Why Machine Translation Humor Works

At some point you've probably read a mistranslation and just lost it. Not a polite chuckle — full, helpless laughter.

It might be a menu in a foreign country offering "Fried Husband" instead of "Fried Pork." A sign warning visitors not to "molest the animals." A product label that confidently declares itself "Untranslatable."

Why is this so reliably funny? And why does running text through a whole chain of languages amplify it even further?

There are real linguistic and psychological reasons. The short version: language is a lossy compression of reality, and watching the compression artifacts pile up is inherently comic.

Language Is Not a Code

The most common misconception about translation is that it's basically a lookup table. Word goes in, equivalent word comes out. If that were true, machine translation would have been a solved problem in the 1960s.

It isn't, because language doesn't work that way.

Every language encodes a slightly different model of reality. Concepts that seem universal often turn out to be surprisingly local. The Finnish word sisu describes a kind of stubborn, stoic inner strength with no direct English equivalent. The Japanese 木漏れ日 (komorebi) captures the interplay of light and shadow when sunlight filters through leaves — a concept English simply doesn't have a single word for. The German Schadenfreude — pleasure at another's misfortune — was so useful that English borrowed it wholesale rather than trying to translate it.¹

When a machine translator encounters a concept with no clean equivalent in the target language, it has to make a choice. It might pick the closest approximate word, restructure the sentence, or occasionally produce something that's technically grammatical but semantically baffling.

The humor lives in that gap between what was meant and what came out.

The Incongruity Theory of Humor

Humor scholars have identified several theories of why things are funny. The most applicable here is incongruity theory: we laugh when something violates our expectations in a non-threatening way.²

The theory has roots going back to Kant and Schopenhauer, and remains one of the dominant frameworks in humor research today. The basic idea is that humor arises from the mismatch between what we anticipate and what we get — provided the mismatch is harmless.

Mistranslations are almost perfectly engineered for this. You read the original — "To be or not to be, that is the question" — and you have a clear expectation of what it means. Then the translation comes back as "True or false, what's the problem?" and the gap between expectation and reality is enormous, absurd, and completely harmless.

The "non-threatening" part is important. We're not laughing at someone in a mean-spirited way. We're laughing at the gap itself — at the inherent difficulty of the task, at the strange world that falls out of the cracks between languages.

Why Chains Make It Funnier

A single mistranslation is funny. A chain of them is exponentially funnier, and the reason is error accumulation.

Each translation step introduces a small amount of semantic drift — a word chosen that's slightly off, a nuance lost, a sentence restructured for grammar in a way that shifts the emphasis. None of these individual steps is dramatic. But they compound.

If each step introduces even a small amount of drift, ten steps introduce far more than ten times that drift — because the errors build on each other rather than averaging out. The final result can be completely disconnected from the original.

This is why telephone-game translations are so much funnier than single mistranslations. "To be or not to be" passed through French, German, Japanese, Swahili, and back to English doesn't give you one small error. It gives you five small errors stacked on top of each other, each one mutating whatever the previous step produced.

The result isn't random. It's specifically weird, in ways that feel almost meaningful.

Three fresh examples from our own test bench, all starting from the same sentence — "…I would have laughed it off — yet here we are, eating our words":

• Through Arabic, "eating our words" came back as "regretting what we said." Correct, joyless, and somehow sadder than the original — the machine translated the idiom into a therapy session.
• Through Hungarian, "I would have laughed it off" returned as "I would have laughed out loud." One small shift, and the narrator goes from politely dismissive to openly mocking. Same sentence, worse person.
• Through three round trips of Japanese, "here we are, eating our words" eroded into "now I take that idea back" — the embarrassed group quietly shrinking to a single person retracting a thought. A whole social dynamic, deleted by grammar.

Accidental Poetry

Not all telephone-game translations are just funny. Some of them are genuinely interesting.

"In the end, it's not the years in your life that count. It's the life in your years" — routed through Luganda, Assamese, Dinka, Basque, and Gujarati — comes back as: "In the end, age doesn't matter. What matters is your life at your age."

That's actually a reasonable sentiment. It's not the original, but it's not nonsense either. It's a genuine reinterpretation — a different angle on the same idea, arrived at accidentally through the distortions of five translation steps.

This happens more than you'd expect. Language models are trained on human text, so even their errors tend to be human-shaped. The drift isn't random noise — it follows the contours of how humans actually think and speak. Sometimes it falls into something surprisingly apt.

Linguists call this kind of emergent meaning a property of language's redundancy: natural languages encode meaning in multiple overlapping ways, which is why partial degradation often still produces something interpretable rather than pure noise.³

The Dignity Collapse

There's one more ingredient in the mistranslation comedy recipe: the deflation of pomposity.

Shakespeare, Churchill, the Bible — elevated language carries authority. It expects to be taken seriously. When you run it through a chain of languages and it comes back as something mundane or absurd, there's a satisfying puncturing of that authority.

"What's going on?" passed through nine languages becomes "Are you OK?" — and somehow that's both funnier and more relatable than the original. The translation stripped away whatever register or implication the original had and left only the core human moment underneath.

This connects to what humor theorists call the benign violation theory: something is funny when it is simultaneously a violation of how things ought to be and benign — that is, not actually harmful.⁴ A Shakespeare quote arriving as a casual wellness check is a perfect violation (Shakespeare! Reduced to texting language!) that is entirely benign (nobody is hurt; the meaning is still vaguely there).

Arguably that's what all great comedy does: strip away pretension and reveal the simple, weird, human thing underneath. Machine translation just does it by accident, at scale, with no malice whatsoever.

Frequently Asked Questions

Why do translations get funny? Because translation isn't a lookup table — every language encodes reality differently, so each hop forces the machine to guess at idioms, dropped pronouns, and concepts the target language doesn't have. The gap between what you expected and what comes back is the textbook definition of incongruity humor: a big, harmless violation of expectations.

Why does translating back and forth change the text? Every round trip forces two sets of guesses — one going in, one coming out. Information the middle language doesn't encode (tense, plurals, who's speaking) has to be reinvented on the way back, and the reinvention rarely matches the original. Chain several languages and the guesses compound instead of canceling out.

Which languages make the funniest translation chains? In our testing, languages that drop information English requires — Japanese and Korean (omitted subjects), Chinese (no tense or plurals) — cause the most drift, while structurally "hard" languages like Finnish often return text nearly untouched. Mixing a subject-dropping language with a distant grammar (say, Japanese → Finnish → Arabic) is a reliable recipe. See our hardest-language test data for the full ranking.

Is the machine making mistakes? Almost never, and that's the best part. Each individual hop is usually a defensible translation. The comedy isn't a bug in any single step — it's the compound interest on twenty reasonable guesses in a row.

Try It Yourself

The best way to understand all of this is to experience it. Take something you know well — a song lyric, a famous quote, a line from a movie — and run it through Translation Mixer. Pick languages at random. See what comes back.

You'll get something funny. You might get something unexpectedly beautiful. Occasionally you'll get something that makes no sense whatsoever, which is funny in its own way.

That's the telephone game. And it never gets old.

---

Want to explore more? Read about the history of machine translation, watch every one of these theories play out on the Duck Song's lyrics in 5 languages, or jump straight to the Translation Mixer and see what happens to your favorite quote.

---

Sources & Further Reading

1. On untranslatable words: Howard, Ella. "Lost in Translation: Untranslatable Words," Babbel Magazine. See also Sanders, Carol. The Cambridge Companion to Saussure. Cambridge University Press, 2004, for background on the arbitrariness of the linguistic sign.

2. Incongruity theory of humor: Morreall, John. "Philosophy of Humor." Stanford Encyclopedia of Philosophy, 2020. The theory traces back to Kant's Critique of Judgment (1790) and Schopenhauer's The World as Will and Representation (1818).

3. On linguistic redundancy: Shannon, Claude E. "A Mathematical Theory of Communication." Bell System Technical Journal, 1948. Shannon's original information theory work established the concept of redundancy in communication systems, later applied extensively to linguistics.

4. Benign violation theory: McGraw, A. Peter and Warren, Caleb. "Benign Violations: Making Immoral Behavior Funny." Psychological Science, vol. 21, no. 8, 2010, pp. 1141–1149.