apple intelligence's language model supports 16 languages as of writing. the model can read them, write them, and — most usefully for ignore — learn how you write them.
every language has its own rhythm of texting. spoken language has rules. text doesn't. text has habits. these are the things ignore picks up on, per-language, per-person.
english
contractions everywhere. "yeah" instead of "yes." periods at the end of a text mean you're upset. ellipses mean you're not done thinking. "k" is a war crime. "lol" doesn't actually mean laughing — it's a softener. all of this varies wildly between contacts, and that's exactly what the on-device fine-tune is learning.
danish
low-key, very informal. "ja, helt sikkert" reads as warm; the same content in english would feel stiff. punctuation runs sparse — no apologies for it.
dutch
text-message dutch loves abbreviations: "ff" for "even" (a sec), "idd" for "indeed." word order can break for a short reply.
french
context-dependent formality. "tu" vs "vous" is non-trivial in text — ignore notices which one you use with which contact, and stays on the right side.
german
"hi" is universal, "moin" is regional, "hallo" is safe. compound words get written long even in shorts. surprisingly little abbreviation overall — germans like the full word.
italian
emoji-heavy but not random. there's a vocabulary of which emojis attach to which feelings. "ti amo" carries weight; "ti voglio bene" is everyday.
norwegian
shares patterns with danish but slightly more punctuation. "takk" appears more than english "thanks" — it's also used as a sign-off, not just a response.
portuguese
distinction between brazilian and european matters. brazilian text reads warmer, with more emojis and softeners; european portuguese is clipped and more formal.
spanish
regional difference is significant. mexican spanish in text is playful with "x" for "por" and "q" for "que." spain leans toward written-fully, with fewer abbreviations.
swedish
minimal capitalization. text starts in lowercase and stays that way (the swedish brand voice, and ignore's, by accident, for entirely different reasons).
turkish
richer suffixes per word, so short messages pack more meaning. the model handles the grammar fine; what it has to learn is whether you use the polite suffix forms or not.
chinese (simplified)
text often skips subjects when they're inferable. punctuation is lighter — you don't need a period to end a thought, a new line will do.
chinese (traditional)
similar to simplified at the casual register; the divergence is in formal writing, which doesn't show up in texts often anyway.
japanese
depending on contact, you'll swing between casual ですます and plain-form 食べる-style. ignore notices which contacts get which.
korean
honorifics are not optional. the difference between "고마워" and "감사합니다" is the difference between texting a friend and texting your boss's parents. the model handles it; your habit decides it.
vietnamese
tone marks matter for meaning. modern texting often drops them when context is clear, but not always — depends on who you're writing.
the punchline
the model is multilingual. your texting style isn't. you have habits in each language, and habits with each person, and those compound. ignore's job is to learn the second-order pattern — the rhythm of you, in that language, with that person.
a multilingual app shouldn't feel translated. it should feel like the version of you that speaks that language. that's the goal.