You may know the feeling: you swear in your second language and the sound comes out with nothing behind it, while the same word in your first language makes your tongue hesitate first.
That impression has an instrument reading behind it, and a more troublesome set of consequences. If emotion runs one notch weaker in a foreign language, then every judgement that emotion pushes around — fear of losing money, fear of dirty hands, fear of saying something too harsh — should shift too. How much, and in which direction, can be measured.
This line has been measured for over a decade, and the result is messier than the headline: the effect is solid on some problems and simply absent on others, and the handiest explanation — that a foreign language sits further from emotion — was punctured by an experiment run in Italy. Three things in order: money, the push, and emotion itself.
Here is the problem as it is usually given. A disease is expected to take 600 lives, and you must pick one of two programmes. First description: programme A saves 200 people for certain; programme B has a one-third chance of saving all 600 and a two-thirds chance of saving nobody. Most people take A.
Second description, same numbers: under programme A, 400 people will certainly die; programme B has a one-third chance that nobody dies and a two-thirds chance that all 600 die. Most people flip to B.
"Saving 200" and "400 dying" are the same event out of 600, yet they push people toward opposite choices. That flip is the framing effect — the same fact, described differently, produces a different choice — reported by Tversky and Kahneman in 1981 and one of the sturdier old results in judgement and decision-making.
Keysar, Hayakawa and An (2012) did one thing to it: they translated it. Three groups — 121 native English speakers who had studied Japanese at school for an average of three years; 144 native Korean speakers with roughly a decade of English; 103 native English speakers studying in Paris who had six years of French behind them. In their own first language, the framing effect showed up as usual. On the same problem in the foreign language, the flip did not appear.
They tested something else too: a bet with positive expected value — you win more when you win than you lose when you lose, so repeated play pays, though a single round can still go against you, which is why many people decline. In a foreign language, acceptance rates went up, including in a version played with real money.
Careful not to read this as "a foreign language makes you smarter". What changes is not arithmetic but sensitivity to wording: the same person does not compute faster in the foreign version, it is just that the phrase "400 will die" no longer pushes them. The scope is narrower than the headline, too — Vives et al. (2018) tested several other judgement biases (outcome bias, the representativeness heuristic) and found no language difference at all.
There is a problem worn smooth in philosophy classes, with two versions that differ slightly and produce very different answers. Version one: a runaway tram is heading for five people, you are standing by a lever, and pulling it diverts the tram onto a side track where one person stands — most people say pull. Version two: no lever; you are on a footbridge beside a person large enough that pushing them off would stop the tram — most people say no. The arithmetic is identical, one life against five.
Costa and colleagues (2014, in PLOS ONE) put both to 725 people, a large share of them native Spanish speakers with English as a foreign language and the rest the other way round. Answering in a first language, 20% of responses endorsed pushing; in a foreign language the figure rose to 33%.
More informative than that main effect is the other half of the result: on the lever version there was no language difference at all. What a foreign language moves, then, is not whether one-against-five adds up, but the resistance to shoving a person off a bridge with your own hands.
Which raises the real question: does a foreign language make you think more, or feel less? Both stories work — reading a problem in a language you do not know well is effortful, and effort might buy you an extra step of deliberation; or the sentence may simply never reach the emotion. Hayakawa et al. (2017) took this apart in six experiments using a technique (process dissociation) that estimates the two forces separately, scoring the "save more people" pull and the "never use a person as a tool" pull independently. The answer came out on the second side: a foreign language weakens the prohibition rather than strengthening the utilitarian pull. Feeling less, not thinking more.
Geipel, Hadjichristidis and Surian (2015) checked it from the other end: they swapped in scenarios with no arithmetic in them at all — offensive violations of social norms that leave nobody actually harmed — and found the same acts judged more leniently in a foreign language. There is no "save five" to compute here, so the only thing left to move is the revulsion itself.
The evidence strength deserves its own line. A 2022 meta-analysis (the kind that pools published studies and recomputes) confirms the effect for the dilemmas where you act with your own hands, and finds it moderated by self-rated reading proficiency: the better people rate their reading, the weaker their shift toward the utilitarian option. Brouwer (2019), testing highly proficient Dutch–English bilinguals, found no difference at all. But a 2021 meta-analysis over 17 studies and 47 experiments found the effect in both the moral and the risk domain while failing to establish proficiency as a moderator. Two pooled analyses giving different answers to the same question is itself the state of this ground.
All the evidence so far rests on questionnaires. Is there a way to measure without asking? There is. Harris, Ayçiçeği and Gleason (2003) recruited 32 Turkish–English bilinguals and fitted fingertip electrodes to record skin conductance — when you tense up, your hands sweat slightly and the skin conducts differently, an old and hard-to-fool index of emotional arousal. The task was flat: read words, rate each one for pleasantness.
The strongest responses came from first-language taboo words and from childhood reprimands — the "shame on you", "go to your room" that adults say to children. English taboo words also drew strong responses, so a second language is by no means emotionless. The widest gap was in the reprimands: the equivalent English sentences read far flatter.
That gives a far more precise statement than "a foreign language is cooler": emotion is attached not to what a word means but to the situation in which you learned it. The reprimand in your first language was said to your face by someone; the English equivalent learned from a textbook mostly came off paper.
Then the puncture. Miozzo et al. (2020) tested two groups in Italy: speakers of Venetian and speakers of Bergamasque. Both regional languages are spoken at home and among familiars from childhood, while Italian is also used in formal settings. Asked the rescue problem and the footbridge problem in the regional language, both the framing flip and the shift on the push appeared, exactly as with a foreign language — yet those languages did not differ from Italian on emotional ratings. They were not "learned"; they are the language of the house.
Emotional distance cannot account for that one. The remaining candidate is where a language is habitually used: the one used in formal, public, account-giving settings and the one used in the kitchen and at the card table do not recruit the same habits. That is a candidate, not a conclusion, but it covers more of the data than "foreign equals cold".
And to keep the attitude straight: no language here is "cooler" than another. What varies is your own history with a language — the same language is somebody else's mother tongue. How to measure that history is precisely what the two meta-analyses are arguing about.
In these experiments it has one much narrower meaning: choices do not flip with the wording. That is consistency, not correctness — a person can be consistently wrong. And the framing effect is not purely an error: wording carries the speaker's hint (someone who says "400 will die" may be signalling how serious this is), and being sensitive to it is useful in real conversation. Turning the sensitivity down buys both things at once.
At least three suspects: self-rated proficiency is a coarse ruler and different studies ask about it differently; the two pooled different sets of experiments (one separates out the dilemmas where you act with your own hands); and publication bias — experiments with significant effects are easier to publish, which pushes a pooled estimate up. When judging this kind of dispute, first check whether the two sides are even counting the same problems.
This is the loosest screw in the whole line of work. The independent variable is written as "native versus foreign", but what seems to be doing the work is something else: where the language gets used, whether it was learned from people's mouths or from paper, what you have lived through in it. The Italian regional-language result is the evidence that the screw is loose. To make this field solid, "foreign" has to be broken into several separately measurable quantities.