I don't trust AI to translate anything accurately to are from a language outside of maybe Spanish and German.
There's way too much nuance in both Korean and Japanese for this to reliably work. Chinese, from my limited study is a bit closer to English in grammar and structure so that might work
>I don't trust AI to translate anything accurately to are from a language outside of maybe Spanish and German.
But you trust scanlation groups? Neither will give you perfect, professional-level translations.
>Chinese, from my limited study is a bit closer to English in grammar and structure so that might work
Mandarin is full of nuance, and it's no closer to English than Japanese is. It has the Subject-Object-Verb grammar structure, just like Japanese and Korean.
>But you trust scanlation groups? Neither will give you perfect, professional-level translations.
I often prefer fan level over professional level because they are targeting different audience. As far as quality goes, there is a range and sometimes I skip something because the quality is too low, but I see plenty that does a good enough job.
Part of it is that there is no such thing as a perfect translation because there isn't an exact equivalent in another language. For someone with no knowledge of the original culture or language, there is some translations that will probably work best, but the more one knows about the language and culture, even a small amount picked up just from consuming other items, the the more likely a different translation works better. For a definite concrete example, how should one handle honorifics like chan, san, and kun.
> Mandarin is full of nuance, and it's no closer to English than Japanese is. It has the Subject-Object-Verb grammar structure, just like Japanese and Korean.
This isn’t correct from what I’ve studied in both Japanese and Mandarin.
> For example, Chinese verbs are not conjugated and only have one form, whereas Japanese verbs have a wide range of conjugations and particles.
Plus, Chinese is an SVO (Subject+Verb+Object) language just like English, so sentences are easier to make and interpret.
Vice versa, Japanese is an SOV (Subject+Object+Verb) language, meaning you do not say: "I eat sushi" but "I sushi eat".
You're absolutely right -- though, while Mandarin’s SVO structure does align with English in basic sentences, what’s interesting is how flexible word order and grammar in general can become in practice, thanks to its reliance on context and particles rather than rigid syntax. For example:
Mandarin often moves the object to the front for emphasis, creating an OSV or SOV structure (e.g., 寿司你吃, "Sushi, you eat" or 你寿司吃了吗 "You sushi eaten?"). This isn’t true SOV grammar but highlights how meaning shifts through word order in ways English can’t replicate without rephrasing.
The nuance in Mandarin often comes from particles that take on very different meanings depending on how they are used (e.g. 了, 的) and contextual cues rather than conjugation. For instance, 吃 "eat" becomes past tense with 了 (吃了), future with 会 (会吃), or continuous with 在 (在吃)—no verb changes needed. But if you say 要吃了, it actually means future tense of "will eat soon"!
Meanwhile, Japanese relies heavily on verb conjugations (食べる→食べた) and postpositional particles (は, を) to mark grammatical roles, in a way making its structure more rigid and easier to interpret. Personally I found Tae Kim's interpretation of "Japanese isn't SOV, it's actually V!" to be useful.
Both languages share subject-drop tendencies (like omitting "I" or "you" when contextually clear), and compound-word formation in both languages from the use of Chinese characters (kanji) adds another layer of contextual interpretation.
I think this isn't about subtle nuances but blatant errors that require full multimodal input to notice. Most MTLs(including LLMs) are done by divorcing text from content and processing it in CSV-like lists, and there's just not enough data in text by itself. This routinely lead to outputs doing something completely different from input.
Not in manga space, but Microsoft had been notorious for nonchalantly shipping crazy MTL errors for past 3-5 years, e.g. "Copilot Child Adoption Kit", "Reply in cost estimate", or "Print in scenery". For manga and entertainments, more likely modes of errors would be wildly fluctuating pronouns, genders, personas, formalities almost in styles of Monty Python satire.
These are less likely to manifest with language pairs that are closer together like English and German, and less likely with human translators who can trivially go through pages to read with full context and/or write with consistency.
If you can learn a foreign language, why not AI. Translation does not rely on grammar structure since these can be learnt. The attention model specifically designed to handle these dependencies in translation, which then led to development of AI in other tasks. You will be surprised.
For old translation systems, you are absolutely right though.
Because Japanese is a notoriously context-dependent language, AI cannot accurately translate even simple sentences like "Daijoubu desu". Its meaning varies wildly depending on the context, such as "I'm fine", "He was okay", "We'll make it", "You can count on this", "No thank you", etc.
And context isn't always clear. It depends on where the conversation is taking place, where the speaker is looking, where speech bubbles are positioned, and if the author intends to mislead readers, the information necessary to understand the context might even be revealed in future episodes.
I use Claude Sonnet to translate Korean web novels to English. It works fine.
The fact of the matter is there's going to be some nuance lost in even a good human translation. For example as far as I can tell in Korean martial arts novels a word is used which can mean either "demonic" or "unorthodox" to describe a group in opposition to the "orthodox" martial artists and a human translator has to basically pick one and go with it removing whatever nuance is in the original.
I'd prefer an official translation of course but the AI one is "good enough" if none is available.
Formal vs informal speech is another thing that doesn't seem to translate- but I don't think you need to understand every nuance to enjoy what is in many cases a straightforward genre story.
Manga is much trickier since the AI if reading the page may not be able to pull from the context of other pages the way it could in a novel.
There's way too much nuance in both Korean and Japanese for this to reliably work. Chinese, from my limited study is a bit closer to English in grammar and structure so that might work