This article is part of the network’s archive of useful research information. This article is closed to new comments due to inactivity. We welcome new content which can be done by submitting an article for review or take part in discussions in an open topic or submit a blog post to take your discussions online.
The woes of transcription and translation are well-known amongst qualitative researchers, particularly those working in multilingual settings or in international teams. With tempting professional transcription and translation services on the rise, it can be easy to offload the tedious work of transforming hours of recorded data into a manageable word document written in a common language for all team members to work with. But for anyone who has been involved in such a process it is obvious that even doing this is far from straight forward; that maximising the researchers’ involvement in transcription and translation is imperative. So what have we learned through actually going through this process? What are the hidden challenges behind the seemingly straightforward instruction to “transcribe and translate” one’s data?
A) TRANSCRIBING: what goes into a transcription?
Transcribing is converting data from an audio or video format into written document format. It involves a native speaker listening to the recorded data and representing it in written form. Doing the transcription can be an invaluable way for the transcriber to really get to know the data. However, recognition that it can take 6-9 hours transcribing 1 hour of audio is a yawn-inducing prospect as well as difficult to fit into other fieldwork plans, even for the most enthusiastic researcher and experienced typist. Here we share some considerations in planning your transcriptions.
Be clear about what kind of transcription you want and who should do it. It is essential when considering transcription to first be clear on what type of transcription is needed for a given piece of work, and why. Generally, there are three types of "transcription."
- Verbatim: Each word is transcribed into text including mumbles such as "uh" or "hum" in conversations needs to be recorded as it is spoken by the speaker in the audio.
- Intelligent: A voice recording is converted into text excluding pauses unnecessary for context or meaningless nods. The transcriber needs not pay attention to those pause which sounds like- " hmmm, know, Got it, you know, ahaan" etc…
- Edited: In addition to the intelligent transcription work, the transcriptionist alters existing sentences into the sentences that make sense.
We suggest that all qualitative researchers should have had the opportunity to experience for themselves what it feels like to do a transcription. Once they have this experience, if they cannot do it all themselves, they can think about who should help with the transcription. Ideally, the person who does the transcription was present at the interview, or is at least part of the research team. This helps ensure the transcriber has a good sense of the study objectives and is able to picture the discussion context. But in many of our studies interviews are conducted in local dialects: native speakers in the team are not experienced or confident transcribers, potentially slowing the process even further. Transcription may have to be done by others not so familiar with the interview or study, or even an external professional company.
Our transcription SOP: agreeing before we start on what a transcription should look like. Whether the study transcriptions are being done by the study team, or by an external transcriber, we have found it really useful to first agree on a transcription and translation Standard Operating Procedure (SOP), including:
- First agreeing on the nature of the transcription needed. Often at our programme we use verbatim or intelligent transcriptions, but even then there needs to be an agreement of exactly what a transcription should look like – when to start a new line, how pauses of different length and laughter are handled, etc.
- Deciding on the language. Usually at KEMRI-Wellcome Trust we aim to generate the transcript in the language in which the interview was conducted (e.g. English to English; Kiswahili to Kiswahili), as heard on the tape, without being simplified or condensed in anyway. If there is a need to translate, we usually do this after the transcript has been produced in the original language to maximize opportunities for cross-checking the English.
- Agreeing on the meta data and additional information to include. It needs to be agreed who will add information at the start and end of the transcript, and what information. For example at the top of the transcription it’s common to include: the name/code/pseudonym of participant(s); characteristics of participant(s); date and place of interview; name of interviewer and note taker (if present). At the end of the transcript any notes or observations made by the study team should be included, for example on the quality of interview, any interesting non-verbal communication, or indicating any important contextual information.
Start early, checking initial transcriptions and discuss any issues needing correction. It is tempting to send off a batch of interviews for transcription and wait for them to come back perfect, as we get on with other pressing fieldwork activities. But we’ve found that after the first few transcriptions it can be really useful to have a feedback session between researchers and transcribers to discuss the transcription and the SOP. Tweaks may be needed on both ends.
And then, even when you’ve got the perfectly agreed process for transcriptions, strange issues and problems can still creep in. For example mis-heard words which confuse the meaning (eg ‘accept’ instead of ‘except’), or words with double meanings depending on spelling (eg ‘weight’ or ‘wait’) where there wrong one has been selected in context or even missing hunks of data. So the research team must read through each transcription in the end to ensure that errors have not been introduced, and consistency has been maintained. Such quality controls are paramount.
b) TRANSLATION: and now for the really tricky part.
Translation is the process of creating a written document from a transcript or a digital/video recording from one language to another language. Where all of the analysis in a project will be conducted by those who are fluent in the language in which the data was originally collected, a translation may not be needed at all. However in our experience it is common to want to share at least a proportion of data with others who will need to work from transcriptions; either as part of collaborative studies or to support trustworthiness checks in analysis processes. Good quality translations will often be absolutely essential but can be surprisingly difficult and time consuming. Here we share some of our lessons in translation from a study aimed at supporting young mothers to re-introduce exclusive breast-feeding following a child’s hospitalisation which involved 20 in-depth interviews with women immediately after hospital discharge. We were keen to understand their experiences with a breast-feeding peer supporter intervention that was being piloted at the hospital.
Who should do the translations? Our research team included a Kenyan post-doctoral nutritional epidemiologist from the local community, two earlier career Kenyan social scientist from elsewhere in Kenya, and a senior British anthropologist. For the team to analyse the interviews conducted in a mixture of Kiswahili and the local dialect Kigiriama, excellent translations of the transcriptions were essential. We know that the most important aspect of translation is maintaining the integrity (meaning and significance) of the audio, video or document being translated. The translation therefore had to be done by somebody who has a complete understanding of both the original language source and the target language source, as well as – ideally - the interview context. Although somebody who is very well trained academically in one of the languages (such as in our case Kiswahili or the local dialect Kigiriama) may offer some advantages in terms of understanding some more obscure words or how those languages are structured, we realised they may not have such a good understanding of local uses of the language, and the importance of keeping local meanings in transcriptions. On the other hand somebody who the local dialect fluently might not be as confident in the language that the transcription is needed in, which can introduce different challenges into the timing and quality of the translation. We opted to conduct the transcription process internally, followed by rigorous quality checks by various team members.
Towards accurate rather than direct translation: We learned that there should be accurate translation as opposed to direct translation because direct translation often does not make sense in English. Meaning should not be guessed, and where there is doubt it should be left in the local language. An example of one direct translation including a mis-understood word was as follows: “I don’t know whether the wait of my child was enough’. In fact following the checking process this was changed to read: “I’m not sure whether my child’s weight was adequate (for his wellbeing)”
Resist the temptation to tidy up people’s comments. There should be no effort to ‘clean’ or ‘clarify’ what people are saying. If it’s broken Kiswahili, it’s broken English. Otherwise the meaning and the possibility of interpretation is lost. For example we had an interviewee say the following: ‘‘The day the child stopped defecating is the same day he started vomiting”. Although this could be tidied up to read “He stopped defecating and vomiting on the same day”. The participants’ way of expressing themselves is changed, and it can be difficult to know where translators will begin and end with their tidying!
Simple punctuation can change a meaning. For example in our context Eeeh has two meanings – either yes, or what. What do we make of this respondent’s response to the moderator asking ‘‘The peer supporters told you to feed well so as to be able to produce sufficient breast milk? R: Eeeh I should feed well to satisfaction so as I would be able to produce enough breast milk.’’ Some indication of the tone would help us understand if what’s being said is an agreement or a querie about whether the question has been adequately heard.
As with transcription above, the value of an SOP for translation and checking the process early. Balancing all of the above is tricky. Translation involves meaning, not translation word for word. At the same time, you need to strike a balance between conveying meaning and preserving the messiness of the conversation as it was spoken. Carefully chosing who does the translation and developing and agreeing an SOP could be really helpful. But also check the process early - the SOP may seem to be an obvious guide to follow; however, inevitable challenges will arise during the translation process. The sooner these are spotted and resolve the better.