1.1 An asymmetry between accuracy and function
Set two passages side by side.
The first is a verbatim transcript of ninety seconds of recorded conversation between two people who know each other well. It almost certainly contains a stretch where both speak at once; several false starts abandoned mid-word; three or four instances of um, you know, I mean; a self-correction; a question repeated because the first version was misheard; a long run of minimal responses (mm, yeah, right) from the listener; and a topic raised and dropped without resolution. It is entirely accurate. It is also, as prose, close to unreadable, and it discloses less about its two speakers in ninety seconds than a competent novelist would disclose in five lines.
The second is a passage of novelistic dialogue, containing almost none of those features. Its speakers do not overlap. They finish their sentences, or break off at points the writer has chosen for effect rather than for the reasons real speakers break off. Every turn advances something. It is, by any empirical standard, an inaccurate representation of how people talk — and readers experience it as lifelike.
That asymmetry is the chapter’s starting point. In no other domain of fiction is the gap between the represented thing and the representation so wide while the illusion of fidelity remains so strong. Descriptions of rooms are not required to be accurate inventories, and no reader expects them to be. Dialogue alone is habitually defended on the grounds that people really do talk this way.
They do. They also produce, in life, enormous quantities of talk that would sink a manuscript: greeting rituals, weather, scheduling, agreement, repetition, the negotiation of who speaks next. Transcribed accuracy is not the standard. What a reader registers as realism in dialogue is not fidelity to the acoustic record but fidelity to a set of social facts about talk — that people want things from each other, that they do not say so directly, that saying and meaning come apart, that every turn positions its speaker relative to the last. Novelistic dialogue is a stylised convention that reproduces those social facts while discarding almost all the surface features of actual speech. §3.6 gives the empirical detail; the practical consequence can be stated at once.
The realism of dialogue is a realism of pressure, not of transcription. A reader who says an exchange “sounds real” is reporting that the wants underneath it are legible, not that the words match a recording.
1.2 The claim in its strong form
Two characters trading lines are a transcript. A dialogue exchange is a transcript with an engine under it: each speaker wants something from the other, wants it now, and the words are the instrument each is using to get it.
The metaphor of the engine will be used throughout, and what it does and does not claim needs stating carefully.
- It does claim that a passage of speech without a want on at least one side is inert regardless of how good the individual lines are, and that this is testable (§7.1, test 1).
- It does not claim that the want must be large. I want you to admit you were wrong. I want to get out of this room without answering. I want you to say my name. These are sufficient. Chapter 1 (The Purposes of a Scene) makes the same point about scene goals, and the argument transfers unchanged: the goal must be legible and resisted, not important.
- It does not claim that the want must be conscious. A speaker may not know what they are after; the reader must be able to infer it. Indeed the gap between what a character is pursuing and what they think they are pursuing is one of the richest resources in the form, and §6 is largely about how it is engineered.
- It does not claim that all speech in a novel must be an exchange in this sense. Ritual talk, phatic exchange and small talk have legitimate structural uses — establishing register, marking the passage of time, providing the neutral ground against which a deviation registers. What they cannot do is carry a scene.
1.3 Where this chapter sits in the series
Chapter 1 (The Purposes of a Scene) established the value-shift test — the demand that something charged be measurably different at a unit’s close than at its open — and the domains of plot, character, theme and texture in which such a shift can occur, to which it proposed a fifth. This chapter applies the test at a smaller scale and adopts that fifth domain, the relational: the negotiation of standing between speakers, which the usual four-domain model folds into character and which dialogue makes too load-bearing to leave there. §4.6 then proposes a sixth.
Three relations to neighbouring chapters should be held in view:
- Dialogue is a mode of scene, not an alternative to it. Everything in Chapter 1 (The Purposes of a Scene) applies. An exchange that fails the value-shift test fails it for the same reasons a scene does, and the remedies — cut, compress to summary, fuse with a neighbouring unit — are the same.
- Dialogue’s two channels have a descriptive analogue. Chapter 4 (The Descriptive Detail) develops the pair denotation and connotation: what a detail names, and what it means. Text and subtext are the same structure in the speech channel, and the two chapters should be read together by any student who finds either pair abstract.
- Who reports the speech is a viewpoint question. All dialogue in a novel reaches the reader through a narrator, and the narrator’s position determines what the reader is permitted to know about the gap between saying and meaning. Chapter 13 (The Viewpoint Shift) handles the machinery; §9.5 below, on Wuthering Heights, shows what happens when dialogue arrives at two removes.