Where can I help translating Joseki Explorer? It is a very useful feature but unfortunately the computer-translated version (of Thai at least) make no sense currently. I helped translate the website before and I want to contribute to the OJE too.
I think the joseki explorer currently uses an auto translate system separate from the usual one on Pootle.
That’s likely why there’s a quality difference in this particular language
2 Likes
In what sense? Better or worse?
Worse.
It’s the same auto translator that we use in things like chat and some system messages.
We send the text to an AI translator, with some prompts, and use what it sends back.
The reason for doing it this way is that in OJE the text-to-be-translated is not “in our code” it is “entered by users and stored in the database”.
Text that is “in the code” can be sent to pootle and translated by humans. It changes rarely.
Text that has been entered by users changes all the time, and is not suited to the pootle approach.
The best results we’ve had in improving the auto-translations is when we provide more “context”.
We haven’t tried to do this for OJE yet.
The way anyone could make a start at helping is like this:
-
Find a text that does not translate well.
-
Ask ChatGPT to translate the English in OJE into the target language, without telling it anything else.
-
See if ChatGPT is equally bad as whatever we’re using.
-
If it is, then try telling ChatGPT something to help it make a better translation.
-
When you find something see if that same something helps with other OJE text
If it does, then it’s a general purpose improvement and we can prompt our translation with it.
A curious thing: I grabbed a couple of the Thai translations that OJE gave me (by setting my language preference to Thai) and asked ChatGPT to translate them back to English.
They were great!
What does this say:
โกบันว่าง ความเป็นไปได้ไม่มีที่สิ้นสุดรอคุณอยู่!
“Go board is empty — endless possibilities await you!”
Full marks.
What does this say:
4-4 จุด หรือที่เรียกว่าจุดดาว; โฮชิ จุด 4-4 มีอิทธิพล แต่จะสับสนได้ง่ายเมื่อพยายามหาวิธีใช้หินจุดดาว
สิ่งสำคัญคืออย่าปล่อยให้ตัวเองถูกผูกมัดด้วยอคติ ธรรมชาติของจุดดาวไม่ได้ทำให้มันไร้ประสิทธิภาพในการรักษาอาณาเขตเสมอไป คุณต้องมีความยืดหยุ่น พร้อมที่จะปรับกลยุทธ์ให้เข้ากับสถานการณ์ที่เปลี่ยนแปลง บางครั้งใช้กลยุทธ์จุดดาวเพื่อสร้างอิทธิพล ในขณะที่บางครั้งมุ่งเน้นการยึดครองอาณาเขต (หน้า 10)
"4-4 point, also known as the star point; Hoshi.
The 4-4 point is influential but can be confusing when trying to figure out how to use the star point stones.
It’s important not to let yourself be bound by bias. The nature of the star point doesn’t make it inherently ineffective for securing territory.
You need to be flexible, ready to adapt your strategy to changing situations.
Sometimes, use the star point strategy to create influence, while other times focus on occupying territory."
(Page 10)
Full marks again!
So I wonder if something else is the problem?
6 Likes
ChatGPT did A LOT of polishing there. Thai sentense read more like this:
Point 4-4, or so call star point, hoshi point 4-4 has influence but easily confused when try to search use for pebble hoshi…
The rest is pretty normal. As you can see, when it come to Go-specific terminology (hoshi, stone, etc.) then ChatGPT simply cannot translate it properly.
I believe this might be the case for other languages as well.
Is it possible that, in the future, there can be different text pools for different languages?
3 Likes
It is definitely the case for Chinese, like the translation for timer in the help is still poorly translated, not only part of it not translated, but in general just uses very weird terms not used in the context of Go
计时器
遊戲時鐘控制著你在遊戲中可用來進行移動的時間。有兩種流行的遊戲時鐘類型：費舍爾時鐘和讀秒時鐘。你會在按鈕的左側找到費舍爾時鐘的設置，而讀秒時鐘的設置則在右側。
費舍爾時鐘的運作方式是給每位玩家初始的時間來進行走棋，然後每做一個走棋還會獲得額外的時間。在遊戲時鐘按鈕上會有兩個數字，第一個是初始時間，第二個是每走一步增加的時間。例如，10 分鐘 + 10 秒意味著每位玩家開始時有 10 分鐘，然後每走一步獲得額外的 10 秒。
Byo-yomi 計時器的運作方式是先給每位玩家一段較長的時間來完成大多數的移動，然後再給予若干個延長時間以完成剩下的移動。如果在一個 byoyomi 延長時間內完成移動，該段時間會在下一個移動中重置。如果移動所花費的時間超過了一個延長時間，則該段計數會減去一，玩家進入下一個延長時間。例如，20分鐘 + 5個30秒代表每位玩家一開始有20分鐘的時間來移動任意次數，然後必須在30秒內完成移動，否則會失去一段延長時間。如果所有5段延長時間都用完，則玩家會因超時而輸掉比賽。
And I don’t even know what this segment is supposed to be
使用指標
棋盤尺寸和遊戲時鐘按鈕的右上角可以有兩個圓點之一：
- 10m + 10s灰色圓點表示這在過去30分鐘內是一個熱門選擇，如果您想快速找到遊戲，這是個不錯的選擇。
- 10m + 10s突出顯示的點表示當前有玩家在線，正在尋找具有相容設定的遊戲。選擇此選項通常會立即匹配成功。
And believe or not, it used to be even worse.
These observations continue to confound me.
Here is ChatGPT’s translation of the first section above:
“”"
Timer
The game clock controls how much time you have to make moves during the game. There are two popular types of game clocks: Fischer clock and byo-yomi clock. You can find the Fischer clock settings on the left side of the button, and the byo-yomi clock settings on the right.
Fischer Clock
The way a Fischer clock works is that each player is given an initial amount of time to make moves, and then gains additional time after each move. The game clock button will display two numbers — the first is the initial time, and the second is the increment time per move.
For example: 10 minutes + 10 seconds means each player starts with 10 minutes, and gains an additional 10 seconds after each move.
Byo-yomi Clock
The byo-yomi timer works by first giving each player a longer period of time to make most of their moves, and then a number of short “overtime” periods to finish the rest.
If a player completes a move within a byo-yomi period, that time resets for the next move. If the move takes longer than the byo-yomi time, one period is lost and the player enters the next byo-yomi period.
For example: 20 minutes + 5 x 30 seconds means each player has 20 minutes to play any number of moves. After that, they must make each move within 30 seconds, or they lose one overtime period. If all 5 overtime periods are used up, the player loses the game by timeout.
“”"
It’s basically spot on.
I guess this might be because if AI translated FROM English to “Wierd Chinese”, then it is happy translating back from “Wierd Chinese” to “English”!
However, interesting as this is, it actually has nothing to do with translating OGS Joseki Explorer.
I’m not actually familiar with where this instructional text comes from 
1 Like
Yes, it’s weird Chinese indeed, like the most basic level. For the most simple term like make a move, in Chinese it would be 落子, not 移動. And a game clock would be 計時器 or 計時鐘, not 遊戲時鐘. It’s like using the phrase a sleeping log to describe a bed. And a completely idiotic use of 費舍爾時鐘, which means a clock made of Fischer, instead of the existing terminology like 加秒制 or simply 費舍爾計時制. Not to mention byoyomi is not translated at all, where the above Fischer timer is translated using the pronunciation. Most of all, the sentence structure is often very weird, Chinese grammar is different, you cannot simply copy and paste the sentence structure from English. Instead of “Chinese English”, it is “English Chinese”.
Remember this joke? This is akin to that (maybe not that bad, but still very off)
4 Likes
I think it depends on the issue you want to solve.
I think the idea of machine translations of text that can change “often enough”, makes sense when you want to make sure that everyone sees the same thing.
If you had separate pools of text, it might be that the Thai OJE is calling the move a mistake, the Chinese move says a pro thinks it’s playable and links to a video, and the English one says it’s a trick move.
Something to that effect is what I mean. You could basically end up with multiple OJE’s that only share a common move pool, but widely different meanings.
On the other hand one could try to allow for manual translations, but it might be quite messy, as people can add whatever as comments, and as you mentioned there’s a lot of Japanese Go terms that don’t translate super well, if they’re written in English by English speakers and then mapped from English over.
Similarly there might be idioms and other expressions in various languages that could be hard to translate, if you don’t have multilingual speakers translating.
Now what we could imagine doing as a half way point might be to provide more context.
What this could be for examples is a subfolder of the OGS repo, which stores context files for different languages.
For example senseis library has things like
and
etc.
Hebsacker Verlag /gospiele provides another list of Chinese go terms
You can imagine letting users contribute to lists like this on the repo. Then the relevant lists are given to Chat GPT as context to see if it aids the translation.
1 Like
I’ve taken @GreenAsJade’s section above (though maybe I should take the original section)
GaJ's section on timer, clock etc
and fed ChatGPT the following instructional prompt as well as the Chinese Go terms pdf linked above and two senseis library pages on Chinese and Japanese go terms :
You are both an expert translator into multiple languages and an expert at recognising terminology from the board game Go, and translating these terms seamlessly from one language to another. To assist you, you have also been given access to some user-submitted Go terms.
It is important to remember that many Japanese Go terms are widely used in English, so it is important to recognise these Japanese loan words and translate them correctly.
This is the new translation @Counting_Zenist if you can tell us if it does any better this time?
计时器
对局计时器控制你在对局中每步棋可用的时间。常见的计时方式有两种：菲舍尔制（Fischer）和读秒制（byo-yomi）。你可以在按钮的左侧找到菲舍尔制的设置，在右侧找到读秒制的设置。
菲舍尔制计时
菲舍尔制的工作方式是：每位棋手在开局时拥有一段基础时间，每下一步棋后再获得一定的增时时间。计时器上会显示两个数字——第一个是基础时间，第二个是每步的增时时间。
例如：“10分钟 + 10秒”表示每位棋手开始时有10分钟，每下一步棋，会额外增加10秒。
读秒制计时
读秒制计时器的机制是：首先给予每位棋手一段较长的时间用于主要落子，之后进入若干个较短的“读秒阶段”（也称为“保留时间”）来完成剩余的对局。
如果棋手在某个读秒阶段内完成了一步棋，该阶段的时间会为下一步重新开始计时；如果超过该时间，就会消耗掉一个读秒阶段，进入下一个阶段。
例如：“20分钟 + 5次30秒”表示每位棋手有20分钟可随意用来落子。时间用尽后，每步棋必须在30秒内完成，否则就失去一个读秒阶段。如果5个读秒阶段全部用完，棋手将因超时判负。
It also asks if we want a traditional Chinese version (I think),
如需繁体中文版本，或希望其中部分术语保留英文，请告知！
so we could try generating that.
計時器
對局計時器控制你在對局中每一步棋可使用的時間。常見的計時方式有兩種：費舍制（Fischer）與讀秒制（byo-yomi）。你可以在按鈕的左側找到費舍制的設定，在右側則是讀秒制的設定。
費舍制計時
費舍制的運作方式是：每位棋手在開局時會獲得一段初始時間，每下一步棋後會額外獲得增時時間。計時器按鈕上會顯示兩個數字——第一個是初始時間，第二個是每步棋增加的秒數。
例如：「10分鐘 + 10秒」代表每位棋手起始擁有10分鐘，並且每下一步棋會額外增加10秒。
讀秒制計時
讀秒制的方式是：先給予每位棋手一段較長的基本用時來進行主要落子，之後進入數個較短的讀秒階段（又稱「保留時間」），以完成後續的對局。
若棋手在某一讀秒時間內完成一手棋，該讀秒時間會重新開始計算；若超過該時間，就會消耗一個讀秒階段，進入下一次讀秒。
例如：「20分鐘 + 5次30秒」表示每位棋手有20分鐘的基本用時，在此期間可自由落子。時間耗盡後，每步棋必須在30秒內完成，否則就會失去一個讀秒階段。若5次讀秒全部耗盡，棋手將因超時而判負。
If it ends up even being a slight improvement, there are a bunch of prompt engineering techniques we can try to tweak and see if we can generally improve it translating into multiple languages.
Another idea might be to ask someone like @JethOrensin if it would be ok to generate lists of Go terms from his project
If we really kick it into gear that is, to make lists of Go terminology across multiple languages so that something like ChatGPT can auto translate OJE or other parts of the OGS page (not the pootle stuff, that seems like it’s working fine.)
Maybe I can make a separate thread later if we want to collect sources and lists of Go terms in various languages.
1 Like
I have no problem with it, at all.
If it is useful for Go promotion and something in my project can be of help, I am almost always in favour of it (with the only potential exception of people proceeding to charge money for it).
1 Like
Better, at least in terms of sentence structure, but subtly wrong in certain meanings and the phrases (as usual for translation using LLM). People who know Go would look at it and know what it is trying to say, but can point out the places that are incorrect or not the most common phrase to use in the common Go terminologies in different contexts, while those who don’t know might get the wrong idea or later on would use phrases that no one would use and wouldn’t understand. Even the first title 計時器 is a very “machine translation” in the context of an online clock, we would often use 用時 for the “abstract concept”, not the “physical clock”. And we have already translated terms for Fischer as I said before, “加秒制” in comparison with “讀秒制”, which are pretty self-explanatory for a Chinese reader. And although the sentence structure is somewhat correct, the content is “weird”, Like it doesn’t convey the same meaning, but something weird, such as “控制你” means “mind-control you” without the associated objective, but “控制(你的)用時” means control how much time [you have] (we would often omit the subject). If you don’t add the object, the meaning would differ, and if the subject is added, it cannot be separated (there are some limits or risk “double meanings” breaking up and emphasizing the sentence in the wrong place). And the translation actually has a different meaning than the original sentence. 控制(你在)对局中每步棋可用的时间 means “control how much time each move (you) can have in a match”, instead of “how much time you have to make moves during the game” it should simply be 控制(你)對局的用時 (proper translation is often much “cleaner” and shorter). It is just the first sentence, and I can go on for all the subtle differences and “clunky” translations and phrases used that might be grammatically correct but have different meanings throughout.
Also, there are terminologies and phrases differ between those used in mainland China and those we use in Taiwan, they are not simply replacing simplify Chinese words with traditional Chinese ones. Although it is a smaller issue then conveying the wrong meanings in the first place.
2 Likes
I also didn’t even realise we had this help section until I went looking for it from this thread
I imagine this is the help from the play page?
I guess with the help text, because it’s so long, just translating words individually won’t work like on the pootle translation.
I’m very slightly aware that not all language are simply about swapping in the right words, the grammar and sentence structure can be very different. I’m just curious about how far we can go with the tools we have available atm.
With the OJE text, which is much shorter maybe it can do better? Let’s try the Hoshi one.
Original English text
4-4 Point
AKA star point; hoshi. The 4-4 point allows a player to establish a foothold in a corner while also enabling development of influence.
Although the 4-4 point favors influence, it is not necessarily inefficient at securing territory. Players should be flexible with this move, ready to adjust their strategy to changing circumstances, at times using it to build influence, at other times going for territory. (p. 10.)
以下是該段文字的中文翻譯，提供簡體中文與繁體中文版本，並使用正確的圍棋術語（如「星位」「勢力」等）：
簡體中文翻譯：
4-4点
又稱星位，日語中為 hoshi。4-4點能夠幫助棋手在角部建立立足點，同時便於發展勢力。
儘管4-4點偏向於獲取外勢，但它在圍地方面也並非低效。棋手應該靈活運用這一步棋，根據局勢變化調整策略——有時用來發展外勢，有時則轉為實地。（第10頁）
4-4點
又稱星位，日文作 hoshi。4-4點可讓棋手在角部建立據點，並有利於發展外勢。
雖然4-4點偏重於勢力發展，但它在確保地盤方面也不一定低效。棋手應對此手保持彈性，根據局勢變化調整策略——有時用來擴張勢力，有時則轉為搶占實地。（第10頁）
Then the other natural question is: Is no translation better than a bad translation?
Or should there be small disclaimer to let the user know when something is AI translated or auto translated, so that they can be aware of it?
1 Like
If we are talking about Go terms, then this is a very tricky question to answer, depending on the language.
For example, I’ve had quite the difficulty to translate the term “ladder breaker” in Greek or find two terms to differentiate between a “clamp” and a “pincer”. Other languages might have words that sound more natural for those terms, but have a difficulty on “tiger’s mouth” or something similar.
As far as I am concerned, I went with the idea that if the translated term sounds silly, then using the term as a foreign import should at least be considered. We do it for terms like joseki, fuseki, haengma etc, so one more would hardly matter, as long as we give it a try to have most terms translated (sometimes people use so many japanese terms in their analysis here and elsewhere and I actually have to look up those terms in google or sensei’s to understand what is being said).
I am not sure if this is relevant, but on the practical side another thing that should also be considered, depending on the interface, is the size of the textboxes. It is reasonable, though not obvious, but the same text in a different language can take an entirely different amount of space (none of the versions of my book are the same, I constantly have to resize or even move textboxes and images to make the translated text fit).
So, if you want to view something on a computer, the size of the textbox might be easily scalable, but in mobile screens, it could be something that might need tweaking.
2 Likes
yes
Less content for it to go wrong doesn’t mean there is nothing wrong.
For the “4-4 point” translation, no one who learn Go in Chinese would use 4-4 in their sentence for Go commentary, let alone using “4-4 point” (4-4點). First of all, 星位 probably would be one of the first terminologies introduced in the very first class, since it is related to the board itself. Second, even if we use it, we would probably use Chinese numerals and no hyphenation (四四) without “點” (also 點 has its own specific meaning in Chinese Go terminologies, several actually). And the translation with “又稱星位” is weird, since 星位 is the proper term in Chinese, not an alias (it would be like a page title “sleeping wood furniture”, AKA a bed), the whole paragraph doesn’t make sense to people learning Go in Chinese natively. And although we do use 勢力, but it is not the same as 外勢 (the proper translation of influence, and used both in simplified and traditional Chinese). Actually, 勢力 means establishing a foothold rather than using 據點 or 立足點 (which are both rarely used in the context of Go game). The whole first paragraph is sort of redundant with the second paragraph, except for Japanese pronunciation. And as a Chinese speaker, the use of English to refer to a term as Japanese is just weird. And it wouldn’t make sense to write in Japanese, since it is the same in kanji 星 as in Chinese.
Also, as I mentioned before, we often omit the subject in a short sentence, hence the insertion of 可讓棋手在 is something redundant. And there are weirdly phrased sentences like 但它在確保地盤方面也不一定低效, it is something no one would say. The proper translation grammatically would be like 下在星位圍地的效率較低. However, it is still not using the “proper” Go terminology, since 子效 or 效率 has its own meaning in Go, and it is not about territory. The proper meaning for the second paragraph should be 星位比較偏重外勢，(相較於小目)較不偏重取地，而是平衡外勢，可以根據(全局)不同的配置轉換。(as I said before, proper translation is often pretty short and concise). LLM translation has nothing wrong in terms of sentence structure, just that people would say, it looks right at a glance on the surface, “but” we wouldn’t phrase them like that (especially not in the Go game context).
My personal opinion is that if they don’t convey the right meanings, or use terms that no one in the target language would ever used, or terminologies have other meanings and just use them incorrectly, it would do more harm than good. Since for someone who isn’t familiar with the terminologies in the first place would just use them incorrectly based on wrong translation, and later on misunderstand actual commentaries from Go players learn them natively. What is the point of creating more communication barriers when the translation is supposedly to improve communication in the first place? It would actually take less effort to find someone who speaks natively and knows Go, than provide a clunky and full of “inconsistency” and sometimes completely wrong phrases and terminologies, and need many times the effort to correct them, where the AI translation disclaimer wouldn’t tell where they went wrong (or how wrong they are)
1 Like
I think they’re kind of speculative barriers to be honest on whether they would cause any real harm (unless the translation was offensively bad).
It’s not that much of a big deal if someone says “I would nobi here” when they actually mean sagari, because some online resource used the wrong word when translating something from Japanese to English for example.
In a real conversation someone would just correct it or understand what they mean.
If it was in OJE, another more knowledgeable person could also fix it too (the English at least).
Basically the real question is, is it much more of a communication barrier for instance to not have OJE in any language other than English?
Or is it better to attempt to translate it? (Or we can leave it as is until some other idea comes along)
At least what I’m seeing is that OJE doesn’t seem to bother trying to translate the text into any form of Chinese atm.
It does do a few other languages, but I haven’t gone through them all to see which ones.
Sure but it’s not some official book or text, it’s some random people on the internet trying to pool together knowledge.
It could also be changed later today or tomorrow on a whim, so it doesn’t really work to have it officially translated somehow.
Then it’s about whether you want to attempt to help people read it, or you’d rather they find their own way to translate or just not use it at all.
I also assume most contributions to OJE so far have been accompanied with English text descriptions.
I wonder what would happen in the reverse situation, or in a situation where we had several people wanting to contribute something to a position, in multiple languages.
I’m going to assume OJE isn’t set up for position descriptions to be in multiple languages.
It’s not down to me how we do it, @GreenAsJade would make a call on that.
I’m just trying to discuss and flesh out a bit what we can do now (like trying to prompt better or provide better context for the auto translations) or whether we can do something else.
1 Like
I’ve stated it was my opinion, and I do believe it is a barrier. I’ve tried to sort of promote OGS in Chinese communities before, and the poorly translated pages make they not just difficult to access and they feel very “unfinished” and “unprofessional” from what I heard back. It is like if you cannot get the fundamental terminologies right, (and for the most part still looked machine translated), they would just assume they wouldn’t be able to communicate with others on OGS, and designed by people outside Go communities.
And correctly wrong terminologies isn’t something that easy, even just using it in the wrong context like 效率 later on when we really need to talk about stone efficiency, it takes extra effort to correct them as teachers. As an interface, it feels not making an effort to make them right, and as a teaching it makes the jobs harder or even impossible. wonder why so few East Asian players who would stick around?
3 Likes
Yeah, but I suppose there is no singular professional involved in all the languages OGS tries to support.
I’m not sure I’ve ever heard people call Fox unprofessional or unfinished because even in the English client a bunch of the buttons are in Chinese.
Maybe that is some insight then as to why OGS isn’t picking up more in other places though - maybe it needs a better all round translation.
Also valid, but that’s just the reality of playing Go on the internet for most people in most countries.
I mean fair enough, I obviously don’t know a whole lot about this, with my main experience being playing Go while talking in English.