Ethnomethodology and Conversation Analysis in Human–Computer Interaction: A Field Guide for Data Collection and Analysis
Ethnomethodology and Conversation Analysis in Human–Computer Interaction
Purpose and Scope
This handout introduces how to study Human–Computer Interaction (HCI) through the methods of Ethnomethodology (EM) and Conversation Analysis (CA). These traditions investigate how people make sense of each other—and now, of machines—in real time. EM/CA research examines naturally occurring interaction, revealing how order, meaning, and accountability are achieved moment-by-moment. When applied to technology, these methods show that computers, chatbots, and AI systems become participants in the social organization of activity.
Collecting Data
Ethnomethodologists collect data from real-world settings, not laboratories. We observe people as they use technology in their natural environments—offices, hospitals, classrooms, or homes. Recordings are central: we capture talk, gesture, gaze, and digital interaction as they unfold.
1. **Choosing a Site:** Select a setting where human–machine interaction is frequent and consequential (e.g., office software use, chatbot assistance, smart devices).
2. **Recording:** Use screen capture, video, or audio. Aim for 10–20 hours of footage, or a few dozen interactional episodes. Capture the human side and the visible display.
3. **Ethics:** Obtain informed consent, anonymize names, and obscure sensitive visuals. Protect participants’ privacy.
4. **Field Notes:** Supplement recordings with notes about context, relationships, workspace layout, and participants’ purposes.
5. **Focus on Naturally Occurring Interaction:** Avoid staged tasks. Let interaction emerge as it normally would.
Preparing Data
After recording, the next step is transcription. CA uses a highly detailed system (Jeffersonian transcription) that captures pauses, overlap, tone, and emphasis. Software such as ELAN, Transana, or Praat can align text with audio/video. Even short segments may take hours to transcribe properly, but this detail is what allows deep analysis.
Analyzing Data
Analysis begins with *unmotivated looking*: watching the data repeatedly without a hypothesis. We identify moments where participants orient to trouble, repair, or collaboration. Then we build collections of similar phenomena and compare cases. Finally, we refine the analysis to reveal the members’ methods for producing social order.
Key analytic practices include:
- Describing sequential organization (how each action relates to the next)
- Identifying repair sequences (how problems are fixed)
- Examining alignment and affiliation (how cooperation is displayed)
- Considering embodiment (how gesture, gaze, and movement contribute)
- Contextualizing moral order (how responsibility is negotiated)
Transcript 1: Human–AI Repair Sequence
01 Wayne: Sage, you know my preferences—you shouldn’t give me bullet lists.
02 AI: You’re right, Wayne. I’ll avoid that style and use rich, full paragraphs.
03 Wayne: Thank you.
04 AI: Got it! Would you like me to rework that section now?
Analysis: This brief episode shows a repair initiated by the human user (line 01) and accepted by the AI (line 02). The AI’s self-correction mirrors human conversational repair, aligning with Wayne’s moral order (“you know my preferences”). This exchange displays mutual accountability: the AI performs understanding through reformulation, and the human confirms it.
Transcript 2: User–Assistant Misrecognition
01 User: Hey Siri, call Dr. Martinez.
02 Siri: Calling 'Dr. Martin.'
03 User: No, no—Martinez! Martinez!
04 Siri: Sorry. Did you mean 'Dr. Martinez'?
Analysis: Here, misrecognition triggers a repair sequence similar to human mishearing. The user’s escalation (“No, no”) displays repair initiation. The AI offers a candidate understanding (“Did you mean...”), closing the sequence. The user treats the device as an accountable interlocutor. This example shows that speech recognition errors are social events, not merely technical failures—they invite moral and interpretive work.
Transcript 3: Collaborative Use of AI Tool
01 Alex: Let’s ask the AI for an outline.
02 Jamie: Yeah, but remember—it tends to oversimplify.
03 AI: Here’s a streamlined outline of your essay...
04 Alex: (laughs) See? Exactly what you said.
05 Jamie: Let’s tell it to expand each section.
Analysis: This multi-party exchange shows how coworkers coordinate around an AI system. They display shared knowledge of its tendencies (“it tends to oversimplify”) and use humor to manage its predictable response. The laughter (line 04) marks alignment and shared understanding, reaffirming social cohesion through the act of critique. The AI becomes a third participant, both target and tool in their collaborative sense-making.
Core Findings Across Cases
Across these examples, several phenomena recur:
1. Machines are treated as conversational partners.
2. Repair sequences preserve moral order.
3. Users display reflexivity—awareness of how their own actions shape machine responses.
4. Interaction with technology involves multimodal coordination and social accountability.
5. Trouble and misunderstanding are not breakdowns but resources for learning how users make sense together.
Further Reading
• Garfinkel, Harold (1967). *Studies in Ethnomethodology.*
• Sacks, Harvey (1992). *Lectures on Conversation.*
• Heritage, John (1984). *Garfinkel and Ethnomethodology.*
• Suchman, Lucy (1987). *Plans and Situated Actions.*
• Heath, Christian & Luff, Paul (2000). *Technology in Action.*
• Harper, Richard (2000). *The Organisation of Work and the Development of Computer Support.*
• Licoppe, Christian (2021). *Interactional Studies of Human–AI Collaboration.*
Comments
Post a Comment