What kind of transcript is most useful?
A transcript is not a neutral, simple rendition of the words on a tape (Ochs, 1979). Different transcription systems emphasise different features of interaction. For example, a researcher with a speech therapy interest will need a system that records phonetics; a sociolinguist concerned with language variety will need some way of showing accent. What system will a discourse analyst need? One common approach claims that discourse analysis concerned with broad content themes such as interpretative repertoires needs only a relatively basic scheme which represents the words and relatively gross features such as corrections and hesitations, while analysis more concerned with interactional specifics will need to represent pause length, emphasis, various intonational features, overlap and so on. Although there is some sense to this, it obscures some tensions that need careful consideration.
In the first place, it is not easy, nor analytically satisfactory, to make a strong distinction between content and interaction in this way. Indeed, one of the consequences of using a basic transcription scheme is that it will often fail to capture those features which show how the content is occasioned by the interaction that is taking place. For example, with a research interview it may obscure just how much the participants’ ‘responses’ are a product of various activities (some very subtle) on the part of the interviewer. Moreover, one of the virtues of discourse analytic work is that readers should be able to assess the interpretations that are made because at least a selection of the original analytic materials have been reproduced. It might be argued that, even if the analyst is not making use of interactional specifics, a reader should have them available so they can make their own judgements.
Having made this strong argument for a fuller approach to transcription, it is important to stress that doing a good transcript is very demanding and time consuming. It is hard to give a standard figure for how long it takes because much depends on the quality of the recording (fuzzy, quiet tapes can double the time necessary) and the type of interaction (a couple talking down the phone present much less of a challenge than a multi-party conversation with a lot overlapping talk and extraneous noise); nevertheless, a ratio of one hour of tape to twenty of transcription time is not unreasonable. However, this time should not be thought of as dead time before the analysis proper. Often some of the most revealing analytic insights come during transcription because a profound engagement with the material is needed to produce good transcript — it is always useful to make analytic notes in parallel to the actual transcription.
The most commonly used system in conversation analysis, and increasingly discourse analysis, was developed by Gail Jefferson. It is a system developed to be easily used with the symbols on a standard keyboard, and records features of interaction that have been found to be important for talk-in-interaction. For fuller accounts of Jefferson’s system see Atkinson and Heritage (1984), Button and Lee (1987), Jefferson (1985) and Psathas and Anderson (1990).
Can you illustrate the value of a good transcript?
There are numerous CA studies that show how features of interaction that are often missed out in more basic transcripts can be analytically useful and interactionally consequential. Most of the work on ‘preference organization’, for example, (the differential marking of responses such as acceptances and refusals, agreements and disagreements) is inconceivable without the availability of high quality transcription (see Levinson, 1983).
Here is a fragment of talk from a relationship counselling interview. The Counsellor (C) is asking the woman (W) about the sequence of events leading to the request for counselling.
Counsellor: Wha- (.) what happened at that point.
Woman: At that poi:nt, (0.6) Jimmy ha- (.)
my- Jimmy is extremely jealous.
Ex- extremely jealous per:son.
Has a:lways ↓been, from the da:y we met.
Y’know?
(DE-JF/C2/S1 — p.4)
| The Transcription Symbols Used in the Extract A dash marks a sharp cut off of the prior word. A colon marks an extension to the preceding sound. A dot in brackets shows a noticeable pause to short to measure. Underlining marks emphasised delivery. A full stop indicates a ‘completing’ intonation, while a comma indicates a continuing intonation. |
Two things may be immediately striking about this extract. First, the transcription symbols may make it hard to read easily. Second, there seems to be a lot of ‘mess’ in it: repairs and changes of gear. On the first point, reading transcript is itself something of a skill which develops with familiarity. After a period getting used to materials of this kind it is the transcript without the symbols that looks odd — idealized, cleaned and shorn of its specificity — while the fuller transcript becomes evocative of the interaction captured on the tape. It starts to be possible to hear, in a sense, the delivery.
On the second point, what seems on first reading to be mess is quickly understandable as something much more organized. The woman breaks off her direct answer to provide a description of her partner. This description is reformulated and emphasised until it has a precise sense that is suitable for the business in hand (briefly and over-simply: starting to display how the relationship problem is his fault — for more on this, see Edwards, forthcoming; Potter, forthcoming). The careful transcript here allows us to see this final version being actively shaped, and gives us a feel for the versions that are rejected as unsatisfactory.
What about reliability and validity?
The notions of reliability and validity have increasingly taken on a mix of everyday and technical senses in traditional forms of psychology. Reliability is taken to be established in a quantitative fashion by techniques such as test-retest correlations or inter-rater reliability. Validity is often treated as established by a congruence between different instruments, or perhaps a triangulation from different research methods. Because of the different theoretical assumptions in discourse work, along with its largely non-quantitative nature, these approaches to reliability and validity are largely unworkable here. Nevertheless, these are important considerations which can be, and have been, addressed in this work.
Reliability and validity are not so clearly separated in discourse work. Various considerations are relevant; four important ones being deviant case analysis, participants’ understanding, coherence and reader evaluation.
Deviant case analysis. A discourse analytic study will often work with a collection of instances of some putative phenomenon with the aim of showing some pattern or regularity. For example, an analyst might claim that news interviewees generally avoid treating interviewers as responsible for views expressed in questions (Heritage and Greatbatch, 1996). One of the most useful analytic phenomena are cases which seem to go against the pattern or are deviant in some way. In this type of work such deviant cases are not necessarily disconfirmations of the pattern (although they could be); instead their special features may help confirm the genuineness of the pattern (Heritage, 1988). For example, when a news interviewee does treat the interviewer as accountable for a view posed in a question it can create serious trouble for the interaction (Potter, forthcoming). The deviant case can show up just the kind of problem that show why the standard pattern should take the form it does.
Participants Understanding. As I noted in the discussion of conversation analysis, one of its important elements is its use of participants’ own understandings. Thus instead of the analyst saying that this turn of talk is a compliment, say, the focus is on how the participants’ treat it. At its simplest, is it responded to with an acknowledgement, perhaps, or a depreciation: ‘Oh that’s very sweet of you, its just an old top I picked up cheap?’ A common critique of discourse analytic work is that there is no check on its interpretations. However, a close attention to participants’ understandings provides one kind of check.
Coherence. One of the features of conversation analysis and increasingly of discourse work is its cumulative nature. A set of studies can be combined together, and can build on the insights of earlier work. For example, work on fact construction builds on the insights about accountability from earlier studies, and its success provides a further confirmation of the validity of those studies (Edwards and Potter, 1993). There is a sense, then, that each new study provides a check on the adequacy of previous studies that are drawn on. Those studies which capture something about interaction can be built on, those that do not are likely to become ignored.
Readers’ Evaluation. Perhaps the most important and distinctive feature in the validation of discourse work is the presentation of rich and extended materials in a way that allows readers of discourse studies to evaluate their adequacy. This has two facets. On the one hand, it allows them to assess the particular interpretation that is made as it is presented in parallel with the original materials. This is not the case in much ethnographic work where the interpretations have to be taken largely on trust and what data that is presented is largely pretheorized; nor is it the case with much traditional experimental and content analytic work where it is rare for ‘raw’ data to be included and more than one or two illustrative codings presented. On the other hand, readers are themselves skilled interactants with a wide range of cultural competencies as viewers of news interviews, members of close relationship, recipients of compliment, and so on. Thus the judgements they can make are not merely abstract ones of the relation between materials and interpretations, but of the adequacy of more general claims.
A final comment on validity. These features are not all present in all discourse studies; nor do they singly or in combination guarantee the validity of an analysis. As sociologists of science have repeatedly shown, there are no such guarantees in science.