Discourse analysis studies language above the sentence level, examining how utterances form coherent texts and conversations. Cohesion refers to the linguistic devices that link sentences together (pronouns, connectives, lexical repetition), while coherence refers to the underlying conceptual unity that makes a text interpretable as a unified whole. Conversation analysis (CA) examines the systematic organization of talk-in-interaction — turn-taking mechanisms, adjacency pairs, and repair sequences. Genre analysis studies how social purposes shape recurring text types with distinctive structural and stylistic features.
Transcribe and annotate a natural conversation, marking turn-taking mechanisms, adjacency pairs, and repair sequences. Analyze the cohesion devices in a written text. Compare genre conventions across text types (news articles, recipes, academic abstracts) to identify structural and stylistic patterns.
When linguists study language, they often focus on sentences in isolation — the grammar of a single utterance. Discourse analysis asks a different question: how do multiple utterances hang together to form texts and conversations? The field operates above the sentence level, treating coherent communication as something that must be actively constructed and maintained through specific linguistic and social mechanisms.
Two foundational concepts organize written discourse analysis: cohesion and coherence. Cohesion refers to the explicit linguistic tools that create surface links between sentences. Pronouns are cohesive devices — "Maria arrived late. She apologized immediately" — because "she" refers back to "Maria," creating a chain across sentence boundaries. Connectives like "however," "therefore," and "in contrast" are cohesion devices that signal logical relationships. Lexical repetition and synonymy (using "vehicle" to refer back to "car") also create cohesive ties. Coherence is different: it is the underlying unity that makes a text make sense as a whole. A coherent text does not just have linked sentences — it has a consistent topic, a logical progression, and meaning that hangs together when you interpret it in context. You can have cohesion without coherence (linking nonsensical sentences with "therefore") and, in some contexts, coherence without heavy cohesion (a sparse literary style that relies on the reader's inference).
Conversation analysis (CA) examines spoken interaction with the same rigor. One of its central findings is that ordinary conversation is not random — it is organized by systematic, largely unconscious rules. The most basic unit is the adjacency pair: a two-part sequence where the first part makes the second conditionally relevant. A question makes an answer relevant; a greeting makes a greeting relevant; an invitation makes an acceptance or refusal relevant. The crucial insight is that omitting the second pair part is itself a social act — silence after a question is not neutral, it is marked and meaningful. Turn-taking in conversation is similarly rule-governed: participants generally take turns without excessive overlapping or silence, using prosodic and syntactic cues to signal when a turn is ending.
Genre analysis adds a third dimension by asking how social purposes shape recurring text types. A recipe, an academic abstract, and a news article all have distinctive structural patterns and stylistic features that reflect their communicative functions. These patterns are not arbitrary — they evolved because they efficiently serve their genre's purpose. Genre knowledge is largely implicit: fluent readers and writers know what to expect from a genre without being able to articulate the rules, but the rules become visible when they are violated (a recipe with no ingredients list, an academic abstract that tells a story instead of summarizing).
An important corrective that discourse analysis provides is against the assumption that written and spoken language are simply formal and informal versions of the same thing. They differ structurally. Spoken conversation is produced in real time under cognitive load, so it features false starts, repairs, backchannels ("mm-hmm"), and turn-negotiation that written text lacks. Written text, produced with revision time, has different cohesion patterns and rarely includes the adjacency pair structure that organizes conversation. Applying spoken-language frameworks to written text, or vice versa, produces analysis that misses what is distinctive about each mode.