Working with contemporaneously produced text & talk: A hybrid problematising workshop

Held on Friday May 22nd, Loughborough University & Online

Workshop convenors: Emma Richardson, Sarah Atkins, and Alexandra Kent

This workshop, hosted by the internationally renowned Discourse and Rhetoric Group and supported by the CRCC brought together international scholars of ethnomethodology and conversation analysis (EMCA) concerned with how to collect, analyse and present data from in-person conversations where contemporaneously textual accounts are being produced. This concern is not only methodological or analytical but is a live concern for participants across a range of interactional contexts as they make choices about what to attend to.

Take aways from us as the convenors

· The gap between talk and text is systematic, not incidental – Across every setting discussed, researchers reported how complex spoken interactions must routinely be compressed into brief, formulaic written entries.

· Tools and templates shape talk – Electronic systems, pop-up prompts and structured forms don’t just record the interaction, they shape what gets asked, in what order and what gets documented.

· AI-generated records raise new questions – AI scribes are already changing how records are produced. Understanding how skilled human operators navigate entextualisation judgements is important groundwork for designing automated systems that can support this.

· Aligning talk and text data is a shared methodological challenge – Even with time-stamped records, tracking log entries relative to the conversation is not straightforward

Talk and text in clinical encounters

To start the day, Graham Johnson, Consultant in Paediatric/Emergency Medicine (University Hospitals of Derby & Burton) and Doctoral Researcher (Loughborough University) showcased the challenges with ‘interruptions’ when documenting ‘referral conversations’ in the Emergency Department. Of central importance is that what is contained in the record ‘sets the tone’ and initiates a trajectory for the patient’s clinical journey. Graham posed the question, “what gets lost between the conversation and the notes?”

AI-scribes in General Practice produce records during the consultations mitigates the issues with interruptions. Thinking methodologically, Deborah Swinglehurst (Queen Mary University of London) raised challenges with the feasibility of gaining access to the ‘AI-scribes– in use’ during the consultation, but also the work that clinicians do to edit the text and codes being produced. Especially relevant given that records are now readily accessible to patients. Deborah raised analytic questions for academics and practitioners around how GPs adapt their consultation practices to accommodate the presence of the AI-scribe, and how we might ensure our research in this area remains current in such a fast-changing technological landscape.

Al Hughes works within an outpatient clinical setting within the NHS, seeing patients prior to major surgical procedures. Liz Stokoe (LSE) and Al are working together on the possibility of having a real-time means of recording complex surgical preassessment appointments. Summarising and documenting (for medical colleagues, patients and their health record) these conversations that might be more effective than the current summary letters produced from the clinician’s recollections after the appointment.

This panel strikingly demonstrated the highly consequential relationship between text and talk in clinical settings and how the future audiences and use of records shapes what can be recorded and how.

Talk and text in other institutional settings

Outside of the clinical setting, Jennifer Watermeyer (University of the Witwatersrand) provided an example the challenges faced by those seeking to make sense and use of multiple available texts, when there are differing participation frameworks. Discursive Psychology has long eschewed the relevance and inclusion of ‘context’ regarding participants cognitive matters and states of mind, focussing on only what is observable in the interaction. However, the question asked was how do we manage the differential access participants have to the texts, especially if we can use them retrospectively.

Katariina Harjunpää (Tampere University), focused our attention on ‘repair organisations’ when correcting texts produced collaboratively during a mediation agreement. Then, Aafke Diepeveen (University of Agder), guided us through how statements are collaboratively produced during police-suspect interviews. Both settings are illustrative of institutional encounters which routinely involve the live production, refashioning, and consultation of written records alongside spoken interaction. In the discussion, Aafke opened up the possibility that other complementary methodologies, frameworks and terminology could be applied.

Our discussion highlighted the interactional and analytic challenge of managing multiple orientations at once, especially when not all the orientations are similarly understood or known to all participants in the interaction.

Methodological challenges and approaches

Saul Albert and Liz Stokoe then offered a story which served as a reminder of the ‘The ordinary inadequacy of ‘conversational AI’ and the importance of the tacit work humans do in institutional settings. They highlighted the inadequacy of LLMs engagement in the delicate interactions we’d seen clinicians and other institutional practitioners manage during the morning sessions. LLMs do not have the same stake in what is recorded and how it is interpreted.

Lorenza Mondada (Univerisity of Basel) guided us to back to earlier consideration on shifting participation frameworks, who has access to what. Lorenza reflected on how video data are generally not made to capture the details of documents manipulated (such as screens which are being read from). In some cases, it is possible to access materials (such as in Jennifer’s case), but as Lorenza showed, their connection to the video records is sometimes hard to establish precisely. Jessica Robles (Loughborough University) extended these analytic concerns by pointing us to the question; how can we analyse something in a social environment that’s not made available to co-participants? And, is there value in reproducing transcripts of simultaneous but seeming-parallel events “on the same page” in some sense?

We arrived at the tentative answer that, if the text engagement relates to the main interactional activity, then it is useful to have it represented in the analysis in a way that more tangential activities might not need to be represented.

Talk/Text in 999 kidnap call data

Finally, Sarah Atkins (Aston University) facilitated an in-person data session featuring an extract from a UK police emergency call where the caller was reporting a kidnap. Working hands-on with this data we were able to apply our earlier discussions. These high-stakes incidents unfold across dual communicative tracks: (1) the spoken interaction between caller and call-handler, and (2) the simultaneous written text of the Computer-Aided Dispatch (CAD) log. The CAD log, initially typed by the call-handler but open to real-time contributions from dispatchers and other officers, becomes a crucial institutional text, shaping the ongoing interaction with the caller, the immediate police response and the longer-term record of the incident. The session focussed on the representational and analytic challenges these talk-text interactions pose for conversation analysis and related approaches.

Conversation Analysis and Ethnomethodology have the ability to provide powerful, grounded insights into these crucial and consequential interactive environments. As analysts we need to think carefully and deeply about the adequacy of the methods we use.