Chat: pretty basic stuff; we can all recognise it and we can all do it. Yet when we come to define chat we have to make decisions about its character. For us, chat is defined by its 'informality' (not that we are capable of defining that), not its modality. Thus it names informal textual interaction as well as informal voiced interaction: holiday postcards, letters to friends and informal emails, along with telephone and dinner table conversation. However, in Conversation Analysis (CA) -- the pre-eminent mode of 'chat analysis' -- textually produced interaction is not considered an altogether appropriate topic, and 'textual CA' remains marginal at best (McHoul; Nelson; Mulkay).
According to CA, conversation (or 'talk-in-interaction') is the primordial mode of social interaction. As a practice, ordinary talk is not considered by its practitioners to be particularly skilled (presumably because it is so basic, so pervasive, so ordinary); yet CA shows it to be a precision instrument, wielded by maestros. Subtle, nuanced and highly sensitive; yet structured, normative and accountable; it displays "order at all points" (Sacks 22), yet is entirely improvised. Moreover, the doing of talk produces and reproduces all the supposedly 'external' phenomena of the socio-psychological sciences: persons, interaction, groups, membership categories (class/gender/ethnicity), the 'sense of social structure' and ultimately society itself. For CA, 'naturally-occurring ordinary conversation' is at once the most mundane and the most consequential social phenomenon (Boden & Zimmerman; Silverman; Hutchby & Wooffitt).
Part of our aim is to provoke a reconsideration of the marginal status of textually conducted interaction as a proper topic for CA. We do this by comparing two forms of CA and their corresponding data -- the CA of face-to-face conversations and the CA of Internet newsgroup messages (for the latter, see Reed, "Being 'OTP'"; Reed, "Newsgroups"; for other insider commentaries and critiques of CA practices, see Anderson and Sharrock; Bogen).
The main axis of comparison between the two forms of analysis is the relevant procedures that each uses to produce its data; and in particular, the subset of these we call 'machinic-productive processes' (see Table 1).
If we concentrate on the two heavily outlined boxes which list each mode's machinic-productive processes we can note first that these processes are assigned to different actors: for the CA of conversation, they are the task of the analyst, whereas for the CA of newsgroup messages, they belong to the participants.
The machinic-productive processes required to transform newsgroup messages into data for CA are carried out, pre-analytically, by newsgroup participants as an integral part of their practice. Note that the term 'participants' includes many more 'actants' than just message writers. There are other humans, mainly message readers of various kinds (active readers/writers, passive read-only 'lurkers', and, sometimes, policing readers known as 'moderators'). And crucially, there are various nonhumans too (computers, newsreading and archival/retrieval software, electronic networks); see, for the participation of nonhumans in social arrangements, Latour; and for 'actor network theory' generally, Law & Hassard.
The threadedness and retrievability process sorts and juxtaposes messages with mutual reference and relevance into interactional streams known as 'threads'. Newsreading and archival/retrieval software produce a visible display of the topical and temporal relations between the messages that comprise a newsgroup. Thus the (human) newsgroup reader, and of course the analyst, is presented with pre-formed sequences of messages. This is how, at the inter-message level, the interactional character of newsgroups is built; without which, they would not be available as potential data for CA with its overriding interest in interaction. A more technical result of this machinic-productive process is the (temporary) preservation of a set of publicly-available messages.
The process of textual composition and editing includes the set of co-operative procedures carried out by message writers and message-writing software. For example, messages with a 'conversational' character can be formed by the insertion of new text into edited portions of the prior message(s) automatically generated by the software's 'reply' command (for a detailed analysis of this practice, see Reed, "Sequential Integrity"). This, then, is one way that the interactional character of newsgroups is produced at the intra-message level. A further, very basic, product of this second machinic-productive process is, quite simply, text.
The analyst of conversations has more (initial) work to do than has the analyst of newsgroup messages. She has to transform the raw material of some bit(s) of talk in the world into something useable, i.e. data. This involves, among other things, the machinic-productive processes of recording and transcribing. These processes are 'machinic' in that they are technologically mediated, requiring the use of audio/video recording machines and codified transcription systems. They are 'productive' because their use results in something new, something that is qualitatively distinct from the (supposedly) 'naturally-occurring' object that is said to be this novel object's original and model.
Recording transforms a private, participants-only piece of talk into one that is overheard: the researcher has conjoined the interested parties; the talk has been bugged. In addition, recording transforms an ephemeral 'been and gone' occasion into a 'frozen moment', preserved out of time. Recording, then, like the machinic practices in newsgroup message production, de-privatises and preserves. These are large and consequential transformations. For example, they enable a continual return to the data for re-listening, re-hearing and possibly re-transcription (for a critique of the 'return to the data' trope, see Ashmore and Reed).
Discussions of the practices of recording talk (or of its productive effects) are noticeably absent from the primary and secondary literature of CA (see, for example, Pomerantz and Fehr; Hutchby and Wooffitt; Silverman). This is not accidental. It is one aspect of a distinctive attitude in CA to its materials, as encapsulated by Harvey Sacks in the following, much quoted, comment:
I started to work with tape-recorded conversations. Such materials had a single virtue, that I could replay them. I could transcribe them somewhat and study them extendedly. (Sacks 26)
Let us simply note the criteria for adequate data implied here:
the data must be portray-able as 'accidentally', or 'irrelevantly' collected.the data must be record-able and re-playable.the data must be transcribe-able.
the data must be re-study-able.
For data to be usable for CA, they must become 'independent objects' detached from their specific origins. This is achieved through their (unattended-to) availability to the machinic recording and transcription processes. Adequate data must have this 'machinic potential'.
The most basic consequence of transcribing is a shift in modality from sound to text. In itself, a text is more distributable, more publicly-available than a tape. Because the transcript is a result of 'hearing work' done on the tape, it also acts as a public display of the analyst's otherwise private and subjective understandings. The particular practice of transcription used in CA is codified -- indeed, frequently capitalised -- as The [Gail] Jefferson Transcription System. A properly formulated CA transcript is not, to the uninitiated, easy to read, being full of abstruse symbols and replete with details of pauses, overlaps and false starts; the features which are usually written out of other transcriptions of talk. To the cognoscenti, however, it is precisely these elements which not only 'speak CA' but also act to produce and display the naturally-occurring character of the data. The more complex the visual 'look' of a CA transcript, and thus the more worked-up it is, the more we are persuaded of its pristine origin, 'untouched by analyst's hand'.
For mainstream CA, newsgroup messages look unpromising as material for analysis. This is because, we are arguing, they do not require the analyst to engage in the data-production processes we have outlined in our discussion of the recording and transcription of conversations. That is, the procedures needed to transform newsgroup interaction into data for analysis are less 'radical' than those needed for conversations. Newsgroup messages are, as it were, pre-'recorded' and pre-'transcribed' as an inherent part of their production: they are already public, already preserved and, of course, already text.
It is our strong impression that one source of the perceived inadequacy of newsgroup material in and for CA, is a distaste for its evidently 'machinic' character; it seems to lack the stamp of a fully human origin, it seems too 'artificially occurring'. Yet, ironically, in one sense of the 'naturally occurring' criterion, newsgroup data would appear to be clearly superior to conversation. In that their worked-up, machinic character is the result of participants' work, they are considerably less mediated, more 'natural' than recorded and transcribed conversations.
For the CA of conversation, we are arguing that, far from having its origin, its model, and its validation in some naturally-occurring, real-time, real-world, locally-specific occasion(s) of talk, its object only becomes recognisable as 'conversation' as a result of the machinic-productive processes we have described. All the things that, for CA, are definitional of conversation -- sequence, turn taking, adjacency pairs and the like -- gain their reality by being worked-up from data made in this way. This is equally true for the CA of newsgroups; in both domains, their respective machinic-productive processes function to manufacture chat. However, in the CA of talk, they produce something else: the myth of an unmediated origin. In formulating conversation as a naturally-occurring phenomenon, their own productive work in so doing is systematically obfuscated. The newly man-and-machine made object erases its original and replaces it with ... itself.