyo.ac.jp, h t tp://www.sanpo,t.u-~ okyo. ac.jp/~nigel/, Mech-lnfo Engineering, Urliversity of Tokyo. Bunkyo-ku Tokyo 113-86.56 Japan. Thanks to Wataru Tsukahara, Dan Jurafsky. and anonymous reviewers for suggestions, and to the Hayao Nakav'ama FoLm- dation and the [naraori Foundation for support. 3 Corpus-based Evaluation We have tested the predictions of the above rule against a corpus of ca~sual, mostly friendly Japanese conversations between mostly students (80 minutes total). The rule gave a coverage of 56(~ (of 873 back- channels in the corpus, the rule predicted 496, using a 500 ms tolerance) and an accuracy of 34~ (of 1447 predictions, 496 were correct). This did better than a rule which makes predictions at random (while obey- ing clauses c, d, and e), both on average (Table 1), and for most speakers in most conversations: in 34 cases out of 36 (2 sides times 18 conversations) the figure of nlerit (namely the product of coverage and accuracy) was higher for the low pitch rule. Analysis of the false predictions for Japanese sug- gests that …

Content truncated to 500 words