If you have already worked out how to translate something 100 times, you have probably noticed something odd: the same sentence, the same hundred rounds, two completely different outcomes. One run comes back as a mildly stiff restatement. The next comes back as a short, confident claim about agriculture.
The rounds count is not what separates those two runs. The languages are.
This is a guide to picking them on purpose — which languages do the most damage, which ones barely leave a mark, and how to build a chain that lands where you want.
Rounds are the wrong knob
It is tempting to think of 100 as the dial marked “maximum chaos.” It isn’t. Thirty rounds through Yoruba, Thai, and Finnish will do more to your sentence than a hundred rounds bouncing between English, Dutch, and German.
That’s because damage per hop varies enormously. A hop between two closely related Germanic languages is a light paraphrase — the translator has seen millions of sentence pairs in that direction and mostly preserves structure. A hop into a language with a different script, a different word-order convention, and no articles is a demolition.
So the useful question isn’t “how many rounds?” It’s “how far does each round travel?” Four dimensions decide that.
Script changes are one-way doors
The single most destructive thing you can do to a sentence is push it out of the Latin alphabet.
Once your text passes through Arabic, Thai, Devanagari, Georgian, Armenian, Khmer, Lao, or Burmese, it is re-encoded in a writing system with different conventions for spacing, capitalization, and punctuation. English capitalization often doesn’t survive the trip. Sentence-final punctuation may be replaced by something the target script uses instead. Spacing rules differ — Thai doesn’t separate words with spaces at all.
Coming back to English, the translator reconstructs a sentence from that re-encoded form. It does a good job, but “good” here means plausible, not faithful. Small things are simply gone, and nothing downstream can restore them because nothing downstream knows they existed.
Stack several script changes in one run — Latin to Arabic to Devanagari to Thai to Latin — and you get the classic 100-times result: a sentence with the right shape and the wrong contents.
Family distance beats hop count
English is a Germanic language. Its closest living relatives in the language pool are Frisian, Dutch, and German, in roughly that order of closeness. Hops among those are gentle: word order holds, cognates survive, and the sentence usually means what it meant.
Then there’s everyone else. Hausa is Afro-Asiatic. Yoruba and Zulu are Niger-Congo, and Zulu belongs to the Bantu family with its noun-class system — nouns take prefixes that change with grammatical role, and English has nothing equivalent. Thai is Kra-Dai, tonal, and analytic. Finnish is Uralic, nowhere near Indo-European.
A hop into any of those forces the translator to make structural decisions English never asked for: where to put the verb, whether a noun needs a classifier or a prefix, how to render a tense the target language doesn’t mark. Each decision is defensible. None is reversible.
Rule of thumb: pick languages from four or five different families and you’ll hit heavy drift well before round 100. Pick languages from one family and you can run a thousand rounds and still recognize your sentence.
Agglutinative languages multiply your words
Finnish, Turkish, Hungarian, Estonian, and Swahili build meaning by stacking suffixes onto a stem. A single English word can become one long word in Finnish carrying case, number, and possession all at once.
The damage runs in both directions. English → Finnish compresses a phrase into a word. Finnish → English has to unpack that word, and the unpacking is a guess: the translator picks the reading it thinks most likely and commits. If it picks wrong, the error is now baked into the text as a normal English phrase, and every later round treats it as ground truth.
This is why Finnish and Turkish punch well above their weight in long chains. They don’t just paraphrase — they change the granularity of the text, and granularity changes are hard to undo.
What disappears never comes back
Some grammatical features exist in one language and simply don’t exist in another. When your text passes through a language that lacks a feature, the feature is dropped — and no later round can infer that it was ever there.
The clearest case is articles. Japanese, Chinese, Russian, Finnish, and Estonian have no articles at all. Send “the dog chased a cat” through Japanese and the definite/indefinite distinction vanishes. Coming back to English, the translator has to choose “a” or “the” from context alone — and it will choose confidently, whether or not it’s right.
Gender works the same way in reverse. Spanish, French, and German assign gender to every noun. An English noun entering those languages gets a gender it never had; the associations that come with that gender can bleed into word choice. When the text returns to English, the gender is gone but its side effects may not be.
Pronoun dropping is a third. Japanese, Chinese, Spanish, and Italian all allow subjects to be omitted. Once a subject is dropped, whoever translates back has to invent one — and invented subjects are where a sentence about your dog quietly becomes a sentence about someone else.
Languages that barely scratch your text
Not every hop is destructive, and knowing the gentle ones is as useful as knowing the violent ones.
- Dutch, German, Frisian — close Germanic relatives. Expect stiff phrasing and occasional word-order oddities, not meaning loss.
- Esperanto — constructed, and deliberately regular. Its morphology is so predictable that the translator has little room to guess wrong.
- Latin — a dead language with no native speakers and a heavily documented grammar. For input that’s already Romance-flavored, hops through Latin tend to preserve structure.
- Spanish, Italian, Portuguese — same family, similar syntax. Good for a run you actually want to read afterward.
A chain built entirely from this group is the “gentle drift” setting: your sentence comes back polished, slightly foreign, and still about the same thing.
Building a chain on purpose
Hypertranslate draws intermediate languages at random from the pool — over a hundred of them — and it never repeats the previous language or lands on your target early. Randomness is the default, but you can steer it.
The language settings page lets you switch languages off, and disabled languages stop appearing as random hops. That turns the pool into a design tool:
Maximum damage. Leave everything on and pick a target language far from English — Japanese, Thai, or Arabic. With the full pool live, most hops will be into unrelated families, and the script changes will pile up fast.
Gentle drift. Disable everything except Dutch, German, Frisian, Spanish, Italian, and Portuguese. Now every hop stays inside Western Europe, and 100 rounds produces something readable and slightly wrong rather than unrecognizable.
One family, extreme depth. Keep only the Germanic set and set the rounds to several hundred. You get a different effect: not chaos, but a slow, cumulative stiffening as the text circles the same grammatical space over and over.
Script roulette. Keep only languages that use non-Latin scripts — Arabic, Hebrew, Greek, Georgian, Armenian, Hindi, Thai, Khmer, Korean, Japanese, Chinese. Every single hop is a one-way door, so even twenty rounds goes very far.
Whichever you choose, the run itself is the same: paste a line, set the rounds, press translate. If you haven’t done a run yet, the three methods for translating 100 times covers the mechanics.
Try it now
Open translate 100 times, then set up your pool on the language settings page before you start — the settings apply immediately and are saved in your browser, so you only configure them once.
Run the same sentence twice: once with everything enabled, once with only Germanic languages. Same input, same hundred rounds, two entirely different fates. That comparison is the clearest demonstration of the point — the languages were doing the work all along.