communicoupling
She says: "I nag because you withdraw." He says: "I withdraw because you nag." Each account is honest; each fits every remembered scene; and they can fight about it forever, because both are locally true. The loop runs — nag, withdraw, nag harder, withdraw further. So what, exactly, are they disagreeing about? Not the facts. The cut.
Watzlawick, Beavin and Jackson named the mechanism punctuation of the sequence of events and made it their third axiom: the nature of a relationship is contingent upon the punctuation of the communicational sequences between the communicants (Pragmatics of Human Communication, 1967, after Bateson and Jackson). An interaction stream has no native episodes. Each party segments it into stimulus→response units — and each, invariably, casts the other's act as stimulus and their own as response. "Who started it" is a quarrel between two segmentations of one loop: undecidable from inside the data, and itself a reliable engine of escalation.
Below, a nag–withdraw engine with coupled dynamics — A's nagging feeds on B's withdrawal, B's withdrawal feeds on A's nagging, each with a lag. The stream renders as alternating acts; a comb of cut-marks punctuates it into episodes; and the instrument scores both stories against the same data, then says out loud whether the data can pick a side.
Two combs, one stream
The engine underneath is deliberately even-handed. Two intensities evolve in coupled steps: A’s nagging rises with B’s withdrawal as it stood τ ticks ago, B’s withdrawal rises with A’s nagging at the same distance, and a slow fatigue term lets every surge blow over — which is why the stream swells and crashes in waves instead of climbing forever. Every twelve ticks an act is emitted, alternating A, B, A, B, its height the current intensity. Nobody starts: the loop was already running when the window opened.
A punctuation is a comb of cuts dropped onto that stream. A’s comb opens every episode with B’s act, so each of A’s moves reads as a response; B’s comb opens every episode with A’s act and returns the favour. Each story is then a causal claim, and the instrument tests it: across the recent exchanges, how often did the “response” actually track its “stimulus” — high answering high, low answering low? With symmetric dynamics the two scores come out equal, run after run, and the verdict line says what the couple cannot: the data cannot pick a story. The second view asks a sharper question. Cross-correlating the two raw series across every lag shows twin peaks, one on each side of zero — B leads A exactly as much as A leads B. That mirror is what undecidability looks like to an instrument. Raise the Δ dial and the mirror breaks: one agent genuinely reacts harder, one peak grows, and the index leaves zero. Punctuation-blame and real asymmetry become visibly different things.
What to try
On the classic preset, toggle A’s punctuation ↔ B’s punctuation. The gold arrows re-pair the identical blocks; both fits sit near 88%, within a few points of each other. Then click between acts to merge episodes — “this fight started last Tuesday” — and watch the scores barely move. No cut settles it.
Drag Δ to 0.80 with reactivity 1.40. A now answers roughly nine times harder than B, and the right-hand correlation peak takes over — index near +0.2. Note the episode scores often still nearly tie: blame bookkeeping can miss a real instigator that the lag instrument catches.
Set lag τ = 36 — the cold war. Answers arrive three beats late, so adjacent-act punctuation pairs each move with the wrong reply: both fits collapse to about 30%, below a coin flip, while the correlation shows the true answer-distance as twin peaks far from zero. Then press The therapist’s move.
The interminable argument
Every escalating relationship keeps two ledgers of one history. The manager: “I check up on you because you go quiet.” The team: “we go quiet because you check up.” The parent tightens the curfew because the teenager sneaks out; the teenager sneaks out because the curfew is tightening. Watzlawick’s own favourite was the arms race — each nation arming, by its own punctuation, purely in defence. In every case both parties describe the same loop correctly at the local scale; each is genuinely responding to the other’s last move. The fight is over which move counts as last — and that is fixed by the cut, which is exactly the thing neither ledger records.
The argument is therefore interminable rather than merely hard: every new act is filed by both books as fresh proof. Her next remark confirms his story; his next silence confirms hers; the quarrel about the punctuation escalates the sequence it is punctuating. The way out that works in the model also works in the room, and it is what a good couples therapist actually does. Refuse the question. Show the stream uncut — one oscillation, two colours — and replace “who started it”, a question about a cut, with “what pattern are we in”, a question about a loop. The first has no answer inside the data. The second has one you can see, and, once seen, sometimes step out of.
Next door
The engine this stream runs on is schismogenesis — Bateson’s name for interaction patterns that amplify themselves, of which nag–withdraw is the complementary textbook case. Punctuation is what each party does to that spiral in memory: cuts it into episodes that flatter their own position. Push the contradiction down a level — a message whose framing forbids every available response — and you reach the double bind. And the move that dissolves the quarrel here is a small instance of second-order observation: watching the cut being made — including your own — instead of arguing from inside it.
The mapping
| In the model | In the world |
|---|---|
| the stream | The marriage’s actual history — continuous, loop-shaped, episode-free. |
| the cut-marks | Each partner’s memory of episodes: where a fight “began” and what merely preceded it. |
| stimulus→response casting | “I was only reacting” — the other’s act filed as cause, one’s own as effect, every time. |
| equal fit scores | Why the argument is interminable: the same data certifies both accusations. |
| the cross-correlation | What a neutral instrument sees — and the honest way to ask whether one side really drives. |
| the uncut view | The systemic reframe: from “who started it” to “what pattern are we in”. |
Where it tears
Real relationships often do have causal asymmetries — power differentials, first movers, one party with far more room to stop. The axiom warns against inferring asymmetry from inside your own punctuation; it never claims asymmetry does not exist. That is what the Δ dial is for: when one agent genuinely drives, the correlation says so, and “it takes two” stops being the whole truth.
Rendering the stream as measurable acts — twelve ticks each, alternating politely — is already a punctuation. A truly uncut history would have nothing to score, no pairs to test, no comb to drop. The instrument demonstrates that cuts are arbitrary by using cuts, a recursion it cannot escape and should own: even the therapist’s “one oscillation” is a description, drawn in someone’s categories.
“It’s the pattern, not a person” can heal a marriage — and can launder genuine wrongdoing into “the system”. Abuse is not a dance; a victim’s resistance is not one arc of a symmetric loop. The reframe is a tool for cases where the loop really is symmetric, and deciding whether it is symmetric is precisely the judgement the reframe itself cannot make for you.