Has anyone put together a decent tutorial anywhere on proper methodology for translation work?

[goofusdoofus]

I'm thinking about putting together a tutorial on how to properly setup and tackle a translation project, probably using an rpgmaker game as an example.

It seems like a lot of folks are either:
-going in via RPGMaker and hunting down text to translate, translating each thing they find by hand
OR
-lobbing it through an mtl script they've cobbled together from other projects and then calling it a day.

Both of these are terrible ideas for all kinds of reasons, with a solid well thought out strategy for translation you can do higher quality work in a shorter period of time, this applies to both human and machine translation projects.
Before I started, I wanted to ask if this existed somewhere else already, and gauge interest a little bit.
IS this something people would be interested in?

 

[Silentce]

As a translator I have my own way of doing things that works for me, and that ive developed through conversations I have had with other translators and personal experience. I always felt like finding a way that works best for you is more important then following a generic guide. That is just me though.

Generally when I start working on a project I usually focus first one the database entries for the RPGM game. (items, actors, Weapons, etc.) and focus on finishing all of those. Once I've effectively created a partial AKA UI translation, I focus on doing the common events.json and then the map files last. (or if I get bored I jump to map files to break up the monotony.) After I've finished common events and map files I will to some play testing look for any variables that are not translated and if I find some I will find where those live and translate them also. Go back and test a bit again and then I am done. This is an incredibly simplified version of how I work but in essence I use that flow and the RPGMaker engine and then Translator ++ for the map files.

 

[goofusdoofus]

As a translator I have my own way of doing things that works for me, and that ive developed through conversations I have had with other translators and personal experience. I always felt like finding a way that works best for you is more important then following a generic guide. That is just me though.

I don't disagree, there's no single right way to do a translation, (although given the number of sloppy halfassed mtl's I've seen there are clearly plenty of wrong ways) but I thought maybe a simple guide might be a good jumping off point for newer translators.

I usually make a point of doing a vertical slice translation first, hitting an early entry in each data type or event type first to identify contextual problem areas in my toolset before I translate a single word (this is how you avoid crashes), and I always tackle any names, special terms, and proper nouns so I can be certain to translate them consistently (the eternal debate between Inma/Sex-Demon/Succubus will never end!).
I'd like to include some guidance for machine translator configuration- but after looking around a bit, I don't think many peopel are actually using configurable translators.

 

[StoneM]

It's hard to make some universal guide, because most games are different and you face unique problems with each of them.

I think learning to use Translator++ and translate the games without bugs is probably enough for beginners.

Going through the engine is more complicated and you can also brake the game easily, without ever realizing.


MTL translation is garbage most of the time, because the machine doesn't understand context. I don't think MTL would ever be good enough (except for the simplest of texts) .

Probably a machine can become better at translation if it actually can learn from the text it translates, but most engines doesn't seem to do that. For example Google spews the same garbage, as it did few years ago (it's slightly better in some things and slightly worse at others), so no it doesn't learn anything (even thought you can suggest better translation, they wont ever use it for some reason) .

My hope was DeepL... it's better than Google for some texts, but not for others. And sometimes Yandex gets really close (I think they use human translators or some crazy AI, because sometimes the translation doesn't have anything to do with the text, it sounds literally like human tries to translate without knowing the context, so the translation is more like "bla bla blyat").

About your suggestion to configure machine translation engines... I don't know how I can do this. I've never used a translation engine installed on my computer and there is no option to configure the ones on the web.

I think we need MTL with hentai brain to process all these Japanese games.

 

[goofusdoofus]

MTL translation is garbage most of the time, because the machine doesn't understand context. I don't think MTL would ever be good enough (except for the simplest of texts) .

Probably a machine can become better at translation if it actually can learn from the text it translates, but most engines doesn't seem to do that. For example Google spews the same garbage, as it did few years ago (it's slightly better in some things and slightly worse at others), so no it doesn't learn anything (even thought you can suggest better translation, they wont ever use it for some reason) .

My hope was DeepL... it's better than Google for some texts, but not for others. And sometimes Yandex gets really close (I think they use human translators or some crazy AI, because sometimes the translation doesn't have anything to do with the text, it sounds literally like human tries to translate without knowing the context, so the translation is more like "bla bla blyat").

About your suggestion to configure machine translation engines... I don't know how I can do this. I've never used a translation engine installed on my computer and there is no option to configure the ones on the web.

I think we need MTL with hentai brain to process all these Japanese games.

Most local translation systems like LEC and ATLAS can be configured to solve the context problem, and then you can get solid translations out of it, but it is a ton of work, you need to write your own custom dictionaries, setup regex expressions to identify and keep it from escape sequences/references/program calls.

I've done some machine translation work from chinese to english and chinese to german with neural networks in Azure, and it's fricking incredible -nearly 100% readable..... but it was part of a million+ USD corporate project, with a team of talented professionals and was heavily tuned for the documents in question.
I know this is a solvable problem, I just don't think anyone can really afford to put the personnel and cash behind training a neural network for smut translation, and there is a real maintenance cost associated with keeping something like that rolling once it's setup. Care and feeding of big cloud networks is a non-trivial cost.

It's hard to make some universal guide, because most games are different and you face unique problems with each of them.

In detail your right, but the broad strokes are the same (or should be), regardless of whether or not it's an MTL or not and no matter what engine the game is in.