3. Why Dialogue Is Not Speech

Craft literature on dialogue is thin in theory and rich in prohibitions. Students are told to avoid as you know, Bob, to use said, to cut greetings, to avoid phonetic dialect, and to write on the slant rather than on the nose. Almost none of this is explained; nearly all of it is correct where it is correct; and the reason it works is available, in detail, in a body of empirical linguistics that the workshop tradition has never absorbed.

This section assembles that case: four theoretical strands — Grice on implicature, conversation analysis on the machinery of turns, Goffman on footing and face, Bakhtin on the novel’s plurality of languages — followed by a historical strand on the contingency of the conventions and an empirical strand on what transcribed talk actually looks like.

3.1 Grice: meaning generated by breaking the rules

The body of theory a writer of dialogue can least do without, because it accounts for subtext, is H. P. Grice’s account of conversational meaning, presented in Harvard’s William James series in 1967 and published as “Logic and Conversation” in 1975.

Grice’s problem was that speakers routinely mean more than they say and hearers reliably recover the surplus. His solution posits that conversation is a cooperative activity governed by a general principle and four attendant maxims.

The Cooperative Principle. Make your contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged.

So far this describes ideal, transparent talk — the dialogue of the maximally cooperative and therefore maximally boring interlocutor. Grice’s decisive move is what comes next.

A speaker may violate a maxim quietly; a lie violates Quality. But a speaker may also flout a maxim — breach it openly, in a way the hearer cannot fail to notice — and when they do, the hearer does not conclude that cooperation has broken down. The hearer assumes cooperation is being maintained at a deeper level and therefore computes an additional meaning that would restore it. Grice calls this a conversational implicature. His classic example is a reference letter for a philosophy post which says only that the candidate’s English is excellent and his attendance regular: nothing false is stated, Quantity is conspicuously flouted, and the reader computes that the referee has nothing to say about the candidate’s philosophy.

This chapter’s central identification follows directly:

Subtext is conversational implicature. What craft writers describe as the gap between what a character says and what they mean is, mechanically, the reader computing an implicature from a flouted maxim. The craft tradition discovered the phenomenon by ear; Grice specified it.

The identification is not decorative. It converts a matter of taste into a matter of procedure, because Grice’s maxims give a generative account of subtext: there are only so many ways to flout, and each produces a characteristic effect.

What matters is the shape of the claim: the reader’s work in reading dialogue is inferential, and the inference is licensed by a shared assumption of cooperation that the character is visibly straining. This is why on-the-nose dialogue is inert rather than merely unfashionable. A line observing all four maxims generates no implicature and leaves the reader nothing to compute.

Two qualifications should be carried.

The maxims are culturally variable. Grice presented them as features of rational cooperative behaviour, but comparative work since — Elinor Ochs Keenan’s fieldwork on Malagasy speakers who systematically withhold information without generating the implicatures an English speaker would compute, and the wider tradition of cross-cultural pragmatics — establishes that their weighting differs across speech communities. A novelist representing a community whose norms of directness differ from the reader’s must establish the local baseline before deviation from it can signify. Implicature requires a norm, and the novelist must supply the norm before flouting it.

Politeness competes with informativeness. Penelope Brown and Stephen Levinson’s Politeness (1987) argues that much apparent inefficiency in talk — hedging, indirection, circumlocution before a request — is the rational management of face, and is therefore not a flouting but the observance of a different imperative. A character who hedges may be concealing (implicature) or may be being polite (face-work), and the two produce different reader experiences. §3.3 develops the face side.

3.2 Conversation analysis: the machinery of turns

Grice describes what talk means; conversation analysis describes how it is built. The founding paper is Harvey Sacks, Emanuel Schegloff and Gail Jefferson’s “A Simplest Systematics for the Organization of Turn-Taking for Conversation” (Language, 1974), and its findings are among the most robust in the human sciences — derived from recordings, replicated across languages, specific enough to be falsifiable. Four bear directly on writing.

1. Turn-taking is locally managed and precise. Speakers do not wait for a pause; they project the end of the current turn and begin at or just before it. Overlap is common but brief; gaps are measured in fractions of a second. The implication for fiction is negative but useful: the rhythm of real conversation is far faster and more interlocking than alternating paragraphs suggest, so a writer wanting speed must produce it by other means — shorter turns, dropped attributions, interruption marked typographically.

2. The adjacency pair is the basic sequence. Certain first turns make a particular second conditionally relevant: a question makes an answer relevant, an invitation an acceptance or refusal. Conditional relevance is the key concept, because it means the absence of the expected second turn is itself an event — a noticeable absence, heard as meaning something. This is the machinery underlying every evasion in fiction, and it is why a character’s failure to answer is among the most economical moves in the form: nothing has to be written for a great deal to be communicated.

3. Preference organisation: refusals are marked, and the marking is regular. This is the finding that should most alter how writers draft. Anita Pomerantz’s work on assessments, and the wider conversation-analytic literature, establishes that the two possible seconds of an adjacency pair are not symmetrical. Preferred responses — acceptances, agreements — come immediately, briefly and without explanation. Dispreferred responses come with a characteristic package: a delay; a hesitation marker (Well —); a token agreement (That’s a lovely idea, but); an account offered without being asked for; mitigation (I don’t think I can, quite); and often no outright refusal at all (I’ll see).

This package is how skilled novelists write reluctance. The delayed, hedged “no” that a reader registers as psychologically acute is not an intuition of the writer’s; it is a documented feature of talk the good writer has absorbed by ear. Conversely, a draft in which characters refuse each other flatly — “No.” “Then I’ll go alone.” — reads as stylised or hostile, and writers rarely know why: they are producing preferred-format turns with dispreferred content. The teaching move follows. A character who agrees instantly is agreeing; a character who agrees after a pause and an account has refused something.

4. Repair. Speakers correct themselves and each other continuously, and repair has a strong preference for self-correction: given the chance, speakers fix their own errors, and hearers give them that chance. Other-correction is rare and socially marked — which is why correcting another character’s name, title, date or word choice is one of the sharpest status moves available in dialogue, and one of the cheapest to write.

A caution. Conversation analysis is a descriptive science of naturally occurring talk. The claim here is not that novelists should reproduce these structures — §3.6 argues emphatically that they should not — but that these structures are what readers have internalised, and therefore constitute the grammar against which a novelist’s stylisations are read.

3.3 Goffman: footing, participation, and face

If Grice supplies the mechanism of subtext and conversation analysis the mechanism of sequence, Erving Goffman supplies the scholarly form of what this chapter calls the relational domain (§4.3).

Face-work. In The Presentation of Self in Everyday Life (1959) and “On Face-Work” (in Interaction Ritual, 1967), Goffman describes interaction as the mutual maintenance of face — the social value a person claims by the line others assume they are taking. Interaction is organised to protect the face of all participants, including that of the person one is opposing, which explains the elaborate mitigation, the offering of exits, and the willingness to accept a transparently false excuse. It also identifies the moment an exchange turns. An exchange turns when one party stops protecting the other’s face.

That formulation gives a writer something to look for. In most drafted arguments the escalation is a matter of volume; in most published arguments that work, it is a matter of face — a name used in full, an excuse refused, a euphemism replaced by the accurate word, a courtesy withheld.

Footing. In Forms of Talk (1981), Goffman dismantles the two-role model of speaker and hearer. A speaker may be the animator (producing the sounds), the author (composing the words), or the principal (whose position the words express), and these come apart: a clerk delivering a company’s refusal is animator but not principal. Footing is Goffman’s term for a speaker’s alignment to what is said, to the other participants and to their own role — and for the fact that it shifts, often within a single turn. A character who moves from I think to people say to she always used to say has changed footing twice, and each change alters who is answerable. This is not a subtle literary effect but something people do continuously, and one of the most under-exploited resources in draft dialogue. The Third-Person Deflection (§6.4) is a footing shift used to shed accountability.

The participation framework. Goffman’s third contribution is that talk has more parties than the two speaking. He distinguishes ratified participants — those the talk is officially addressed to, whether addressed or unaddressed — from bystanders: overhearers, known to be present, and eavesdroppers, not. Every utterance is designed with reference to the whole framework.

The literary application is large enough that §4.6 builds a proposed domain on it. Consider how much great dialogue is designed for a third party: Elizabeth Bennet speaking to Darcy in a room containing Miss Bingley; Nelly Dean listening to Catherine while Heathcliff sits unseen; any exchange conducted in a servant’s presence, where the servant’s status as unratified participant is what makes the room usable. Who is in the room but not in the conversation is a structural fact about the exchange, not a detail of the setting.

3.4 Bakhtin: the novel as a contest of languages

Bakhtin supplies the largest frame, and the one that underwrites the prohibition against characters who all sound the same.

His argument, developed across The Dialogic Imagination (English translation 1981) and Problems of Dostoevsky’s Poetics (English translation 1984), is that the novel is constitutively different from poetry and epic in its relation to language. Poetry aspires to a unitary language; the novel is built out of heteroglossia — the actual stratification of a national language into the discourses of professions, generations, classes, regions and genres, each carrying its own evaluative charge. The novelist does not write in a language but orchestrates languages, and the book’s meaning arises from the friction between them.

This is the scholarly ground for voice differentiation. The craft instruction that characters must sound different is usually justified on grounds of clarity. Bakhtin’s justification is stronger: a character’s speech carries a whole social position — the words they reach for, the register they default to, what they consider sayable, the authorities they cite. Two characters who sound the same occupy the same social world, and a novel in which everyone occupies the same social world has forfeited the form’s central resource. This is the argument against the ventriloquist exchange (§8.2) in its strongest form, and it does not depend on realism at all.

Polyphony. In the Dostoevsky book Bakhtin distinguishes the monologic novel, where all voices are finally subordinated to the author’s evaluative position, from the polyphonic novel, where characters’ ideological positions retain real independence. Ivan Karamazov’s argument against God is not defeated by the narrator; it is answered, if at all, by another character’s life. This is the basis for the theme domain’s requirement (§4.4) that a staged dialectic be open — both positions given their best case, the narration declining to arbitrate.

Bakhtin also supplies a term for something writers produce without naming: the character zone, the field around a character in which the narration itself becomes contaminated by that character’s speech habits, so that the boundary between narration and speech goes porous. Anglo-American criticism approaches the same phenomenon through free indirect style, which is why §2.4 lists free indirect speech as a boundary case rather than an exclusion.

3.5 The historical strand: quotation is a convention with a date

Students routinely receive the contemporary conventions of dialogue presentation as though they were properties of narrative. They are properties of a period.

Four observations follow, each a caution against universalising the present.

1. The typographic apparatus is not old. M. B. Parkes’s history of punctuation, Pause and Effect (1993), traces the slow emergence of the marks now used to signal quotation, which stabilise into their modern function only in the eighteenth and nineteenth centuries. A convention two centuries old is not a law of the form.

2. The preference for said dates from roughly the mid-twentieth century. The nineteenth-century novel is full of rejoined, ventured and expostulated, used by writers of the first rank without embarrassment. The modern proscription — codified for a general readership by Elmore Leonard’s rules, but current in teaching for decades before — rests on the premise that attribution should be invisible so the reader’s attention rests on the speech. That premise is defensible and this chapter largely endorses it (§7.1, test 9). It is not timeless, and a student who reads he expostulated as an error rather than a period convention has mistaken a fashion for a fault.

3. Austen’s practice would fail a contemporary workshop. Much of Pride and Prejudice’s social interaction is not rendered as dialogue at all but as summary, indirect report and free indirect discourse, with direct speech reserved for what matters. Norman Page’s Speech in the English Novel (1973) — the standard study, and the most useful single book for a student of this chapter — documents how varied and deliberate the nineteenth-century handling of the speech/report boundary was. Contemporary practice over-uses direct speech, because it has been taught that showing beats telling. Austen’s ratio is a technique, not an archaism; §9.1 shows what it buys.

4. The abolition of quotation marks is a formal contract, not a tic. Saramago’s Blindness (1995; trans. Giovanni Pontiero) runs dialogue into continuous paragraphs, marking speaker changes with a comma and a capital letter. Chapter 9 (The Opening Pages) treats this as a formal contract declared on page one. What this chapter adds is that the device has a semantic consequence: without the visual separation of quotation marks, speech loses its ontological privilege over narration, and the reader stops experiencing speech as a distinct kind of material. In a novel about a society dissolving into undifferentiated crisis, that is not a typographic preference. §9.6 develops the reading.

3.6 The empirical caution: transcribed talk is not dialogue

Work in discourse analysis and corpus linguistics — Wallace Chafe’s studies of spoken and written language, Jefferson’s transcription conventions, and the spoken components of the large reference corpora — establishes that spontaneous conversation has pervasive features that novelistic dialogue almost entirely lacks.

The conclusion is not that novelists are inaccurate. It is that literary dialogue is a stylised convention that reads as real, in the way stage lighting reads as daylight. The illusion is produced not by reproducing surface features but by retaining a small number of structural ones — the preference organisation of refusals, the conditional relevance of adjacency pairs, shifts of footing, the flouting of maxims — which are the features that carry social meaning. A writer who imports the surface features instead produces something that sounds less real, because disfluencies signify differently on the page: on a recording um is invisible; on a page it is a mark, and per the rules of notice discussed in Chapter 9 (The Opening Pages) every mark on a page is read as intended.

Deborah Tannen’s Talking Voices (1989) supplies the final turn of the argument. Studying how speakers report others’ speech in ordinary conversation, Tannen found such reports are rarely verbatim: speakers invent, compress and dramatise, producing what she calls constructed dialogue. The stylisation novelists perform on speech is a stylisation ordinary speakers also perform on speech. The novelist’s dialogue is not a departure from how people talk but a formalisation of what people already do when they retell talk.

3.7 The first principle, restated and qualified

A dialogue exchange earns its place when speech is doing something that could not be done otherwise: when the words are the instrument of a want, when the balance of information, standing or feeling differs at the close from the open, and when what is said and what is meant do not entirely coincide.

Four qualifications, each of which the rest of the chapter honours.

1. Not every exchange must run on two channels. The gap between text and subtext is a resource with a limit, and §6.5 argues that closing it is a climactic move spendable once. A book in which every line is oblique is as monotonous as one in which every line is direct, and more irritating.
2. The subtext requirement is culturally and historically local — a twentieth-century Anglo-American ideal, shaped by Chekhov’s reception, by Hemingway, and by the workshop system that codified them. §16 develops this at length and should be read as part of the argument, not as a disclaimer appended to it.
3. The tests are diagnostic, not evaluative. The protocol in §7 identifies what an exchange is doing, not whether it is good. A writer who converts its outputs into verdicts will cut the most interesting material in the draft, because the most interesting material is where the conventions are being strained.
4. Realism is not the standard, but plausibility is a constraint. Rejecting transcriptional accuracy does not license dialogue no human register would produce. What must remain true is the social grammar: that speakers manage face, that refusals are marked, that people design their utterances for everyone in the room. A character may speak in perfect epigrams — Wilde’s do — provided the epigrams still do the social work talk does.