communicoupling
At the MIT Radiation Laboratory, physicists who thought in field equations had to work with engineers who thought in circuits and tolerances. Neither converted the other; conversion was never on the table. Instead, Galison observed, they did what neighboring cultures with no common tongue have always done at their borders: they built a trading zone — a thin, local, jointly owned language, rich enough to coordinate the exchange and silent about everything else. Agreement on use came first; agreement on meaning never came, and never needed to.
This is the working answer to incommensurability. Kuhn worried that paradigms cannot fully translate; the radar builders showed that full translation is not what cooperation runs on. Below, two communities with disjoint native languages meet at a border. Watch a pidgin assemble itself out of failed exchanges — and compare it with the expensive alternative everyone proposes first: merging the languages.
the mechanism
Each community speaks a full native language over its own concepts; six concepts are trade-relevant — the overlap of needs. Each round, every boundary pair attempts an exchange requiring two of those concepts; a concept goes through only if the pair holds a shared token for it, and when one is missing the exchange fails — but with probability λ the pair coins a token on the spot, and the border group keeps it. That is the entire machine. Run it and the pidgin assembles along a saturating curve: fast at first, when every failure meets an uncoined concept, then done — with these dials, full trade coverage by about round eight, and the shared vocabulary stops growing at exactly 6 / 6. Nothing enforces the stopping. The pidgin is thin because coining only happens where trade fails, and trade only fails where trade happens.
The merger strategy answers the same problem the way a reorganization would: converge everyone — both whole communities, all concepts — on one common tongue. The instrument runs it honestly and the price appears on two curves at once. Trade climbs slowly (an exchange needs both its concepts already merged, so early progress compounds against itself), taking on the order of a hundred rounds to become reliable. And internal efficiency — each community talking to itself — sags through the transition as half-converted speakers mismatch their own colleagues, bottoming near 69%, then settles permanently below where it started, because the merged tongue coarsens each side's specialist distinctions. The pidgin bought the same trade in a week, for free.
what to try
Set boundary agents to 0: no contact, no pidgin, no trade, forever — proximity alone builds nothing. Then set b = 1: one interpreter per side. The pidgin still completes (around round 17), and every exchange now flows through one pair — count the absolute throughput and you have measured the interpreter bottleneck.
Run the border market for 200 rounds, note all three curves, then run the merger. Same eventual trade — reached in ~8 rounds versus ~100, with internal efficiency dipping to 69% along the way and never recovering fully. The merger buys nothing the pidgin didn't, and pays with both languages.
Press creole drift: at round 150 the trade widens from six concepts to twelve. Trade success dips as the new needs arrive uncoined — and the pidgin grows to meet them, settling at 12 / 12. Galison's progression in miniature: pidgins deepen into creoles exactly as fast as the traffic demands, and no faster.
the zone, everywhere
The pattern generalizes far past radar. Machine learning people and clinicians run wards together on a pidgin of "sensitivity", "flag", and "override" while meaning different things by all three at home. Product and engineering coordinate through tickets whose fields — estimate, priority, done — are pure trading-zone tokens: precise enough to transact, silent about each side's deeper ontology. A good contract is a treaty-grade pidgin; so is a well-run interface spec, and so, between a couple from different worlds, is the small vocabulary of truce words that both can use though neither could define to the other's satisfaction. In each case what makes the zone work is what naive collaboration advice would fix: its thinness. The parties stay expert because the shared language does not reach their cores.
The zone's honest rival is not the merger but the boundary object — coordination stored in a thing rather than a tongue — and real institutions run both: the trauma bay has its pidgin ("airway", "line", "push") and its boundary objects (the chart, the board). Both solve the same problem operational closure poses: systems that cannot share meanings still have to share work. The trading zone is the answer for the parts of the work that must be said — and its designers are whoever shows up at the border first and starts failing usefully.
the mapping
| In the model | In the world |
|---|---|
| native languages | Each discipline's full conceptual apparatus — what "field", "priority", "risk" mean at home. |
| trade concepts | The overlap of needs: what actually must cross the border for the joint work to run. |
| boundary agents b | The people who show up at the seam — liaisons, tech leads, the bilingual nurse. |
| coining λ | Turning a failed exchange into a shared word instead of a grievance. |
| the pidgin's size | Exactly the trade's vocabulary — thinness as the design, not the deficiency. |
| the merger | The unified-ontology project, the forced common tongue — slow, total, and lossy. |
where it tears
The radar and bubble-chamber collaborations had wartime urgency, shared machines, and institutions that put the communities in one building with one deadline. Coining is not free in life: it takes repeated contact, tolerance for failed exchanges, and someone funding the border. The model's λ dial compresses all of that into a probability — which is honest about the mechanism and silent about who pays for it.
Whose terms seed the shared vocabulary is rarely symmetric: the dominant side's words tend to become the zone's words, and the junior partner does the semantic commuting. Linguists know natural pidgins carry their power relations in their lexicons; disciplinary ones do too — ask which side's jargon your joint tickets are written in. The model's symmetric coining flatters the fair case.
Some joint work eventually needs what the pidgin cannot say — deep model assumptions, values, the reasons behind the tokens. Creoles grow, but they need governance, teaching, and time the model waves through in one preset. And the philosophical dispute stands: whether trading zones refute incommensurability or merely route around it is exactly the kind of question no simulation settles.