Comment on Anon tries to learn Japanese

• quixotic120@lemmy.world ⁨11⁩ ⁨hours⁩ ago

  I never know what to tell people like you. DeepL is the best one at translating Japanese and it still is mixed

  Japanese is a contextual language. It is difficult for machines to translate because it is not a language where a word simply means this or that.

  Take a very short sentence:

  がくせいです - gakusei desu

  Gakusei is “student”

  A literal translation of this would be “am student” or “is student”

  But if I say this to you you will infer things based on context and です/desu takes on a different role. If I am clearly referring to myself then gakusei desu in this context becomes “I’m a student” (though to be fair I would probably have said it with a first person pronoun and particle like わたしはわ (watashi wa)

  But if I’m referring to another singular person in the room this would be inferred through context, eg “he’s a student.” But this is where it starts getting confusing, the copula (desu) doesn’t differentiate singular or plural, so context is also used to derive plural forms, eg “they are students”.

  This copula also applies to other situations outside of he/she like “it” eg コンビニです, konbini desu, konbini being “convenience store”, “its a convenience store”

  This is a very very basic idea of why. It gets more complex obviously once you move past these extremely basic examples but honestly someone more knowledgeable at Japanese should explain at that point, I’m self taught and mediocre (thus my use of 0 kanji, I’m pretty sure at a minimum there’s kanji for gakusei and watashi but I suuuuck at kanji. I at least know meat. 肉肉肉 although that’s mostly thanks to anime and the local Asian grocer lmao).

  I think AI can probably do it eventually but it will need to be able to do much better job of understanding what the source material is actually talking about and that’s the challenge to overcome. And that’s why it will probably never be able to really accurately translate a paragraph copy pasted into it

  As to the sanitation that’s a separate issue about corporate control of AI. If they don’t want their translation services to sound “vulgar” that’s their prerogative I suppose but it also means they sound less human and realistic because people are gross and ugly when they speak. Vernacular is ugly and lexicon adapts quickly in ways that people don’t always love. But to be clear I don’t mean it’s just about slurs and bad words, I mean it’s about slang in general. Words that are generally inoffensive but not considered proper. English equivalents would be what we consider zoomer speak, hella sus yeah bruh type shit (I’m not good at this part)

  source

  • kalleboo@lemmy.world ⁨11⁩ ⁨hours⁩ ago

    ChatGPT is way way better at this than DeepL or Google Translate, because you can give it context before. Like “A is a x and B is a Y, when A says z, what does that mean”

    source

    • quixotic120@lemmy.world ⁨1⁩ ⁨hour⁩ ago

      And yet feeding it Japanese it usually gives the worst outcomes even if you do this and deepL gives much better ones even if you don’t. I don’t know what magic deepL does to make that happen but when I scanlate manga deepL almost always gives the best result even when I do what you’re doing or feed ChatGPT the images directly

      source
  • Eyekaytee@aussie.zone ⁨11⁩ ⁨hours⁩ ago

    Sorry I thought you meant local context like:

    chat.mistral.ai/…/44fa329b-99db-4cb6-969c-4134ee3…

    So when you ask it to translate something in context with local slang it will use it rather than the actual definition of it

    source
  • tehmics@lemmy.world ⁨11⁩ ⁨hours⁩ ago

    Sup brah I’m trying to learn this shit you got a blog or something

    source

    • quixotic120@lemmy.world ⁨1⁩ ⁨hour⁩ ago

      no but there are plenty of of resources much more knowledgeable than I am

      Recommend genki if you’re good with a book

      If you just want to get started like today duolingo isn’t horrible but I would suggest going into setting as soon as possible and turning off romaji (the English pronunciation guide written over the hiragana). It will make it significantly harder initially but using romaji will stop you from learning the characters. You won’t read the hiragana, you’ll read the alphabet you already know.

      That said I did duolingo to brush up a bit and I really dislike that they don’t explain much when you make errors. Given it’s an app they have the ability to bring up so much information for errorless learning. I guess they want to make it “easy” and prevent information overload but it tends to be that they present you with a screen when a concept is introduced then that screen is gone forever. So then if you make the error you just have to kind of figure out the grammatical rules behind what you’re doing wrong

      In the beginning it’s not so bad but eventually you’ll get very confused, why does は (wa, kind of like “is”) sometimes go here and sometimes go there? Why is it sometimes は and sometimes が (ga)? Duolingo never really clearly explains particles, wa marks the topic and ga marks the subject. Duolingo just throws dozens of examples at you until you consistently get it correct without necessarily knowing why unless you dig through the app or research independently

      It bothers me because I am in mental health by trade and my research interests and background is in human behavior and more specifically learning and skill acquisition. This is an inefficient model for skill acquisition. Allowing the learner to make this many errors as part of your model slows learning, potentially significantly, and allows the learner to learn bad habits and mistakes. So as you’re doing it and mistranslating you may ingrain that

      それら** は**わたしのあかいコートですか (sore wa watashi no akai koto desu ka) are those my red coats?

      Should be

      それらわたしのあかいコート** は**ですか

      (sore watashi no akai koto wa desu ka) which is grammatically incorrect and makes you sound like a goddamn fool

      But when you’re learning concepts like doko/どこ and soko/そこ (where, there, basically) you learn that “wa” goes towards the end of your phrase like whatever wa doko desu ka similar to the incorrect example above

      Sore/それ and kore/これ are demonstrative (this and that, basically, though the above example again shows how contextually this falls apart and it becomes “those”) and are typically introduced shortly before doko/soko

      Fun times! Remember these are real basic examples and I am also real dumb

      source