Craft advice on dialogue is mostly prohibition: cut the greetings, use said, avoid phonetic dialect, never let characters tell each other what both know, write on the slant. Most of it is right where it is right, and almost none of it comes with a reason. The reasons exist, in fifty years of research on recorded conversation that the teaching of fiction has seldom drawn on. Each strand below opens with the question a writer has, then gives the source that answers it.
Why does an evasion mean something?
A character asked where he was on Tuesday talks about the car, and every reader hears a refusal. The philosopher H. P. Grice explained why in talks given at Harvard in 1967 and published as “Logic and Conversation” in 1975. Hearers assume that a speaker is following the Cooperative Principle: make your contribution such as the talk requires at the stage where it occurs. Grice divided the principle into four maxims: quantity (say as much as is needed and no more), quality (do not say what you believe false), relation (be relevant) and manner (be clear, brief and orderly). A speaker who breaks a maxim covertly, as a liar breaks quality, simply misleads. A speaker who breaks one openly, so that the hearer cannot miss it, is flouting it, and the hearer does not conclude that cooperation has failed. She assumes it continues at a deeper level and works out the extra meaning that would restore it. Grice called that meaning a conversational implicature. His example is a reference for a philosophy post that says only that the candidate’s English is excellent and his attendance regular: nothing false is said, quantity is flouted, and the reader concludes that the referee thinks nothing of the candidate’s philosophy.
This chapter’s central identification follows. What writers call subtext is conversational implicature: the reader computing a meaning from a maxim visibly flouted. The identification turns a matter of ear into a procedure, since there are only four maxims and therefore a small number of ways to flout one, each with its own effect, which The Mechanism sets out. It also explains why on-the-nose dialogue is inert rather than unfashionable: a line that keeps all four maxims generates no implicature and leaves the reader nothing to compute. Two cautions travel with it. The weighting of the maxims differs between communities (Elinor Ochs Keenan’s fieldwork in Madagascar in the 1970s found speakers withholding information without implying what an English speaker would infer), so a novelist must establish a community’s norm of directness before a departure from it can signify. And much indirection is politeness rather than concealment, as Penelope Brown and Stephen Levinson argued in Politeness (1987): a character who hedges may be hiding something, or protecting the hearer.
Why does a hedged yes read as a no?
Conversation analysis, the study of recorded talk founded by Harvey Sacks, Emanuel Schegloff and Gail Jefferson (their paper on turn-taking appeared in Language in 1974), supplies the machinery of sequence. Its basic unit is the adjacency pair: a first turn, such as a question, an invitation or a greeting, that makes a particular second turn expected. The technical term for that expectation is conditional relevance, and its consequence for fiction is that a missing second turn is itself an event, heard as meaning something. A character’s failure to answer costs the writer nothing and tells the reader a great deal. The second finding is preference organisation: the two possible answers to an invitation or an assessment are not symmetrical. Acceptances come at once, briefly and without explanation; refusals come with a recognisable package of delay, a hesitation such as well, a token agreement, an unrequested account and softening, and often with no outright no at all. That package is how skilled novelists write reluctance, and it explains why flat refusals in drafts (“No.” “Then I’ll go alone.”) sound stylised: they put refusing content in the format of agreement. A character who agrees after a pause and an explanation has refused something. The third finding is repair, the correction of trouble in talk. Speakers strongly prefer to correct themselves, so correcting another person’s name, date or word is rare and marked, which makes it one of the sharpest status moves a writer can give a character.
When does an argument turn, and whom is it for?
The sociologist Erving Goffman described talk as the mutual protection of face, the social value a person claims in an encounter. People soften, offer each other exits and accept transparently false excuses to protect even an opponent’s face, which is why an exchange turns at the moment one party stops doing so: a full name used, an excuse refused, a euphemism replaced by the accurate word. In Forms of Talk (1981) Goffman added footing, a speaker’s alignment to what she says. The person who utters a line, the person who composed it and the person whose position it expresses can come apart, as when a clerk delivers a company’s refusal; a character who moves from I think to people say to Ma said has shifted footing twice, and each shift changes who is answerable. Goffman’s participation framework completes the picture: every utterance is designed for everyone present, the addressed, the unaddressed and the overhearers, so that who is in the room but not in the conversation is a fact about the exchange’s structure, not about its setting.
Why must characters sound different?
The usual answer is clarity. The Russian theorist Mikhail Bakhtin gave a stronger one. The novel, he argued, is made of heteroglossia (Greek for “different tongues”): the stratification of a language into the speech of professions, generations, classes and regions, each carrying its own valuations. A character’s speech carries a social position, and two characters who sound alike occupy the same social world, which forfeits the form’s main resource. Bakhtin also distinguished novels in which every voice is finally subordinated to the author’s from polyphonic novels, his term for Dostoevsky’s, in which characters’ positions keep their independence; that is the ground for requiring that an argument staged in dialogue be left open.
The conventions have dates
Today’s conventions of presentation are properties of a period. Quotation marks settled into their present function only in the eighteenth and nineteenth centuries; the rule that tags should be said dates from the mid-twentieth century, and he rejoined in a Victorian novel is a period convention, not an error; Austen rendered much of her social interaction in summary and free indirect report, keeping direct speech for what mattered, a ratio that a modern workshop would mark down. Transcribed talk, as linguists such as Wallace Chafe have shown, is full of overlap, false starts, filled pauses and abandoned sentences, and the dialogue that reads as real keeps almost none of them. It keeps the structural features instead: refusals marked, absences heard, footing shifted, maxims flouted. Deborah Tannen’s Talking Voices (1989) found that people retelling a conversation do not quote it either; they invent and compress, producing what she called constructed dialogue. The novelist’s stylisation of speech formalises something ordinary speakers already do.
See also: the Reference theory section, with the maxims tabulated against the subtext mechanisms, the period table of conventions and the sources; the consolidated glossary.