Transcription

There are always mistakes in transcripts – here’s how to avoid them

01. August 2024 3 minutes reading time

There are always mistakes in transcripts – here’s how to avoid them

A great article by Isabella Chiari has been accompanying us in our workshops for a long time. Chiari specifically and systematically measured how many and which errors occur during the typing of speech.

To this end, she had the participants in the study listen to and write down 22 short passages from messages and spontaneous dialogs. The passages were between 1.5 and 13 seconds long. Chiari then drew out the red pencil, searched for all the errors and systematized them.

Alarming results

The frightening thing: An average of 1.6 errors were produced per passage! That’s really a lot. Each passage written down therefore contained an average of one error. And this was roughly equally distributed both in the controlled news texts and in the spontaneous dialogs.

Chiari also examined the extent to which the transcription errors change the meaning of the statement. If I type “an apple” instead of “an apple” in the transcript, it is not significant. But the difference between “one” and “none” is very clear. And in fact, 64% of the errors distorted the meaning of the statement. (Similar results can also be found with Daniela Oppermann, Susanne Burger and Karl Weilhammer)

That’s tough stuff to begin with. It can therefore be assumed that every second passage in a transcript without a correction run contains a semantic error!

This is how you can explain it – empathy and thinking for yourself

The explanation for the many mistakes is a conciliatory one. Chiari concludes that most errors are not due to poor hearing or inattention on the part of the transcriber.

On the contrary, the erroneous passages are usually semantically very similar to what is actually spoken. For example, minor grammatical errors made by the speaker are corrected or nonsensical statements are written down as they were understood. Chiari concludes that empathic thinking and compassion on the part of the transcribers must therefore play an important role here.

This little experiment shows wonderfully that transcribing is not a blunt process, but that it is an active (re)construction. The errors make it visible, so to speak, that transcribers actively think and empathize during transcription. This – let’s call it pathetically “human” – ability to disregard transcription and really only write what you hear requires a very high level of concentration and self-discipline. This ultimately makes typing extremely tiring.

Conclusion … and what we can learn from it

It is a platitude, and yet – as we have seen – it is intrinsically important, especially in transcriptions: making mistakes is human.

And if you know about it now, you can arm yourself and take action. Therefore, even if it seems annoying, you should always plan a correction run. We recommend that this should ideally be carried out by a second person. If this is not possible, you should allow some time to pass between the first pass and the correction (simply to take your mind off things).

In addition, a little distrust of the typed word is quite appropriate during the analysis. For important passages, it is advisable to listen directly to the original audio recording again. If you have set time markers when typing, you can also do this later directly in the evaluation software.

The latest posts

4 minutes reading time
Documentary method – Proposal for the use of QDA software.

A simplified example, implemented with f4analyse. Introduction Here we show an example of the technical implementation of the evaluation steps …

3 minutes reading time
Automatically convert Zoom meetings to text

Zoom is a wonderfully practical tool for online meetings. And since the meeting is held on the computer, it can …

4 minutes reading time
Qualitative interviews – free checklist for data protection

How to conduct your interviews in compliance with the GDPR Qualitative interviews usually contain personal data, and these need to …

Back to overview