Topic: Establishing a Translation Guide/Policy for e621

Posted under Tag/Wiki Projects and Questions

Over the last week and a half, I started on a wiki project to collate various bits of translation-related information.
In particular, there are a few bits of pieces on translation-related policy that you can pick up if you browse old wikis like translated and hard translated (which previously fulfilled the purpose that translation edit does now, records related to translation misuse, and certain deletions. There's also the comparatively recent introduction of translated description, but I am not currently aware of any incidents related to people adding unofficial non-English description translations to posts.

The Translation Guide is a wiki page that covers:

  • important translation-related precedents (deletion of bad translation edits, penalizing users who add non-English translations incorrectly, penalizing users who consistently add bad translations, etc.)
  • establishing translated character/copyright tags (when to adopt localized names, when to ignore localized names, translating literally-descriptive names, etc.)
  • how to resolve translation-related information with tags (translations can be a basis for lore tags like trans_(lore), but very little otherwise.)
  • how to resolve issues with incorrect translations (meet translators halfway with "translator's notes", write well-explained corrections in comments, slowly build up a case against users with a history of bad translations.)
  • various useful templates and formatting for translation-related stuff.
  • relevant language resources.

At this time, the translation guide is not de jure policy, but it does cover various de facto policies that were in effect before I became janitor. Other staff members have had the opportunity to review it for a while now, and they provided some minor feedback, but I figured non-staff translators also deserve a chance to provide input.

This document might later become the basis for more in-depth translation policy later on, but until then: consider this thread a chance to weigh in on translations, tag translations, translation descriptions, translation edits, etc.

If you've got nothing important to say, but think you can contribute some good translation-related tools like OCR, dictionaries for obscure slang, etc., that's fine too.

Updated

Nothing to add, just a bit curious, should description_check really imply description_request?
description_request sounds more like when a post has no description, and needs one.

Also, translation_check doesn't imply translation_request at the current moment.

What I mean by that is, this implication feels a bit unnecessary or confusing. And I'd like to suggest the use of description_translation_request / description_translation_check instead, they're longer but more descriptive, which may make their use more clear at a first glance.

I noticed that earlier. Looks good!

I'm not sure if all of these should be added, but here are some tools I've used for translating:

  • http://thejadednetwork.com/sfx/ - Look up Japanese onomatopoeia (including sound symbolism like キラキラ, etc.)
  • https://ejje.weblio.jp/ - En ↔ Jp dictionary with example sentences, sometimes providing example sentences next to each individual definition.
  • Google Translate's handwrite input - for inputting kanji you are unfamiliar with, so you can copy them into a proper dictionary*.
    • ...or you could OCR it. KanjiTomo (for Windows) can usually work reasonably well with text that's typed (vertically or horizontally) in a standard font, though it's far from perfect. Due to lack of experience, I can't say how it compares with alternative OCR programs.
  • Japanese Dictionary Takoboto - Jp -> En dictionary app for Android.
  • If you want to see examples of a word or phrase in sentences with translations, there are many different websites for this, though I recommend Weblio as previously mentioned. You can also just search the phrase on Twitter, though that obviously doesn't come with translations.
  • This goes without saying, but Google and fan wikis are also very helpful. Don't know what pokemon オーダイル is? Google it. You have a hunch that a phrase might be a meme, but you're not familiar with the meme? Google it.

*EDIT: Or more practically, just use a browser add-on/extension that let's you mouse over the word for a definition, such as Yomichan (Firefox, Chrome)

Updated

kora_viridian said:
3. I understand the hate-on for machine translation, but another way to interpret that section is "if you don't have near-native proficiency in the language that is to be translated, then don't bother". This probably results in better translations, but it probably also results in many fewer translations, IMHO.

I am a native speaker of English, have had formal classes in Spanish, and have self-taught a little bit of German. My experience with translating both Spanish and German into English is that machine translation isn't a terrible place to start, but you must follow it up with further editing, dictionary look-ups, etc, to get something that makes sense in English.

I realize that a lot of people will copy the foreign-language text, paste it into Google or DeepL, push the button, copy the generated English text, paste it somewhere, and write "I TRANSLATED IT, LOL" at the top. If you want to avoid that whole scene, then banning machine translation is one way to do it.

Machine translation has valid use for certain words and set phrases (and it's much better for languages that aren't Chinese, Japanese, or Korean), but my thinking is that if people can tell you are openly using machine translations, then that's generally a bad thing. In practice it would rapidly prove impossible to fully police machine translations, but in my time here I've learned that there are people who are rather uncritical about what kinds of translations they'll post. They'll accept "Meskemo" or similar as a character name despite it being a relatively common word in site-relevant art, they'll accept DeepL jank like "I'm not sure if it's a good idea, but it's a good idea.", etc.

Basically: it's only against the rules if you get caught.

In a perfect world I could just point to a bunch of real-world examples of machine translations being passed as final on e621, and explain what the issues were with each one, but that'll require some pretty thorough note/comment searching.

lafcadio said:
Machine translation has valid use for certain words and set phrases (and it's much better for languages that aren't Chinese, Japanese, or Korean), but my thinking is that if people can tell you are openly using machine translations, then that's generally a bad thing. In practice it would rapidly prove impossible to fully police machine translations, but in my time here I've learned that there are people who are rather uncritical about what kinds of translations they'll post. They'll accept "Meskemo" or similar as a character name despite it being a relatively common word in site-relevant art, they'll accept DeepL jank like "I'm not sure if it's a good idea, but it's a good idea.", etc.

Basically: it's only against the rules if you get caught.

In a perfect world I could just point to a bunch of real-world examples of machine translations being passed as final on e621, and explain what the issues were with each one, but that'll require some pretty thorough note/comment searching.

I admit that I use those tools as a start, but try to find better wording and definitions than the default. Also, machine translation fails hard for longer sentences compared to common phrases (which you should get familiar with, pretty quickly). The idea of blindly trusting what amounts to a string-matching algorithm with no context recognition (because that's an open problem) is insane. If I'm not able to figure something out, I ask others in comments. I fixed a few of my mistakes that way while learning! It wasn't the words BTW, but having trouble reading the handwriting! XD

European language translations have their own challenges. Like this example, where there's a song called Muchas Veces (Many/Much times/often; pronounced Ve-chez not vee-cees or the like) that without understanding the context, you can get a misleading translation of the title. You could translate it as Quite Often, which given some of Clutch's lyrics and song titles actually makes more sense than the machine (and dictionary!) translations. There are words that are not actually remotely related but look close between languages. These can confuse polyglots who translate multiple languages. The old joke about Cat/Cake in Latin languages.

https://www.thoughtco.com/mucho-and-its-variations-3079138 There's more like this where it uses Latin suffixes for slightly different words. There are a lot of German words that work similar to English ones that are phonetically close, but are obsolete in English or used in a barely different way. Examples like: Has/Have/Had

A funny bit mentioned in that link on muchas is that if you read the lyrics... He's referring to his problems with women. XD

Updated

Since machine translations were brought up here, I made a couple of edits to that section: Revised the text around machine translations to come off as a bit more formal, provide real-world examples of DeepL being crap, and communicate the idea that unedited machine translations are effectively low-quality translations.

Updated

  • 1