jefferson transcripts, produced from the audio.

what jefferson notation is

the jefferson transcription system, developed by Gail Jefferson for conversation analysis, marks features that grammatical transcripts miss: pause length to the tenth of a second, overlapping talk with brackets, in-breath and out-breath, stress, lengthening, and pitch movement. it's the standard for CA, sociolinguistics, and most discourse-analysis work that takes the talk seriously.

producing it by hand takes roughly an hour per minute of audio for a trained transcriber. tools that exist today (DOTE, transana, plain text editors with jeffersonian macros) help with the symbols but don't time the pauses or detect the overlaps. the analyst still spends hours with a stopwatch and a waveform.

what we ship

a transcription pass that emits jefferson-style marks automatically:

• timed pauses in parentheses, to the tenth of a second: (0.4), (1.2), (.) for micro-pauses.
• overlap brackets [ and ] placed at the start and end of overlapping talk between speakers — derived from word-level timestamps with sub-200ms resolution.
• in-breath and out-breath .h and h. from voice-activity detection inside silence regions.
• latching = when one turn begins immediately at the end of another with no measurable gap.
• stress, lengthening, pitch shift from the model's prosody embedding. these are flagged as suggestions — final calls stay with the analyst.

workflow

1. drop the audio file. transcription runs and produces a first-pass jefferson transcript.
2. review in the browser editor. pause marks are clickable — click (0.7) to play the silence; the editor scrubs to the timestamp.
3. adjust where the model is wrong. the editor knows the jefferson symbols and tab-completes them.
4. export. UTF-8 plain text (the standard CA-publication format), jefferson-PDF for handout circulation, or our own JSON for programmatic re-analysis.

what we don't claim

jefferson transcription is interpretive work. the analyst's ear is the source of truth for prosody calls, for what counts as overlap, for whether a stretched syllable is doing work in the interaction. our tool gets you to a first draft faster. the analyst still does the analysis.

the model is best on english. for other languages we pass the audio through the multilingual cloud model and the jefferson marks come from word-level timestamps and VAD only — symbol tab-completion still helps but the prosody suggestions are english-only at launch.

citation

for conventions and the citation we use to anchor this work: Jefferson, G. (2004). Glossary of transcript symbols with an introduction. In G. H. Lerner (Ed.), Conversation analysis: Studies from the first generation (pp. 13–31). John Benjamins.