Description
Analyses chatbot or voice assistant transcripts turn by turn using conversation analysis concepts such as adjacency pairs, repair and preference organisation.
Conversation analysis (CA) studies how people organise talk-in-interaction: turn-taking, adjacency pairs such as question and answer, repair of misunderstandings, and preference for certain responses over others. Applied to chatbots and voice assistants, CA examines real transcripts turn by turn to see where the system violates conversational expectations, for example by failing to recognise repair initiations, answering a question that was not asked, or producing dispreferred responses without mitigation. Because breakdowns are located precisely in the sequence, the findings map directly onto dialogue design changes. Rigour comes from systematic transcription, analysing sequences rather than isolated utterances, building collections of comparable cases, and checking interpretations against what the next turn reveals about how the user understood the system.
Attributes
- Discipline
- Linguistics, Semiotics & Discourse
- Loss type
- Latent
- Stage
- Evaluation, Post-launch
- Data
- Qualitative
- Effort
- Days
Steps
- Collect 50+ chatbot conversation transcripts. Code each turn for:
- adjacency pair completion (does the bot's response constitute an appropriate second pair part?),
- repair (does the bot handle user corrections?),
- topic closure (does the conversation reach a coherent ending?). Compute violation rates per category. Target: adjacency pair completion > 85%, repair success > 70%. Violations are redesign priorities.
Sources
- Conversational Design — Erika Hall
Also known as
- CA for conversational UX
- Turn-taking analysis
- Dialogue transcript analysis
Categories
- Content
- Linguistics, Semiotics & Discourse