Anais G. Rivera and Nigel Ward, University of Texas at El Paso
Prosodic Cues that Lead to Back-Channel Feedback in Northern Mexican Spanish
In order to demonstrate attentiveness during a conversation it is necessary for the listener to provide feedback. Most commonly the listener produces a back-channel utterance to let the speaker hold the turn while at the same time demonstrating interest and understanding. To some extent, the times when back-channel feedback are welcome are determined by the speaker and conveyed to the listener with prosodic cues. The ability to back-channel correctly is specially important to prevent misunderstandings between speakers of different languages like English and Spanish (Berry 1994). In this study the prosodic features of conversations between native Spanish speakers were analyzed to identify these cues.
The corpus consisted of two male-male conversations, one male-female conversation and two female-female conversations (Acosta 2004). All the subjects were from the northern Mexican region. The corpus was first manually labeled to identify the back-channel utterances. In order to be considered a back-channel the utterance had to: (1) respond directly to the content of an utterance of the other, (2) be optional, (3) not require acknowledgement by the other (Ward & Tsukahara 2000).
Analysis focused on the prosodic behavior of speakers previous to the back-channel utterances. In Northern Mexican Spanish the cues are characterized by a pitch downslope followed by a pitch rise accompanied by a rate reduction on the last syllable and a drop in energy leading to a slight pause. More specifically the low pitch must last for at least 50ms and for no more than 200ms and be below the 28 percentile line. The pitch rise must last at least 80ms and no longer than 300ms and end higher than the 75 pitch percentile. The rate reduction is measured by a long vowel of at least 100ms which must occur before the energy drop.
The rule gives an output of 28.7% coverage and 14.2% accuracy over the entire corpus. There are significant differences in performance between conversations involving two males and those in which there was a female speaker involved. The total coverage for the male-only conversations was 18.8% and the accuracy was 9.4%. On the other hand the coverage for the female-female and male-female conversations was 30.2% and the accuracy was 17.8%.
Errors, that is, misses or false positives, were subsequently divided into four sub-categories for classification purposes: the first consisted of yes-no questions which are characterized by a drop in pitch followed by a pitch rise. The prosodic similarities between questions and back-channel cues make them challenging to distinguish (questions tend to be longer, and final lengthening is less common, but there are many exceptions) and this produced many false positives. Another sub-category consisted of overlapped speech, which in Spanish is more common than in English (Berry 1994), which caused the rule to miss some back-channels due to the lack of a pause. The other two categories consisted of a pitch drop feature; this seems to also sometimes serve as a cue for a back-channel response. Some of these were due to an amusing utterance which made the listener back-channel with laughter. Pitch-drop based back-channels were more common in the male-male dialogs, also in one the speakers appeared less focused on the conversation and the back-channels appear in less consistent places.
References
Acosta-Reyes, Luis H., 2004. Prosodic Features that Cue Back-Channel Responses in Northern
Mexican Spanish. Masters Thesis, Univ. of Texas at El Paso, Dept. of Computer Science .
Berry, Anne, 1994. Spanish and American Turn-Taking Styles: A Comparative Study.
Pragmatic and Language Learning Monograph Series, Volume 5, pp 180-190.
Ward, Nigel and Wataru,Tsukahara, 2000. Prosodic Features which cue Back-Channel Feedback in English
and Japanese. Journal of Pragmatics, 23, pp 1117-1207.
