↩ SocioRhetoric · A New Catalogue of Speech Acts
sociorhetoric · a speech act
framed misfire → manipulation · the effect depends on not being disclosed
How can every word you say be true and the impression you leave be false? A palter tells no lie — it engineers a true statement whose pragmatic inference is wrong, and it works precisely because the listener reasons about why you chose those words.
In 2011 Lance Armstrong looked into a camera and said he had taken “500 drug controls worldwide… never a failed test.” Every clause was literally, verifiably true — he really was tested that often, and he really never failed one. Years later, confessing at last to systematic doping, he did not even flinch at the memory: “that’s not a lie, that is the truth.” And it was. The sentence was true. What it manufactured — the belief that he was clean — was false. That split, between the truth of the words and the falsehood of the impression, is the whole of the palter.
The palter is the specimen that forces this catalogue’s third dimension, Motive, apart from its second, Orientation. A lie inverts the map: it is keyed to reality and reports the opposite of it. A palter never touches the map at all — it is keyed to reality, every word answerable to the world, and yet wholly manipulative. So the honest reader’s reflex, is it true?, sails right past it; orientation-checking cannot catch a statement that is oriented correctly. The catch has to happen one level up, in the listener’s reasoning about the speaker — which is exactly where the palter does its work, and exactly where, as the instrument below shows, it can be undone.
the honest instrument
This is a live Rational Speech Acts model (Frank & Goodman 2012). A hidden world state is drawn from a small set; the speaker picks an utterance with a literal truth-value in each state — the truth table below, which you can edit. Click a cell to flip its truth. Every distribution to follow — literal listener L0, pragmatic speaker S1, pragmatic listener L1, and the cynic L2 — is computed from these numbers on every change.
The safety palter
2 states · 3 utterances · α = 1.0
The truth table — click any T / F cell to flip it
Live distributions
true words, false impression
Four of this catalogue’s five dimensions would happily collapse the palter into something they already know. Function calls it an assertion and moves on. Orientation checks it against the world, finds every word answerable to reality, and stamps it true. Only Motive can see the act for what it is — because Motive asks a question the others cannot: would this effect survive being disclosed? Say the quiet part aloud — “I’m telling you I passed 500 tests so that you’ll conclude I’m clean, which I am not” — and the manoeuvre dies on contact. That dependence on non-disclosure is the signature of manipulation, and it is invisible to any check that only asks whether the sentence is true.
The catalogue therefore keeps Motive and Orientation separate. A lie is keyed to deception and serves a hidden interest — the two dimensions agree, so you could imagine reading either one and catching it. The palter is the counter-example that breaks the shortcut: keyed to reality on the Orientation axis, manipulative on the Motive axis, the two coordinates disagree. An act can be perfectly oriented and thoroughly crooked at once. Keep the axes separate and the palter has a home; merge them and it becomes literally unspeakable — a true statement that somehow deceives, with no coordinate left to name the crookedness.
A lie inverts the map. A palter leaves the map alone and edits you.
what to try
Start on the safety palter. The speaker, actually in the unsafe world, says the literally-true 500 tests, never failed
— the truth badge confirms every word holds. Yet the naive listener’s belief in safe climbs to 66.7%, well above the 50% the words literally license. A true sentence has manufactured a false belief. That +16.7 gap is the palter.
Now flip the listener to cynical. The same true sentence, the same speaker — but the posterior on safe snaps back to 50%. Modelling the motive drains the true statement of its false pull. The palter “depends on not being disclosed”: disclose the incentive and it dissolves.
Flip the speaker to cooperative instead. Now the speaker in the unsafe world stops paltering and discloses — it failed a safety review
— and the listener lands on the truth. Remove the manipulative goal and the very same model becomes honest and informative.
Add an utterance and set it true only in the safe column (click the unsafe cell to F) — a flat it is certified safe.
Watch the palter collapse toward 50%: once the honest speaker has a strong safe claim available, the listener expects it, and the weak true statement stops implying safety.
the listener does the lying to themselves
The speaker never asserts the falsehood. The belief that Armstrong was clean was manufactured inside the listener, by a piece of reasoning the listener is normally right to trust. A cooperative speaker who knew the product was unsafe would warn you — that is what the disclosure norm, Grice’s maxim of quantity, leads you to expect. So when the speaker offers only it passed 500 tests
and no warning, you reason backwards: if there were a problem, a decent person would have said so; they didn’t; so there probably isn’t. The inference is valid — against a cooperative speaker. The palterer’s entire trick is to borrow that assumption while violating it.
That is why the fix is not more scrutiny of the words — they will pass every time — but a change in the model of the speaker. The cynical listener in the instrument is a higher-order reasoner: it knows the speaker may be strategically selecting true-but-flattering statements, so it stops treating the absence of bad news as evidence of good news. Against that listener, the palter carries no information — belief falls back to the prior. This is the catalogue’s thesis about manipulation made mechanical: the effect lives entirely in the gap between how the listener models the speaker and how the speaker actually behaves. Close the gap and there is nothing left to exploit.
model
The engine is textbook Rational Speech Acts, three nested layers, each a normalised probability distribution. Write [[u]](s) for the truth-value of utterance u in state s (the 1s and 0s in the table), P(s) for the prior, cost(u) for how effortful the utterance is, and α for how sharply the speaker optimises:
Each line is normalised so its distribution sums to one; that is all the ∝ hides. The naive listener in the instrument is exactly this L1 — and because L1 assumes S1 is cooperative, it is the one the palter fools. The cynical listener is a second speaker model, Smanip, whose objective is not to convey the true state but to maximise belief in the flattering one: it scores each true utterance by ln L0(goal | u) instead of ln L0(s | u). Invert that speaker — L2(s | u) ∝ Smanip(u | s) · P(s) — and you get a listener who has priced in the manipulation. The manipulative speaker toggle chooses, among utterances literally true in the actual state, the one that most raises the naive L1’s belief in the goal state; the cooperative speaker toggle instead speaks the most informative true utterance. Everything you see — L0, S1, L1, L2, and the literal-versus-inferred gap — is these formulas evaluated on your current table, live in the browser.
the act ↔ the model
| The act | The model |
|---|---|
| a world state | the fact actually in question — is the product safe, is the athlete clean — hidden from the listener, weighted by the prior. |
| an utterance’s truth-value | whether the words are literally true in that state — the 1s and 0s you can flip. A palter keeps every cell in the spoken row set to true. |
| the pragmatic speaker S1 | choosing words for their effect on a listener as well as their truth — the act of selection that carries the manipulation. |
| the pragmatic listener L1 | the impression formed by inference from that selection — belief reasoned out of why these words and not others. |
| the literal-vs-inferred gap | the palter itself: the distance between what the words entail (L0) and what the listener comes to believe (L1). |
| the cynical listener L2 | modelling the motive — a higher-order listener who assumes strategic word-choice, and so is not moved. The disclosure that disarms the act. |
how this act misfires
misfire · manipulation
This is the catalogue’s named risk for the Motive dimension, and the instrument computes it directly. A palter’s effect is its non-disclosure: the moment the listener models the speaker as someone who selects true-but-misleading statements — the cynical L2 — the true sentence stops carrying its false implication and belief collapses back to the prior. The very rationality the palterer exploited re-derives the truth. Say the quiet part aloud, or address a listener who already suspects the game, and there is nothing left to steal.
caveat · stylised rationality
Real listeners are not consistently cynical; they flip between naive and sophisticated within a single conversation, and no one runs a clean softmax with a stable α. So read the exact posteriors — 66.7%, 50% — as illustrative, not measured. What the model gets right is structural and robust to the knobs: a statement can be entirely true and still drive a belief its literal content does not license, because listeners reason about utterance choice; and that same reasoning, aimed one level higher, is the defence. The numbers are a stylisation; the shape — true words, false belief, dissolved by disclosure — is the honest point.
A palter tells no lie — it lets you deceive yourself with a true sentence.
can you use it?
RECOGNITION — A doped athlete says: A: “I've never failed a drug test.” B: “I've never taken a banned substance.” A clean athlete says: C: “I've never failed a drug test.” Which is the palter?
A. Every word survives fact-checking while the impression it engineers — clean — is false. B is false in the actual world, a lie; C opens no gap between literal content and inferred belief.
THE NEAREST NEIGHBOR — One criterion separates the palter from the lie. Name it.
Where the falsehood lives. A lie puts it in the sentence, false in the actual world; a palter keeps every word true and lets the listener's own inference manufacture the false belief.
PRODUCTION — Your product failed a safety review. A customer asks whether it is safe. Compose the palter, then check it.
“It has passed five hundred bench tests without a single failure.” Yours works if every clause is true, the customer leaves believing “safe,” and saying your reason for choosing those words aloud would kill the effect.