Reddy’s conduit metaphor, run as an experiment
Three people and one crossing. Two describe what they need; the third builds from their words and may ask nothing. Nobody but the builder may name a part — so a need has to be rebuilt out of what it would be like to live with it.
340 conversations, judged against what they were for. Each person is given an aim in plain words; the need behind it is scored on the crossing that ended up standing, and shown to nobody.
Do they end up where they started?
of what a speaker meant on its first turn is still in what it means on its last. More of it than that survives into the thing itself — 57% of what they first asked for is in the crossing that finally stood.
Partly. What they say moves further than what they get.
Does the task survive being misunderstood?
Runs where the builder caught a third or less of what was meant served 44% of needs. The twenty runs where it caught most of it served 31%. Being understood sentence by sentence does not predict getting what you needed — though twenty runs is a thin end to lean on.
Yes, and that is still the surprise.
Does it serve both of them?
of conversations leave both people at least half served. In 47% one of the two is sacrificed for the other, and in 22% neither is.
Usually not. Half the time somebody pays for the other’s crossing.
Talking is worth something. A need reaches the built crossing more often than chance would put it there — 41.7% against a 35.3% floor, +6.4 points, p < 0.0001 — and the instrument contains its own control: one person alone with the builder gets 32.8% of what they need, where two people in any arrangement get 42.9%. It holds best for the same aim, unknown to both (67.6%) and worst for two references (23.5%), where each is gesturing at a crossing the other has never seen. The floor is measured differently from before: every argument now carries one stated aim per role, so shuffling within an argument has nothing to permute and reports no effect whatever the truth. It is shuffled across the whole set instead.
340 conversations. What each person said they meant is asked afterwards, turn by turn, and never shown to anybody — so an idea can be followed from the mouth it left to the crossing that ended up standing.
After they have spoken to somebody, which words do they stop using?
of the ideas a person opens with are never said again. 78% of everyone who spoke more than once abandoned at least one of them. The ones most often dropped are where they are — the ground is water (75 times) — and what they are afraid of: it goes slick (65), it moves underfoot (50).
People stop describing their situation almost at once, and never return to it.
What replaces them?
times a person took up something at the hand having not asked for it at the start — more than any other idea. Then takes loads (139), does not go slick (139), ends hold in soft ground (131). The commonest exchanges: it goes slick → takes loads (32), the ground is water → something at the hand (30), the ground is water → it goes slick (27).
A condition becomes a fitting. They arrive describing a place and leave asking for parts — which is the one thing they were told they could not name.
Does going through somebody else change the size of the crossing?
One person alone with the builder gets 5.2 matchsticks. Put anybody else in the room and it is 13.1 — and the builder rebuilds 2.4 times instead of once. The order is almost monotone: straight to the builder 4.5 and 5.8, relayed through the builder 7.3 and 8.7, through the other person 9.3, all three talking 11.7, everything heard 14.2, and 17.1 when it waits and builds once at the end.
Yes, and in one direction. Every additional party makes a larger crossing, whatever anybody asked for.
The three answers are the same answer. What a person can say — a place, a fear — drops out of the conversation; what a builder can act on takes its place; and the thing that gets built grows with every mouth the meaning passes through. Rebuild counts are not comparable across cases: five of the seventeen receive only one or two lines, or hold off and build once at the end, so their “1.0” is a property of the route and not a finding. Matchsticks are comparable, and that is the figure above.
Nothing below is scripted. What each of them meant is asked afterwards and never shown to the builder.
+ one who needs to cross · − the other, who needs something else of it · ○ the builder · boxes are rooms.
Role 3 builds from a workshop of parts, with no way to ask anything: it gets sentences, and out of those and its own materials makes what it can. Reddy’s toolmaker rather than a device — the limit is what is to hand.