↩ SocioRhetoric · A New Catalogue of Speech Acts
sociorhetoric · a speech act
Why is the phrase that binds a group so often the one that costs outsiders something to say — awkward, extreme, or embarrassing? A signal that is cheap to fake tells you nothing; belonging is proved by a cost only a real member will pay.
In the Book of Judges, the men of Gilead hold the fords of the Jordan and stop each traveller with a single test: say the word shibboleth. The word itself means an ear of grain — but its meaning is beside the point. What the guards are listening for is a sound the fleeing Ephraimites cannot make; their dialect has no sh, so they say sibboleth and are killed on the spot. Forty-two thousand die, the text says, on a phoneme. The password does no informing whatsoever. It works because it is free for the right mouth and expensive for the wrong one — and expense, not content, is the whole of the signal.
Economists have a name for this shape. A signal separates two kinds of people only when it is cheap enough for the genuine and dear enough for the faker — Michael Spence's insight about why a degree can sort workers even if it teaches nothing, and Amotz Zahavi's handicap principle about why a gazelle wastes energy leaping in front of a predator. Make the marker costless and everyone adopts it, so it tells you nothing about who is real. Make it costly in exactly the right band and only members will bear it, so its mere presence becomes proof. The instrument below is a working costly-signalling model. It never reads the words. It only ever weighs their price.
the honest instrument
Two kinds of people can say the in-group phrase: a genuine member (cheap for them) and an outsider trying to pass (expensive — the phrase is awkward, or means abandoning another allegiance). A receiver rewards anyone who signals with a membership benefit B. Each type says the phrase only if that benefit beats its own cost. The displayed values — who signals, the receiver's belief that a signaller is real, the information the signal carries, the separating band — are computed live from that rule. The literal phrase is not an input to the model.
The shibboleth model
two types · one signal · a single-crossing equilibrium
the phrase on trial
“Let’s go Brandon”
Literal content is not a model input. Scramble it and every number holds.
what it costs a real member to say the phrase
the handicap — awkwardness, or a rival allegiance given up
the reward the receiver grants a perceived member
the base rate of real members in the crowd
the phrase that costs
Slide the cost to faker up and watch the receiver's belief snap from the base rate to certainty. Nothing about the words changed; what changed is who is willing to say them. This is the strange arithmetic of a shibboleth: the more embarrassing, extreme, or costly the phrase, the better it works as a marker — because embarrassment is precisely what an outsider will not pay. A slogan that is a little too much, a in-joke that would mortify a normal person, a loyalty oath that burns a bridge to some other camp: these are not failures of taste. They are the handicap doing its job. The phrase is calibrated to be just bearable for a member and just unbearable for a pretender.
Which is why the literal proposition drops out entirely. Ask what “Let’s go Brandon” asserts about the world and you have already misunderstood the act; its truth-value is not in play. The model makes this concrete by refusing to store the phrase at all — the only inputs are two costs, a benefit, and a base rate. Press scramble the words and the sentence becomes nonsense, a rival faction's creed, a made-up password; every readout holds to the last digit. When a marker is oriented to belonging rather than reality, meaning is ceremonial and cost is everything.
what to try
Start from the shibboleth and drag cost to faker down below the benefit B. Watch the outsider start signalling too, the posterior collapse from 100% to the base rate, and the information line fall to zero bits. A cheap phrase is one everyone can say — so saying it proves nothing.
Now push the faker's cost back up past B. The outsider goes silent, the posterior jumps to certainty, and the phrase becomes a reliable badge again. The whole separating band is the stretch of faker-cost above B; the chart shades it, and the readout names its lower edge.
Press scramble the words and watch nothing move. Then raise cost to genuine above B: even real members fall silent and the crowd goes quiet — the third regime, where the phrase is too dear for anyone to bear.
separation lives in the cost gap
Spence and Zahavi share one structural idea: single crossing. A signal sorts types only if the same act costs the two types differently, so that there is a price the genuine will pay and the faker will not. Fix the member's cost and the benefit; then sweep the faker's cost, as the chart does. Two thresholds carve the line into regimes. Below the benefit B, the outsider finds the phrase worth faking, so everyone signals — the pooling-loud equilibrium, where the marker is universal and therefore worthless, its information exactly zero and the receiver left with nothing but the base rate. Above B, the outsider abstains and only members speak — the separating equilibrium, where a signaller is genuine with certainty and the signal carries its full bit.
There is a third regime the sliders can reach from the other side. Raise the member's cost above B and even the genuine fall silent — pooling-silent, no one signals, and no amount of faker-cost can conjure a badge out of a phrase nobody will say. So the credible shibboleth lives in a band: costly enough that fakers won't pay, cheap enough that members will. Too cheap and it is noise; too dear and it is silence; the signal is honest only in the gap between. This is not a metaphor laid over the numbers — it is the numbers. The posterior is a step function with its riser at exactly cf = B, and the information value is the binary entropy of the prior, earned in full the instant separation begins.
model
There are two types — genuine with prior P(g), faker with 1 − P(g) — and one available act, the signal, which costs cg for a member and cf for an outsider. A receiver who takes a signaller for a member confers benefit B. The behavioural rule is the whole model: a type signals if and only if B ≥ its own cost. That yields the genuine's best response (signal when B ≥ cg) and the faker's (signal when B ≥ cf), and their four combinations are the four regimes the tags report.
From those two switches, two quantities follow by definition. The posterior on a signaller is Bayes' rule on the signalling strategies — P(g)·sg / [ P(g)·sg + (1−P(g))·sf ] — which is 1 when only the genuine signal, the flat prior when both do, and undefined when neither does. The information value is the mutual information between type and the send/stay-silent decision, in bits: it equals the binary entropy of the prior under perfect separation (a full bit at an even prior) and drops to exactly zero under either pooling. Everything recomputes on every drag. Treat it as a toy model that isolates one mechanism — costly separation — not as a full account of how real slogans spread, which the caveats below insist on.
the act ↔ the model
| The act | The model |
|---|---|
| a type | which kind is speaking — a genuine member for whom the phrase is cheap, or an outsider for whom it is dear. |
| the signal | saying the in-group phrase; a single act with no content the model ever reads, only a price it charges. |
| cg vs cf | why the phrase is a handicap: the same words cost the two types differently, and the gap is what does the sorting. |
| separating equilibrium | only members will pay, so a signaller is believed — the phrase is a credible badge, posterior on a signaller ≈ 1. |
| pooling-loud | a cheap phrase everyone can mimic; the marker goes universal, the posterior falls to the base rate, and it means nothing. |
| the content | irrelevant once cost does the work — not an input, which is why scrambling the words leaves the equilibrium untouched. |
how this act misfires
misfire · orientation mismatch
The catalogue's named misfire for this act. A tribal signal is keyed to belonging, not reality; its truth-value has dropped out and the act announces membership instead. The mismatch is what happens when a listener parses it literally — takes “Let’s go Brandon” as a proposition to be fact-checked, or answers a loyalty chant with a rebuttal. They are reading coordinate one, orientation-to-reality, when the speaker is playing coordinate two, orientation-to-standing. The outsider who argues with the content has already lost the thread: there was no content to argue with, only a price being paid in public.
misfire · the badge goes cheap
A marker is only a badge while it is costly. Let the phrase spread — let it become safe, quoted, merchandised — and the faker's cost falls until everyone can say it. The model calls this the pooling-loud regime: the posterior slides from certainty back to the base rate, the information drains to zero, and yesterday's password becomes today's noise. This is the same drift the catalogue tracks as Surrogation — a token outliving the substance it once indexed — seen from the signalling side. A group that wants its badge to keep working must keep it expensive, which is why in-groups so reliably escalate: the moment a marker is cheap enough to be comfortable, it no longer proves anything.
A badge that costs nothing to fake is worn by everyone and means nothing; belonging is a price only members pay.
can you use it?
RECOGNITION — Which is a working tribal signal? A: “turnout in the third district rose four points.” B: a fan bellows the club's mortifying victory chant at the rally. C: the same chant, now printed on mall t-shirts everyone wears.
B. The chant costs an outsider embarrassment a member pays gladly, so voicing it certifies membership. A answers to reality; C has gone pooling-loud — universal, costless, and proof of nothing.
THE NEAREST NEIGHBOR — A listener fact-checks the chant and rebuts it. What single coordinate did they misread?
Orientation. The signal is keyed to belonging, so its truth-value has dropped out; the act charges a price and asserts no proposition. Arguing with the content answers an axis the speaker never played.
PRODUCTION — Design a membership phrase for a small hobbyist forum. Compose one, then check it against the band.
An in-joke a normal person would be embarrassed to type — cheap for regulars, dear for pretenders. Yours works if cg ≤ B < cf: members will pay, fakers won't, and the words themselves can be nonsense.