Monday, 1 March 2021

Transcripts and translations

I have a number of students who come to me suggesting that they will ask participants questions in one language, then simply transcribe the data, then translate it... as if this piece of work will be no problem. We can be completely naive about the volume of work we are - unwittingly - proposing that we take on!

For a short Master's project, we need to budget for roughly an hour's worth of interview per interviewee, to have 8-10 interviewees, thus collecting at least between 8-10 hours of data. While more data is good, 8-10 hours should give us the findings complexity and depth required for a short Masters project at 30 credits. 

A single hour of interview recordings will contain roughly 5,000 to 10,000 words. If we use a transcription service, it will take an experienced transcriber between 4-10 hours to transcribe each hour of interview (IndianScribes, 2018). We need to brief the transcriber. We need to provide a key for what level of transcription we need (see University of California Irvine, 2014). We need to provide a sample format (e.g. here) so the transcriber knows what we want to get back. Once the transcripts come back, we must review and correct them (see Choi et al., 2012). 

It has been documented that inexperienced data transcription has a factor of around 60 to 1 for a "plain" transcription (McCulloch, 2019). A full transcript, which includes tone, phrasing, interrupters etc, may take longer (see Saldana, 2009). We need to explain what our transcription process was, and why. An estimate of times are as follows:

  • If no tools, apps or shortcuts are used (see here for some ideas), this is likely to take between 480-600 hours to transcribe 8-10 hours of recordings (McCulloch, 2019)
  • If tools such as otter.ai are used, there can be a saving of a factor of 6: so 10 times, or 80-100 hours (my own method can be read about here, and Pogue, 2017)
  • An experienced transcriber will take so 32-100 hours for 8-10 hours of data, not including researcher time in checking and rechecking the work (IndiaScribes, 2018). However, we also need to be aware that "Accented speech is often charged at a higher fee to standard interview transcription and this needs to be acknowledged in research budgeting" (Fryer, 2019, p. 1670), so we are better to assume our transcription to be on the high side

As we can see by the hours listed above, if we are doing a short course and are under time pressure, it is worth paying for professional transcription. As it is, we are still going to have to create the brief, decide on the key, and consider the type of transcription we desire. Fryer has a great chapter which details the process very well (2019). We also need to carefully review the transcriptions to check the accuracy, validity, the time stamps, and the formatting. Although we will have been listening to our sound files, it is only once our transcripts are done that we can really start our data analysis.

However, we have not yet considered translation. When we ask questions in one language, then have to not only transcribe, but to translate so we can collectively analyse all findings. If we are planning on transcribing AND translating, this adds considerably to our workload.

Calculating hours for this task is difficult, as there are so many variables. It depends how literate we are in both languages: how quickly we can work across both to accurately capture meaning. It depends on how closely related both languages are: colloquialisms in French can be remarkably similar to English, while colloquialisms in Mandarin are extremely difficult to meaningfully translate into English. But I would estimate that each interview will take as long to translate as to transcribe. 

To wrap up, we must ensure that our documentation is correct. Our research ethics application encompasses any use of transcription or translation services. In our methods, as well as noting how long each interview is and how many interviews we complete, we must also note how long interview transcriptions took, describe our method for creating accurate translations, and how we ensured that both transcription and translation were accurate (for some stellar insight into these processes, see Choi et al., 2012; Fryer, 2019; Regmi et al., 2010). 

Due to time pressure on short projects, it is easy to see how transcription and translation can become a significant element of the research project, and is often a choke-point for completion.

If we are supervising, we need to have a realistic view of just how much effort and time will be required; only then will we be able to clearly support our students so they make good quality decisions.


Sam

References:

No comments :

Post a Comment

Thanks for your feedback. The elves will post it shortly.