Introduction: Why transcribe?

Why transcribe?

Transcription is a form of documentation which creates a word-for-word text version of what was said. A commonly-used approach is intelligent verbatim, where redundant words, repetitions, misspeaking, and “ums and ahs” are all removed, but no summarising or other intervention is done. Other types of documentation are also based on the words spoken, but are different because the documenter does intervene to summarise or shorten what was said.

Intelligent verbatim transcripts serve several purposes. They can be used as a basis for translation or subtitling; they’re also a valuable archival adjunct to a primary source (such as video) whose format may eventually become obsolete. They are also useful as an accessibility tool - people can find it helpful to watch a video alongside an intelligent-verbatim transcript.

Transcription is also useful for all the reasons that any kind of documentation is useful: for accountability, transparency, making material more discoverable /searchable /indexable, keeping track of decisions, making it easier to share content, and avoiding the need for people to sit and watch entire 2-hour recordings.

Here is an overview, originally produced for Governance Guild, of what documentation is and why it’s important.

...and why not?

Intelligent-verbatim transcription is labour intensive and time-consuming. AI-generated transcripts are not yet good enough to be left uncorrected; and often (especially in our global communities, with many accents and dialects) their output is so garbled that correcting them to a standard that is useful takes almost as long as transcribing from scratch. In particular, AI tools have noticeable cultural biases - they work reasonably well for middle-class US male voices, but struggle with non-US accents, working-class accents, and women’s voices. So if an AI transcript is going “on the record”, and is official and archival in some way - particularly for a meeting in which governance decisions are being made - it needs correcting. A very high level of skill in the relevant language is needed to do this - in this project, we tried giving the task to upper-intermediate level English speakers, who worked using translation tools, but the results were not effective.

For the more ephemeral meetings we hold in Catalyst, intelligent-verbatim transcripts are an unnecessarily heavy overhead, and other approaches are better - for example, summaries, timestamping, layered notes, etc. But for these, the transcriber needs a very good understanding of the topic of the meeting. In fact, because summarising involves more intervention from the documenter, more trust in the documenter’s skill is required.