Transcrire l'oral — Transcribing Speech

Spoken French on Paper

Spoken and written French differ more than learners often expect, and transcribing a recording makes the differences impossible to ignore. In everyday speech the ne of negation is dropped most of the time, subjects and objects are detached and picked up again with pronouns (moi, je …; ce film, je l'ai vu), on very often replaces nous, ça replaces cela, and the speech is full of discourse markers such as ben, bon, enfin, tu vois, du coup and quoi. None of this is a sign of carelessness: these features are found in every social group and are part of the grammar of the spoken language. Researchers who work on speech transcribe interviews according to explicit conventions, so that the transcript stays faithful to what was said and can be checked by others. Conventions vary from one research team to another, for instance in how pauses, overlaps, laughter or inaudible passages are marked, and a student should follow the guide used in their own lab. Automatic speech recognition has improved a great deal, and some researchers use it for a first draft, but it tends to tidy speech into written norms and to struggle with accents, overlaps and informal speech, so the result has to be checked carefully against the recording. Transcription is slow: several hours of work for one hour of recording is common.