utterancewhen

Utterancewhen is a term used in linguistics and natural language processing to denote the temporal context of a spoken utterance. It refers to when an utterance occurs within a discourse timeline, including its position in turn-taking, its simultaneity with other speakers, and its relation to preceding events or statements. The term blends utterance with when to emphasize timing information accompanying speech.

In annotated corpora and transcripts, utterancewhen is stored as metadata attached to each utterance, allowing researchers to query timing information alongside the textual content. Representations can be absolute clock time, such as a timestamp like 00:02:15, or relative timing, such as timing relative to the start of a recording or to the previous utterance. Annotations may also indicate overlap, interruption, or simultaneous activity between speakers.

Examples of use include labeling a line such as: Speaker A: Let’s begin. [utterancewhen: 00:01:23], Speaker B: Okay. [utterancewhen: 00:01:24]. In transcripts, this enables precise alignment between spoken text and audio, and supports analyses of pacing, latency between turns, and the handling of interruptions.

Applications and status: Utterancewhen supports studies of turn-taking dynamics, discourse pacing, and the synchronization of transcripts with audio or video. In dialogue systems and conversational AI, it can aid in modeling timing patterns and in aligning synthetic speech with real interactions. The concept is used in varying degrees and does not correspond to a single standardized annotation scheme across corpora.

See also: timing in discourse, turn-taking, timestamping, discourse annotation, speech transcripts.