Preparing Audio Data for Analysis: Transcription PDF

 7 Preparing audio data for analysis: transcription OVERVIEW Ort...

 7 Preparing audio data for analysis: transcription OVERVIEW Orthographic transcription and the messiness of language use Understanding what a transcript is, and what it is not What makes a (quality) transcript? Producing the transcript Giving yourself enough time to transcribe With your data collected (hoorah!), you’re sitting at the edge of analysis. But before you jump in and get started you may need to prepare your data ready for analysis. This chap- ter outlines and discusses the main way audio (and audio-visual) data are made ready for analysis – through transcription. Transcription is an important part of qualitative research using audio data and, at first sight, it might seem like a straightforward process: you play a recording in very short bursts and type up what you hear. Indeed, transcription is often (implicitly) treated as a ‘minor or merely technical concern’ (Nelson, 1996: 12) and given little, if any, coverage in qualitative textbooks. But it’s not just a technical concern. When transcribing, we have to choose how – and what – we translate from speech and sounds into written text, making transcription a theoretically influenced practice (Ochs, 1979). And you soon discover that transcription’s not simple when you try to do it. Before reading any further, we recommend trying the first transcription exercise at the end of this chapter; our discussion of transcription will be much more meaningful if you’ve already had a go. ORTHOGRAPHIC TRANSCRIPTION AND THE MESSINESS OF LANGUAGE USE There are many different styles of transcription, which suit different analytic methods. We outline a style of audio transcription often called orthographic or verbatim. This 07-Braun & Clarke_Ch-07-Section 3.indd 161 28/02/2013 7:40:27 PM  162 Successfully analysing qualitative data style, which focuses on transcribing spoken words (and other sounds) in recorded data, can be contrasted with audio transcription styles that include more phonetic or paralin- guistic features, where the transcript aims to record not only what was said, but also how it was said (that form of transcript – Jeffersonian – is used in discursive psychology [DP] and conversation analysis [CA]; see Boxes 8.1 and 8.2 in Chapter 8), or visual elements (e.g. Heath & Hindmarsh, 2002; Norris, S., 2002; Peräkylä, 2002, 2006). Orthographic transcription records what was said. Even doing this form of transcrip- tion isn’t simple, as spoken language and written language are very different. When we speak, we don’t use punctuation to make ourselves understood. We use pauses and into- nation; we vary our speech in pace (faster, slower), volume (louder, quieter) and many other ways. Spoken (natural) language is ‘messier’ than written language: we hesitate when we speak, we stumble over our words, start a word or phrase and don’t finish it, and say the same word or phrase a number of times. One of the most startling research experiences we’ve had was hearing how we actually speak, how inarticulate we actu- ally are, in real life; transcribing your first interview can be a horrifying yet eye-opening experience! The orthographic transcripts from the weight and obesity and body art focus groups (FGs) on the companion website show how messy real speech is. UNDERSTANDING WHAT A TRANSCRIPT IS, AND WHAT IT IS NOT Most qualitative analysis uses transcripts, rather than the original audio(visual) record- ing, so it’s important that your transcripts are thorough and of high quality. We avoid using the term ‘accurate’ in relation to transcripts because there is considerable debate among qualitative researchers about what constitutes an accurate transcript and whether such a thing is possible (see Potter, 1996; Sandelowski, 1994a). You can avoid these debates, and still produce a good transcript, or rather, a ‘good enough’ transcript, as it’s important to know when to stop. You need to be aware that a transcript of audio(visual) data is not a facsimile; it’s a representation. Just as an audio recording of an interview is not the same as the actual interview experience, a transcript of an audio recording is not the same as the audio recording, making a transcript two-steps removed from the actual interview experience. With each step, information is lost or changed in some way. Rather than seeing a transcript as raw data, it can be seen as ‘partially cooked’ (Sandelowski, 1994a: 312) data, already prepared and slightly altered from its original stage. So far from being ‘a neutral, simple rendition of words’ from audio to written form (Potter, 1996: 136), a transcript is a ‘selective arrangement’ (Sandelowski, 1994a: 311), produced for the purposes of analysis. The transcript is the product of an interac- tion between the recording and the transcriber, who listens to the recording, and makes choices about what to preserve, and how to represent what they hear. A transcription notation system allows you to clearly and consistently translate spo- ken language into written language, meaning your approach to transcription is thor- ough and meticulous. With no definitive notation system for orthographic transcription 07-Braun & Clarke_Ch-07-Section 3.indd 162 28/02/2013 7:40:28 PM  Preparing audio data for analysis: transcription 163 (unlike CA transcription, e.g. Atkinson & Heritage, 1984; Jefferson, 2004), qualitative researchers most often construct their own notation systems (we provide ours, with instructions, in Table 7.1; we show a section of annotated transcript in Box 7.1, explain- ing the features in situ). Idiosyncrasy can create confusion if authors don’t include a notation key so the reader can decode their transcription system. For example, research- ers often use three full-stops ‘…’ in transcripts. Some common uses of ‘…’ are to signal a pause, hesitation, or trailing off, or that a section of data has been deleted. The reader needs to know which it is. If you’re not doing your own transcribing, you need to make sure the person doing it follows your transcription notation system, as well as all other instructions (including confidentiality; see the transcriber confidentiality agreement on the companion website). WHAT MAKES A (QUALITY) TRANSCRIPT? Vitally, a transcript needs to both signal what is said and who is speaking. A good orthographic transcript records in written form all verbal utterances from all speakers, both actual words and non-semantic sounds – such as ‘erm’, ‘er’, ‘uhuh’, ‘mm’ and ‘mm-hm’. Your aim is to create as clear and complete a rendering of what was uttered as possible. Nothing should be ‘corrected’ or changed – for example, don’t translate slang or vernacular terms into ‘standard’ English (if a participant says ‘dunno,’ it should not be transcribed as ‘don’t know’). If you ‘clean up’ or edit your data, your participants will sound more fluent and more like they are using written language (DeVault, 1990), but the whole point of collecting spoken data is that we capture how people express themselves. Some researchers signal some more significant paralin- guistic features of the data (e.g. laughter, crying, long pauses, or strong emphasis – see Table 7.1). Very simple errors or mishearings in transcription can radically change the mean- ing of data. For instance, in the weight and obesity FG, the line ‘Oh god it’s home ec [home economics] again’ (Line 887; see the companion website) was originally tran- scribed by a professional transcriber as ‘Oh god it’s homework again’. In the context of a discussion about schools teaching students how to cook, the meaning is quite different (home economics traditionally teaches skills relevant to domestic life, such as cooking). Canadian public health researcher Blake Poland (2002) identified four common types of transcription error: 1 Sentence structure errors: as noted, people don’t talk in sentences – indeed, the concept of a ‘sentence’ doesn’t translate well into spoken language – yet some peo- ple use punctuation in a transcript, as if it were written language. But in so doing, they make decisions about punctuation use, such as where to begin and end sen- tences, that can alter the interpretation of the text. For example, ‘I hate it, you know. I do’ carries a different meaning from ‘I hate it. You know I do’. For this reason, we tend to include little or no punctuation in our orthographic transcripts. If you want to add punctuation to extracts of data included in written reports and presentations, to aid ‘readability’, always check the audio, so you can get a sense 07-Braun & Clarke_Ch-07-Section 3.indd 163 28/02/2013 7:40:28 PM  164 Successfully analysing qualitative data from intonation and language use where ‘sentences’ stop and start, and where the pauses really are. 2 Quotation mark errors: these errors fail to capture a feature of talk called reported speech. Reported speech is where a person reports what someone else said (or thought), or indeed what they themselves said at another point in time. Our exam- ple of ‘oh God, it’s home ec again’ is reported speech, because Anna is reporting on what her parents used to think or say: Anna: My parents used to dread that day ?: Yeah ((General laughter)) Anna: ‘Oh god it’s home ec again’ Whether or not someone uses reported speech is meaningful (e.g. Buttny, 1997), especially for some forms of qualitative analysis such as conversation analysis (CA; see Box 8.2 in Chapter 8) and discursive psychology (DP; see Box 8.1 in Chapter 8); in transcripts, it should be signalled with inverted commas or quota- tion marks. 3 Omission errors: these errors are ones where words (or vocalised sounds) are not included. Sometimes they can be inconsequential; at other times, crucial. Poland gives an example from his research on smoking cessation of a transcriber who missed the vital word lung from ‘I lost a very close friend to [lung] cancer’, an obviously a significant omission given the focus of the research. 4 Mistaken word/phrase errors: these errors are ones where an incorrect word or phrase is used – our homework/home ec error is a good example of this. Although these words sound somewhat similar, their meaning is different, and in our data, important analytically. Omissions and mistaken words can occur because you’re tired (transcription requires intense concentration and focus; it’s important to break at regular intervals), or because you’ve left too long between collecting the data and transcribing it. As noted in Chapter 4, if collecting interview (or FG) data, we strongly recommend scheduling time to transcribe it as soon as possible (ideally the following day). A surprising amount of detail from an interview or FG remains clear for a few days; this memory rapidly fades. The quality of the recording and the nature of participants’ speech (volume and speed, accents and overlap) can also lead to errors (see below); if you can’t understand what was said, your transcript might not only contain errors; it will also contain lots of blanks, making it far less viable as data. In the actual transcript itself, each speaker needs to be identified by a name (a pseu- donym rather than their real name) or role (e.g. interviewer/moderator); each time a new speaker says something (technically called a turn of talk), it’s presented on a new line. Using a hanging indent style, with each speaker’s name or role followed by a colon and a tab, before what they said, makes the transcript visually clear and easy to analyse (see Box 7.1). 07-Braun & Clarke_Ch-07-Section 3.indd 164 28/02/2013 7:40:28 PM  Preparing audio data for analysis: transcription 165 Table 7.1 O  ur transcription notation system for orthographic transcription (adapted from Jefferson, 2004) Feature Notation and explanation of use The identity of the The speaker’s name, followed by a colon (e.g. Anna: ) signals the identity speaker; turn-taking of a speaker (use Moderator/Mod: or Interviewer/Int: for when the in talk moderator/interviewer is speaking; or the moderator/interviewer’s first name); start a new line every time a new speaker enters the conversation, and start the first word of each new turn of talk with a capital letter Laughing, coughing, ((laughs)) and ((coughs)) signals a speaker laughing or coughing during etc. a turn of talk; ((General laughter)) signals multiple speakers laughing at once and should be appear on a separate line (to signal that no one speaker ‘owns’ the laughter) Pausing ((pause)) signals a significant pause (i.e. a few seconds or more; precise timing of pauses is not necessary); can also use (.) to signal a short pause (a second or less) or ((long pause)) to signal a much longer pause Spoken abbreviations If someone speaks an abbreviation, then use that abbreviation (e.g. TV for television; WHO for World Health Organization), but do not abbreviate unless a speaker does so Overlapping speech Type ((in overlap)) before the start of the overlapping speech Inaudible speech Use ((inaudible)) for speech and sounds that are completely inaudible; when you can hear something but you’re not sure if it’s correct, use single parentheses to signal your best guess or guesses as to what was said – for example (ways of life) or (ways of life/married wife) Uncertainty about who Use ? to signal uncertainty about the speaker – just ? for total uncertainty, is speaking F? or M? if you can identify sex of the speaker, or a name followed by a question mark (e.g. Judy?) if you think you might know who it is Non-verbal utterances Render phonetically and consistently common non-verbal sounds uttered by your participants. For English-as-a-first-language speakers, these include ‘erm’, ‘er’, ‘mm’, ‘mm-hm’, but note that how these are written is context-dependent. In Aotearoa/New Zealand, the first two would be written ‘um’ and ‘ah’ Spoken numbers Spell out all numbers (and be mindful of the difference between ‘a hundred’ and ‘one hundred’) Use of punctuation It is common to use punctuation to signal some features of spoken language (such as using a question mark to signal the rising intonation of a question or a comma to signal a slight pause but with the intonation of continuing speech). However, adding punctuation to a transcript is not straightforward and it is important to be mindful of the ways in which adding punctuation can change the meaning of an extract of data. Equally, punctuation enhances the readability of spoken data, especially extracts quoted in written reports (see Box 11.5 in Chapter 11) (Continued) 07-Braun & Clarke_Ch-07-Section 3.indd 165 28/02/2013 7:40:28 PM  166 Successfully analysing qualitative data Table 7.1 (Continued) Feature Notation and explanation of use Cut-off speech and This level of detail is not necessary for most experiential forms of speech-sounds analysis, although it can be useful to signal moments when participants are struggling to articulate their thoughts, feelings etc.; to signal cut-off speech, type out the sounds you can hear, then add a dash (e.g. wa-, wor-, worl-); try to capture this at the level of phonetic sound Emphasis on particular Again, this level of detail is not necessary for most experiential forms of words analysis, although it can be useful to indicate words or sounds that are particularly emphasised by underlining (e.g. word) Reported speech Reported speech is when a person provides an apparent verbatim account of the speech (or thoughts) of another person (or reports their own speech in the past). Signal this with the use of inverted commas around the reported speech (e.g. … and she said ‘I think your bum does look big in that dress’ and I said ‘thanks a bunch’…) Accents and It’s important not to transform participants’ speech into ‘standard’ abbreviations/ English; however, fully representing a strong regional accent can be a vernacular usage/ complex and time consuming process. A good compromise is to signal mispronunciation only the very obvious or common (and easy to translate into written text) abbreviations and vernacular usage, such as ‘cos’ instead of ‘because’ or a Welsh speaker saying ‘me Mam’ (instead of the English ‘my Mum’), unless it is absolutely critical for your analysis to fully represent exactly how a speaker pronounces words and sounds. Don’t ‘correct’ mispronunciation or misspeaking of works, such as ‘compostle’ instead of ‘compostable’ Names of media (e.g. Should be presented in italics (e.g. The Wire, Men’s Health) television programmes, books, magazines) Identifying information You can change identifying information such as people’s names and occupations, places, events, etc. in one of two ways (see also Box 7.2): By changing details and providing unmarked, appropriate alternatives (e.g. ‘Bristol’ to ‘Manchester’; ‘my sister is 14’ to ‘my sister is 12’; ‘I’m a really keen knitter’ to ‘I’m a really keen sewer’) By replacing specific information with marked generic descriptions (indicated by square brackets, so ‘London’ might be replaced with [large city]; ‘Michael’ with [oldest brother]; ‘running’ with [form of exercise]) PRODUCING THE TRANSCRIPT Transcription is often thought of as a chore, and boring; but although it can be hard to do, it’s a really good skill to develop and essential for any qualitative researcher work- ing with audio data. We’re going to talk about transcribing digital data, as most data are digital these days (previously, most were recorded on audio tape). To transcribe digital data, the minimum you need is computer software to play it. Transcription soft- ware (rather than Media Player or iTunes) has features that allow you to speed up or slow down the playback pace. This can be really useful for deciphering what is being 07-Braun & Clarke_Ch-07-Section 3.indd 166 28/02/2013 7:40:28 PM  Preparing audio data for analysis: transcription 167 BOX 7.1 ANNOTATED EXAMPLE OF ORTHOGRAPHIC TRANSCRIPTION Start the first word of a new turn of talk with a capital letter. You can use the interviewer/ moderato’s name, but it is often You can use this symbol to clearer to use their role (often indicate very short pauses if you abbreviated to ‘Int’ or ‘Mod’). are doing pattern-based DA. Moderator: Okay so (.) I want to focus on (.) obesity Underlining rates (.) within individuals (.) so why do used to Give each new speaker/new you think people become fat or obese indicate emphasis. turn of talk a Sally: I think there are a number of reasons erm new line. I think one o-erm one of the main reasons I became obese was because ((pause)) erm UK English I had to go through a number of various speakers tend Cut-off to say ‘erm’ orthopaedic surgeries which actually speech or whereas people meant that I was in a wheelchair for quite a speaking US or sounds. few months at a time and unfortunately (.) New Zealand part of when you’re stuck in a wheelchair English tend to you suddenly feel like a little bit depressed say ‘um’. as well so (well) you tend to have the Best guess. knock-on effect of eating so you’re eating A longer more than you should do cos really when pause. you’re very immobile you shouldn’t eat very much at all because hence you do gain weight and you get there and that’s one of the reasons a lot of people erm who were generally quite fit people who I’ve spoken to have gained weight through ((pause)) mainly things like surgery and various life impacts like that life events Moderator: What does everyone else think Rebecca: I think that I think you’re right about life events even if you know nothing to do with surgery I think Sally?: Oh A question mark after a name signals your best guess as to who is speaking – unless you know participants’ voices very well, it can be hard to identify the speaker of very short bursts of speech and sound, especially when there are multiple speakers as in a focus group. 07-Braun & Clarke_Ch-07-Section 3.indd 167 28/02/2013 7:40:28 PM  168 Successfully analysing qualitative data said. The basic software Express Scribe, free to download, is more than adequate. More advanced programmes can include features like tone adjustment, and an auto- reverse option. Other useful equipment includes: decent quality headphones – vital if you’re transcribing on a public computer, for confidentiality reasons, but important for potentially increasing clarity and volume of the playback; a specially designed transcription foot pedal, which allows you to control play, pause, rewind and fast forward with your foot – this can speed up transcription considerably, as you don’t have to use your hands to press play, pause and rewind, and they can stay dedicated to typing the transcript; you also don’t need to switch between different active windows on your computer to play/pause the recording, and type up the transcript. To actually transcribe the data, you play a very short segment of the recording (a few seconds) and type what you hear, using your notation system to guide you. You need to rewind the recording a little, to avoid missing anything, then play another short segment, transcribe it, etc. It’s not this straightforward though. You may need to play each snippet of the recording a number of times to make out what has been said (slow- ing down or speeding up the recording can help here), and you should always go back and double check what you’ve transcribed. We have a natural tendency to ‘correct’ what we hear, and this creeps into transcription all over the place. One of the key bits of advice we give our students who are learning to transcribe is to try not to listen to the meaning of the words, just to the sounds of the words. But it is tricky to learn this skill. There is no right way to manage the process of producing a transcript. When and how frequently you go back and check the ‘accuracy’ of your transcribing is up to you. You’ll soon learn how many errors you are making, and how much checking you need to do. Even if you thoroughly check your transcribing as you go, we recommend checking (and if necessary editing) each transcript in full at least once, after it has been completed (it’s very easy to miss tiny details in the gaps between stopping and starting the recording). Leaving a bit of time between producing the transcript and checking it can be helpful, particularly for hearing things that you couldn’t make out the first (or tenth!) time you listened to the recording. If you’re missing something you think is important, you can also get your supervisor or a co-researcher to listen; it’s amazing how another person can instantly and clearly hear something you think is completely incomprehensible. Once you’re done, it’s generally a good idea to password-protect or encrypt transcript files, even if anonymised, for confidentiality reasons. Just don’t forget the password! The main aim of doing an orthographic transcription is to produce a thorough record of the words spoken. Non-semantic sounds, hesitation, repetition, false-starts, pauses, laughter and so on are less important (they can be really time-consuming to transcribe). How much of such detail you need relates to your analytic, methodological and theoretical positions. An orthographic transcript is going to provide enough infor- mation to analyse data using the methods we discuss in depth in this book. Experiential 07-Braun & Clarke_Ch-07-Section 3.indd 168 28/02/2013 7:40:28 PM  Preparing audio data for analysis: transcription 169 analytic methods, like interpretative phenomenological analysis, and some versions of thematic analysis (TA) and grounded theory (GT), focus on the words spoken by the participants – what was said rather than how it was said. At most, the analysis might discuss moments when a participant cried, struggled to articulate themselves or was hes- itant. Critical analytic methods, like constructionist TA and GT, and pattern-based dis- course approaches, are also interested in how things are said. However, a very thorough orthographic transcription, such as the one we’ve outlined, generally captures enough detail for these forms of analysis. Finally, another exception to the rule of ‘recording everything that was said’ relates to anonymising data as you transcribe it (see Box 7.2). Anonymity of participants is an important ethical issue (see Chapter 3) and participants are typically informed that their data will be anonymised (see the participant information sheet in Chapter 5, and the others on the companion website). BOX 7.2 ANONYMISING TRANSCRIPTS Anonymising data means removing or changing any information that could identify the participant. You should always change participants’ names, and the names of other people mentioned in the data, by giving them a pseudonym (fake name) (unless you have people’s express permission to use their real names and this doesn’t compromise the anonymity of other participants). As a basic rule beyond this, consider what information may potentially make the participant identifiable – such as their occupation, their age, the fact that they have three sisters. It’s important to consider what information may be potentially identifying not in isolation, but as a cumulative effect. For instance, information that a participant is a teacher might not be identifying; information that a participant is a disabled 50-year-old Chinese male teacher who lives in East Grinstead is far more potentially identifying. Whether you remove or change such information, and the degree to which you change it, depends on a number of factors including the extent to which a participant is identifiable by others who may read your research, the importance of complete anonymity to individual participants, and the importance of such information to your analysis. For instance, if you are interested in understanding experiences of eating difficulties in the context of family relationships, preserving information about the age and gender of participants’ siblings may be important. As shown in Table 7.1, there are two main ways of anonymising data – a marked, generic description, or an unmarked equivalent description. In either case, but especially when anonymising via unmarked changes, it’s important to keep a separate (password-protected) document that contains the non-anonymised data, and/or a record of all changes (so you can check what has been changed if necessary). 07-Braun & Clarke_Ch-07-Section 3.indd 169 28/02/2013 7:40:29 PM