You are in a paint shop holding two blue swatches, hesitating for three seconds. Now imagine something else: if your first language treated those two blues as two unrelated words, the way 红 (hóng, "red") and 粉 (fěn, "pink") are two unrelated words for you — would you be any faster?
That is the ground linguistic relativity — the claim that language influences thought — is actually fought over today. Not "speakers of language L cannot think X"; nobody defends that version any more. The live question is much narrower and much harder: do the distinctions a language forces on you keep working when you are not speaking?
Three areas have been dug over for thirty years: colour, space, time. The three answers come back at completely different strengths, and that difference is itself the most informative thing here. Each gets the same ruler: how big is the effect, can you see it in a task that needs no words, and is it still there once language is blocked.
English blue covers the whole range from pale to dark, and so does Mandarin 蓝 (lán, "blue"). Russian has no such neutral superordinate: you have to choose between голубой (goluboy, the light kind of blue) and синий (siniy, the dark kind), the way no single English word covers both red and pink.
Winawer and colleagues brought this into the lab in 2007, with a task deliberately built to allow no talking: three patches on screen, one on top and two below, one of which exactly matches the top one. Press a key for left or right. Pure discrimination speed.
The result: Russian participants were faster on cross-boundary pairs — one patch falling on the goluboy side, one on the siniy side — and slower on within-boundary pairs, where both were light or both were dark. Show English speakers the same patches and the difference does not appear. And the smaller the colour difference, the clearer the effect.
The second step is what matters. They added a concurrent task: one group held a string of digits in mind while discriminating (tying up verbal resources), another held a spatial figure (tying up spatial resources). The digit group's cross-boundary advantage vanished; the figure group's survived. So the advantage is language joining in online, not different eyes.
Note the size of the effect: tens of milliseconds. Not "they see different colours" — one judgement got a shade faster. And the controversy runs the other way: Martinovic, Paramei and MacInnes re-ran the same task in two experiments in 2020 and found no cross-boundary advantage in reaction times, plus evidence that the siniy/goluboy boundary itself gets pushed around by how often and in what order the patches appear. This evidence is currently contested, not established.
While we are here, one widely circulated item deserves dismantling: the claim that the Himba "cannot see blue". The demo where eleven green squares hide one blue square and they supposedly cannot spot it comes from a 2011 television documentary, and no matching experiment appears in the literature; the published work (Roberson et al.) reports no discrimination advantage at the blue–green boundary their language does not mark, and an advantage at the boundaries it does mark.
"The cup is to the left of the book" hides something you do not normally notice: whose left. Change seats and the same scene needs a different sentence. That coordinate system, anchored to the speaker's viewpoint, is the relative frame of reference — which is to say, what you measure direction against.
Guugu Yimithirr, spoken at the Hopevale community in Queensland, Australia, does not use it. The language has four direction roots — gungga (north), naga (east), jibarr (south), guwa (west) — and all talk of location and motion runs through their inflected forms. Not "the ant on your left" but "the ant north of you". Tzeltal, a Mayan language of Tenejapa in Chiapas, Mexico, takes another route and uses terrain: ajk'ol "uphill" (roughly south, there), alan "downhill" (roughly north), jejch "across" — the same system for a coin on a table and for a village kilometres away.
The point is not the exotic-sounding "they have no words for left and right". The point is that to say one ordinary sentence in such a language you must know at every moment where north is. So Levinson and colleagues built wordless tasks: lay out a row of small animals all facing one way on a table, walk the person to a second table across the room, and after a delay have them lay it out again. Only two arrangements are possible — preserve "left to right relative to me", or preserve "actually facing north". Guugu Yimithirr and Tzeltal participants leaned towards the latter, Dutch participants towards the former.
Then comes the prettiest rebuttal in this whole area. Li and Gleitman ran the same table task with American students in 2002, varying exactly one thing: the laboratory blinds. With the blinds down and no view outside, nearly everyone chose the relative solution; with the blinds up and the scenery visible, or with the whole thing moved outdoors, people split roughly half and half. One more detail makes the case: in the conditions with a view, about 70% asked before starting which solution was wanted, against 20% with the blinds down. Not a word of language changed; the environmental cues changed, and so did the answer.
Levinson and colleagues hit back the same year, arguing that this conflates the geocentric frame with the object-centred one. The exchange has no clear winner: Li and colleagues later, in 2011, re-tested Tzeltal speakers with several different rotation tasks and did not reproduce the dominance of the absolute frame.
What survives is this one sentence: the frame of reference is not issued to you by your language, it is calibrated jointly by environmental cues and habit, and language's role is to make one of them obligatory, so that you have to maintain a running bearing estimate. That is much weaker than "their spatial cognition is different", but it can be measured. Evidence strength: cross-cultural non-verbal tasks, replication failures in both directions, samples of only a few dozen people.
Time has no direction of its own, yet you can barely talk about it without borrowing space. English says "the week ahead". Mandarin says 上周 (shàng zhōu, literally "up week", meaning last week) and 下周 (xià zhōu, literally "down week", meaning next week) — a vertical axis. Same thing, two different axes.
Boroditsky reported in 2001 that English and Mandarin speakers making temporal judgements are primed differently by spatial cues in different orientations. That result then met three replication failures: Chen tried four times in 2007 and got nothing, January and Kako ran six experiments in 2007 with the same outcome, and Tse and Altarriba in 2008 likewise. Chen also pointed out something more basic — horizontal time expressions in Mandarin, such as 往前排 (wǎng qián pái, "move it forward") and 往后推 (wǎng hòu tuī, "push it back"), are actually more frequent than vertical ones.
In 2011 Boroditsky, Fuhrman and McCormick changed tack: no linguistic judgement at all, just have people point out a series of time points in the three-dimensional space around their own body. Both groups showed a left-to-right axis, matching writing direction; only the Mandarin speakers additionally showed a vertical one. The task itself needs no words, which makes this one of the firmer pieces here, though far from settled.
Aymara, a language of the Bolivian and Peruvian highlands, offers a different kind of surprise: it uses nayra — basically "eye, front, sight" — for the past, and qhipa — basically "behind" — for the future. Núñez and Sweetser in 2006 looked beyond the words and filmed the gestures: speakers point in front of themselves for the past and flick backwards for the future. One explanation they offer hangs on the language's evidential marking, which requires you to say whether you witnessed something: what you have seen goes in your field of view, what you have not goes behind your back. This is the only arrangement of its kind documented so far.
The cleanest piece is from Pormpuraaw. Boroditsky and Gaby in 2010 handed speakers of Kuuk Thaayorre a shuffled stack of cards — a man ageing, a crocodile growing, a banana being eaten — and asked them to lay them out on the ground in the correct order. They laid them east to west: facing south that runs from their own left hand to their right, turned to face north it runs right to left, and facing east it runs towards their own body. Nobody mentioned direction to them, and nobody told them the experiment had anything to do with bearings.
Why this piece is firm: laying out cards needs no words at all; one person turning to a new orientation produces a new layout, which makes them their own control; and no previously documented population laid time out this way — everything recorded before was body-relative. But do not read it as the strong version: they can perfectly well follow a left-to-right layout. What changed is the default, not the capacity.
They can, and that is the hardest thing here to state properly. Tens of milliseconds is not "small enough to ignore", it is "small enough that it must be described correctly": it is about judgement speed, not about what you see. To judge whether a write-up has overreached, see whether it dares to print the effect size.
This is where those blinds really bite. Environmental cues alone are enough to make people switch frames, and communities speaking absolute-frame languages also tend to live where landmarks are clear and people move around all year — language and environment come bundled, and the field cannot separate them. Only one design can: find two communities with similar environments but different languages. Such pairs are rare, which is also why the samples here stay so small.
Possibly neither is language itself. In those very datasets the left-to-right axis lines up with writing direction — and writing direction is a property of a script, not of a grammar. So "language influences cognition" has to say what "language" means: vocabulary? An obligatory grammatical category? Or the writing habits this language happens to come packaged with? Blurring the three is the commonest slide in this field.