birdgirlbeak

HIGHLY IMPORTANT CROWBOLD

MOST POWERFUL BIRDCHOSTER ON PLANET MIRTH
𝚯𝚫
I AM THE "PEST" IN "PESTO"
I AM AN ADULT
cawcawcawcawcawcawcawcawcawcawcaw
pfp by crowtaur


0xabad1dea
@0xabad1dea

you could reasonably translate this as either peeping or squeaking (as in cute small noises) but apple translate has an innovative idea


0xabad1dea
@0xabad1dea

not only have I discovered another new species of crift (Apple Translate absolutely cannot render 蛐蛐儿 as cricket and comes up with a new spelling every time), it has decided that goatees are out, goateaches are in

(Little Pudding is the girl’s name and there’s nothing here grammatically to distinguish her from a literal cup of pudding, so I’ll give the translator a pass on that one)


You must log in to comment.

in reply to @0xabad1dea's post:

Ah, the joys of crowdsourced translations.

I wonder if that portmanteau pops up enough in the internet corpus they fed into the model, or if enough people gamed some 'suggest a better translation' functionality like they used to do with Google Translate.

I don't think this is caused by crowdsourcing or anyone having ever actually said this ever, but by the recent switch to LLM translators that work on tokens that approximate morphemes. It's just kind of losing track mid-word and using the front half of one acceptable translation and the back half of another. This same translator renders 蛐蛐儿, an informal but normal Chinese word for cricket, as crift, crips, crike, criet... it always gets the "cri-" right and then just kind of slaps a random token on the end.

Interesting, I would have expected a LLM translator to still be weighted by the context window of the output language towards sequences of tokens that occurs more frequently in the target language. (The Chinese word examples looks a lot more like that with several of the possibilities being words or subsets of words.)

If they do lose the output language context when translating certain tokens, that seems like a serious regression towards the behavior of extremely early machine translation systems.

i am a native speaker and i'm with abadidea on this, yeah. it's the continuous form

i'd also say it can be translated as 'beeping', a more neutral sound, since it's also used for something like an alarm clock. also figuratively for 'whining'

in reply to @0xabad1dea's post:

it's a children's time travel novel called Oracle Script School and the Great Ming Dynasty. A better English rendition would be: "An old man with a goatee gently squeezed through the crowd to Little Pudding's side, saying: 'Young lady, be careful or you'll crush that cricket.'" (she is holding a cricket she found who happens to be a magical time traveling talking cricket)