Backchannels, as an exemplary dialog phenomenon, have received much attention, but the details of their forms and functions are still not well understood. We examined the prosodic properties of 280 backchannel tokens annotated as "un" or "u:n". In addition to the expected difference in length, there was a difference in pitch: whereas the "un" tokens had falling pitch, the pitch of the "u:n" tokens was closer to flat. One way to describe this difference is to postulate that "un" has a lexical accent but "u-n"does not. We discuss the implications for transcription standards and for modeling.
