This was an fun read, as someone who's both a Korean BW player and a speech recognition researcher.
It's interesting to note that the original Korean transcription already has many errors, seemingly (and impressively) corrected by LLMs later on. For example, 12 안마당 빌드 (12 courtyard build) is actually 12 앞마당 빌드 (12 frontyard build), which might have been more understandable to BW players. Similarly 투에처리 빌드 (processing-at-two build? makes no sense lol) should have been transcribed 투해처리 빌드 (two-Hatchery build).
Therefore it may also be helpful to directly feed the slang dictionary into Whisper's inference process using contextual biasing. There are lots of ways to do this, but the simplest would be to increase the probability of slang words in the dictionary in the final prediction layer of Whisper by a constant factor. This is fairly easy to implement, for example by using HuggingFace's library: https://huggingface.co/docs/transformers/en/internal/generat...
I was able to understand the Google Translate version well, but I am very familiar with the intricacies of BW and zerg 12hatch openers.
Chatgpt and Claude did an incredible job translating the korean text:
Claude:
Today I'll teach you about the 12 Hatchery build. I'll explain the types of 12 Hatchery builds, their advantages and disadvantages, and the build orders in a simple but detailed way.
Against Protoss, this is the build you use when you want to start with the most economic advantage. Against Terran, there are several builds you can do with 12 Hatchery, so I'll explain some of the most commonly used builds.
The first is the two-hatchery build that starts with 12 Hatchery:
12 Hatchery
11 Spawning Pool
10 Gas
This build uses early gas, and it's often used when you want to quickly transition into a three-hatchery build with three gas bases.
The second build is:
12 Hatchery
12 Pool
12 Gas
This build allows for moderately fast tech tree and moderately fast three-hatchery expansion. This build is commonly known as the "safe three-hatchery" build, and you can think of it as a build that enables both quick Mutalisks and quick third base.
Don’t let the title fool you: this is anextremely thorough and creative take on translating and making more approachable the commentary of StarCraft.
As the author rightly points out, in its 27 years of existence, commentary around the game has become a domain specific language. Not just Korean or English.
This approach of automated scripting and using AI to understand roughly what was said and then make it coherent is really cool.
LOL, as a non-native English speaker, reading this reminds me of EXACTLY the same problem of translating many things, but more precisely, computer articles and software development.
There’s a huge amount of terms that are difficult to translate (sharding? Hash?). The only real solution is to adopt them to your language, more or less adapted, which is what happens over time. But it requires a community that, to some degree, is able to cross the gap between the languages. In this case, learning English.
Talking about software development in Spanish (my native language) is a succession of imported terms from English.
I don’t think there’s a good way of doing that, and I’m interested to see how automatic translations deal with it, because the only way this can work is with a process of mixing both language in a social way and see what terms evolve from that process.
And you need, in the terms the post describes, people that know Korean at least in a non-fluent way. And the game itself, of course.
Even when Google Translate got pretty good I was not really able to effectively translate Chinese or Japanese text about Go (the game). I had similar issues to the ones mentioned in this post. Many Chinese and Japanese words (e.g., "ko") have a very specific meaning in the context of Go, but they also have regular meanings (e.g., "robbery") in more normal contexts, so Google Translate would translate text in a generic way, which made everything unintelligible. With modern LLMs, I can now preface my translation requests with instructions such as "I am going to ask you to translate some Chinese text accompanying weiqi diagrams. Your translations should be idiomatic and not shy away from Go jargon. For example, 拆 = extension, 夹 = pincer, 刺 and 觑 = peep.", and it does a fantastic job, enough for me to basically read anything I want. It was lucky for me that evidently enough Go material already existed in the training set that I didn't have to do anything more special.
(Some chess corrections, in case the author is reading: the moves at the start of chess games are called openings in English, not openers; there are not distinct white-piece and black-piece openings, although of course an individual player will probably study a given opening from the point of view of one side or the other; their study is considered fundamental all the way up to the highest level, in fact more so as you increase in skill; and the Sicilian variation in question is the Najdorf, not Najdork.)
jaeyounkg ·7 hours ago
It's interesting to note that the original Korean transcription already has many errors, seemingly (and impressively) corrected by LLMs later on. For example, 12 안마당 빌드 (12 courtyard build) is actually 12 앞마당 빌드 (12 frontyard build), which might have been more understandable to BW players. Similarly 투에처리 빌드 (processing-at-two build? makes no sense lol) should have been transcribed 투해처리 빌드 (two-Hatchery build).
Therefore it may also be helpful to directly feed the slang dictionary into Whisper's inference process using contextual biasing. There are lots of ways to do this, but the simplest would be to increase the probability of slang words in the dictionary in the final prediction layer of Whisper by a constant factor. This is fairly easy to implement, for example by using HuggingFace's library: https://huggingface.co/docs/transformers/en/internal/generat...
Show replies
diziet ·41 minutes ago
Chatgpt and Claude did an incredible job translating the korean text:
Claude:
leshokunin ·8 hours ago
As the author rightly points out, in its 27 years of existence, commentary around the game has become a domain specific language. Not just Korean or English.
This approach of automated scripting and using AI to understand roughly what was said and then make it coherent is really cool.
jaimebuelta ·4 hours ago
There’s a huge amount of terms that are difficult to translate (sharding? Hash?). The only real solution is to adopt them to your language, more or less adapted, which is what happens over time. But it requires a community that, to some degree, is able to cross the gap between the languages. In this case, learning English.
Talking about software development in Spanish (my native language) is a succession of imported terms from English.
I don’t think there’s a good way of doing that, and I’m interested to see how automatic translations deal with it, because the only way this can work is with a process of mixing both language in a social way and see what terms evolve from that process.
And you need, in the terms the post describes, people that know Korean at least in a non-fluent way. And the game itself, of course.
Show replies
dfan ·3 hours ago
(Some chess corrections, in case the author is reading: the moves at the start of chess games are called openings in English, not openers; there are not distinct white-piece and black-piece openings, although of course an individual player will probably study a given opening from the point of view of one side or the other; their study is considered fundamental all the way up to the highest level, in fact more so as you increase in skill; and the Sicilian variation in question is the Najdorf, not Najdork.)
Show replies