Conversation Analysis: Key Concepts & Research Domains in Linguistics

Professor Ontake has identified six different conceptualizations of applied CA: (1) Foundational applied CA seeks to reframe or repurpose foundational concepts within disciplines like psychiatry, such as Edward and Potter's discursive psychology which positions constructs like memory as fundamentally interactional rather than internal. (2) Social problem-oriented applied CA aims to better understand social problems and institutions, addressing macro issues like class, gender, and power—exemplified by Kitzinger's work on heteronormativity and feminist CA. (3) Communication applied CA attends to disorder talk to understand its features and challenge how disorders have been classified, highlighting communicational competence of individuals who might otherwise be assumed incompetent. (4) Diagnostic applied CA attempts to correlate speech organization with underlying disorders, though this is controversial—Schwab's work on diagnosing epilepsy exemplifies this approach. (5) Institutional applied CA focuses on studying how institutions carry out their work, examining courtroom, classroom, and mental health interactions. (6) Interventionist applied CA attends to pre-existing problems with the aim of intervening with findings to provide solutions.

Conversation analysis is an approach to analyzing spoken discourse that examines how people manage their everyday conversational interactions, originating from ethnomethodology at UCLA in the 1960s; it investigates social order through systematic patterns in talk such as adjacency pairs, turn-taking, feedback, and repair, and has been applied to diverse settings including doctor-patient consultations, legal hearings, and classroom interactions.

Sequential effects in healthcare communication are largely causal effects rooted in interaction rules rather than individual differences in personality or psychology. Across numerous studies, psychological and demographic variables rarely affect these effects. Participants can be trained to employ CA-derived strategies, and these training effects endure over time. These sequential effects can then be associated with distal health outcomes like patient satisfaction and treatment compliance. CA accounts for embodiment including body language, eye contact, and physical positioning, which affect interaction flow. Cultural factors are difficult to identify in interaction since people rarely explicitly invoke cultural identity, though cultural understanding matters in how medical advice is implemented—for example, doctors recommending vegetables didn't realize patients only ate fried vegetables until explicitly asked about preparation methods.

Conversation Analysis has been applied across various institutional settings with significant findings: (1) In language speaking tests, examiner behavior (enthusiastic vs. disinterested responses) affects candidate performance; (2) In counseling, counselors use paraphrasing to demonstrate active listening and redirect conversations; (3) In clinical communication, doctors' online commentaries during examinations shape parents' treatment expectations by validating concerns while managing expectations for non-aggressive treatments. These applications demonstrate how CA reveals hidden interactional dynamics that influence institutional outcomes.

Discourses are not just ways of speaking and writing—they are bound up with institutional practices, ways of organizing, regulating, and administering social life. These include the medical profession, scientific profession, legal profession, politics, and media. Being positioned as a patient in a medical discourse means one's body becomes an object of legitimate interest to doctors and nurses, who may expose, touch, invade, treat, operate on, and administer drugs to the patient. These institutional practices come with specific discourses and subject positions (doctor, nurse, patient, victim).

Modern CA increasingly integrates non-vocal elements (eye gaze, gestures, body position) with vocal elements, recognizing they are laminated onto each other rather than separate phenomena. Some transcription systems now show non-vocal elements alongside talk lines, showing precise timing of gaze shifts and gestures relative to speech. However, integration remains challenging and varies by research question complexity.

The Lyon school introduced a significant concept in conversation analysis called the multimodal approach to conversation. This approach enabled researchers to move beyond the intrinsic analysis of linguistic units by integrating acoustic and gestural aspects as parameters of analysis, which had previously been neglected. The fundamental hypothesis is that the conversational flux is constantly managed by coordinated activities of the speaker and listener. Verbal exchanges allow describing interaction as a progressive construction assured by joint and coordinated activities of partners. This perspective requires describing and comparing all regulatory signals—both non-verbal (expressive gaze, smile, motor head nods) and verbal (verbal regulators like 'oui', 'ouais', 'c'est vrai')—which punctuate all verbal production and serve as reference points for listeners.

CA can extend to capture nonverbal communication through video recording, including gaze direction, pointing gestures, body positioning, and subtle movements. This is particularly important for studying groups like autistic children who rely heavily on nonverbal cues. Capturing these elements requires mapping visual aspects onto transcripts, making analysis even more labor-intensive than traditional Jeffersonian transcription. Researchers like Paul Dickerson have used video-based CA to study how children interact with robots using gaze and pointing.

Conversation analysis, originating in the US in the mid-1960s, studies turn-taking and the roles of verbal and non-verbal elements in social interactions. The interest in multimodal approaches emerged in recent decades to analyze additional resources beyond speech, including posture, body movements, gestures, eye contact, and proxemics (use of space). This multimodal conversation analysis examines how people coordinate multiple semiotic resources simultaneously during face-to-face interactions, representing a sophisticated form of multimodal communication.

Multimodal analysis is an approach within conversation analysis that examines communication resources beyond spoken language. This includes gestures, eye gaze, body positioning, and objects in the immediate environment. These elements are treated as communication resources that participants combine and coordinate during interaction. Multimodal analysis became a significant focus in conversation analysis research around the 2000s, particularly in Japan around 2010.

Conversation analysis provides empirically grounded guidance for improving professional communication across healthcare, law enforcement, and service industries. Research reveals that real police interviews differ significantly from simulated ones, with real officers using modal verbs and specific reference points that elicit better accounts. Similarly, GP telephone consultations with high patient burden correlate with poor patient experience scores. These findings demonstrate that systematically analyzing naturally occurring data reveals practical improvements that intuition cannot identify, enabling evidence-based communication training and policy development.

Conversation design, which focuses on how humans communicate with computers through voice assistants and chatbots, can be significantly improved by applying insights from conversation analysis—the scientific study of how humans talk to each other in natural settings. By analyzing real-life conversations between people, designers can identify key patterns such as question intonation variations, appropriate interruption timing, and self-repair mechanisms that should be incorporated into conversational interfaces. This approach recognizes that while humans engage in conversation daily without consciously understanding the underlying rules, conversation analysts have systematically studied these patterns, providing valuable data that can inform more natural and effective human-computer interactions.

This comprehensive section establishes the theoretical foundation of conversation analysis—a 50-year-old interdisciplinary field spanning sociology, psychology, and linguistics that investigates real-world social interactions through detailed transcription and analysis. The panelists introduce key concepts including segments (high-level customer categorizations) and personas (fictional character representations), then pivot to recipient design—the critical observation that effective interactions are designed as best fit for participants through word selection, intonation, prosody, and embodied resources. Through comparative analysis of human and AI interactions, the section demonstrates how recipient design enables frictionless, efficient interactions that adapt to each participant's emerging evidence. The research reveals that conversation has systematic order beneath apparent messiness, and that effective practitioners continuously adapt their approaches based on unfolding interaction dynamics rather than applying uniform strategies.

This comprehensive section establishes the theoretical foundation for approaching AI through conversation analysis. It begins with Sachs' seminal metaphor of the 'talking and doing machine,' illustrating how disciplinary assumptions create artificial research problems—researchers who understand only one aspect of a system assume its purpose, while sociologists explore discrepancies between linguistic descriptions and practical actions. Garfinkel's criticism extends this critique, arguing that sociological models exist primarily to be studied rather than to explain reality. The section traces how Lucy Suchman's ethnomethodological work at Xerox PARC demonstrated fundamental differences between designers' ontologies and users' situated actions, inspiring researchers to reframe technology design questions as studies of work practices. Button and Dourish articulated twin paradoxes: modifying technology based on user studies reflexively changes the practices that generated those insights, while without existing systems, where does design begin? The section culminates with empirical application to Google Duplex, revealing how CA methods can identify structural features like the 'anchor position' without anthropomorphizing AI, demonstrating that AI systems can reproduce natural conversation patterns including prosodic elements.

Applied Conversation Analysis (ACA) is a qualitative research methodology that analyzes naturally occurring spoken interactions to understand how social actions are performed through language in real-world settings. Unlike traditional research methods that rely on self-reports or structured interviews, ACA focuses on the sequential organization of talk, examining how participants negotiate meaning turn-by-turn in everyday and institutional contexts. The approach uses Jeffersonian transcription to capture not just what is said but also how it is said, including intonation, pauses, and other interactional features. ACA has been successfully applied across diverse fields such as healthcare (mental health assessments, clinical consultations), education (online discussion forums, classroom interactions), and social work. Key practical considerations include obtaining appropriate ethical clearance for recording interactions, collecting sufficient data to identify patterns, and disseminating findings to both academic and practitioner audiences. Learning ACA requires extensive practice with real data, engagement with seminal literature, and participation in collaborative data analysis sessions.

Membership Categorization Analysis (MCA) is a sociolinguistic methodology that examines how members of society use category systems (such as occupational roles like 'police officer') to make sense of each other's actions; these categories come with attached predicates (activities, knowledge, motivations) that create bidirectional inference—when we observe someone performing a category-tied activity (like arresting), we identify them as belonging to that category, and conversely, when we identify someone as belonging to a category, we expect them to perform its associated activities, enabling members to observe, identify, and report on social actions in real-time.

The MCA expansion included 6 new ministers and 2 promotions, totaling 8 new faces. The composition was: 5 OBC (backward classes), 2 Dalits, 1 Brahmin. This breakdown is presented as evidence of BJP's 'social engineering' to balance the MCA with PDA communities. The total MCA size became 60 ministers, with 8 being new additions.

CA has been employed in feminist linguistics and integrated with membership categorization analysis (MCA), which examines how categories hierarchically link together in social organization. Membership categorization devices (MCDs) represent central machinery of social organization, storing much of society's knowledge about itself. For instance, 'mommy' is interpreted as the mother of the baby by native speakers, demonstrating how categories enable practical inferences about social reality. This compatibility with feminist studies allows examination of how social categories shape interactional practices and power relations. Key phenomena in CA literature include turn-taking (allocation of participation rights), repair (mechanisms for handling interactional troubles), preference organization (structural preferences for certain actions), and sequence organization (ordering of interactional actions). These phenomena represent fundamental building blocks for understanding systematic organization of social interactions.

This segment explores how individuals construct and negotiate their social identities within group contexts. Characters demonstrate how group membership shapes self-perception, how social labels affect how others treat individuals, and how individuals seek to establish their identity within social hierarchies. The segment shows how group belonging can provide identity and security, but also how social categorization can create vulnerability and discrimination.

MCA results are interpreted by examining positions of categories in the factorial space. Categories close together have similar profiles. The contribution of each category to each axis indicates its importance in defining that dimension. The cos² values (squared cosines) indicate how well each category is represented in the projection. Categories far from the average profile show strong associations.