This is English! All we have is loan words 
2 Likes
I was just trying to explain, why I don’t think it’s a good idea to change the original English text.
What other options are there?
- Wait for the AI to improve
- Try to improve the AI
- Use the AI to provide translations but let humans suggest different translations.
- Use the AI to translate its output back to English and compare it to the original input to automatically detect potentially bad translations and show a warning to the readers in such cases.
- Or maybe change the way how the text is provided to the AI? How is that done currently? Whole text at once or comment by comment?
I guess I just don’t agree with this point. Obviously its a yucky work around, but this doesn’t make it a bad idea.
There is no evidence of it creating new problems, and my instinct (albeit just instinct) is that it is likely to help more than hinder.
-
I think that simplifying text where it translates wrongly is better than waiting and doing nothing
-
We have no input to the AI that I am aware of (other than pressing the “dislike” button on wrong translations)
-
This is the same as “find a good way to let humans provide translations”, which is the exact thing I’ve said we need to do, we intend to do, but are not quite up to yet, and it will be a little while until we do.
-
This would be difficult and as much effort as #3 for a worse result.
-
It’s provided “whole text at once”, but often there is not a lot of text in a Joseki Description. The real context that might help is “This is a description of a Joseki”. I guess we could try providing that each time, without including the translation of that part?? Maybe that would help the AI!?
Fundamentally, I prefer these two things:
- Simplify descriptions that are known to cause problems, in the short term
- Create a good way for humans to provide translations, as the real solution.
I think that it might not be worth bothering to do so, when it’s not clear that it helps more than it hurts. But I do know and understand your view on this. And I think you know and understand mine. So maybe it’s time to test the hypotheses.
I don’t know how to simplify:
Black plays here to avoid having Black’s corner surrounded.
If White has a stone in the upper left, White can play a pincer at around 1.
And your workaround
doesn’t work in this case, because the translation on the website is already different from the one in OJE and doesn’t leave room for misunderstandings.
So, how a) do we simplify the description or b) can we get that different translation into OJE?
Theoretically we could contact the developers. They should be interested in having good translations, but probably don’t want to be bothered about every translation that needs improvement. So, yes, not the way to go.
Right, I just wanted to have it in the list. It would be nice if having that functionality wasn’t necessary, but maybe it is. Do you know of any other translations, that are not satisfying?
Hm, I would think it’s not that hard to achieve. (When does the translation happen?) When the API sends the response, OJE requests to translate the response back to English and compares the result with the original text. Well, it might flag a lot of okay-translations as not okay. Anyway, if it worked well it would be nice to have, but it’s not necessary.
It seems to be possible to provide context like that. The problems with “Empty board” actually happens for the German translation too. It translates to “Leere Tafel” and back to “Blank slate”. It seems to interpret empty as “nothing written on it”. Translating “Empty board. Players place stones on the board.” gives “Leeres Brett.” (and the second sentence). Wich fits the context of board games.
Question: Is it safe for OJE to split sentences at ‘.’?
Hah! Well this certainly warrants a closer look.
What’s the position on OJE with this description?
Thanks!
But I’m not sure the website is always producing the same result.
1 Like
Darn AIs 
(But I will keep looking into it)
… so, I’ve looked into this, and I don’t see what you see. I get the same result from the website and OJE using the website:
I changed the position description to something easier to parse:
If you are really keen to shoot down this approach, you could try some other languages and see if any are now broken on this page 
Yes, because you’ve only entered the first sentence. That’s why I’ve asked how the text is sent to the API earlier. With the second sentence the website prints a different text. Maybe the OJE requests the API to handle all sentences on their own? I’ve seen an option for that.
I only know German and English. How can I check other languages? 
It’s an interesting thought - I’ll poke around to see if we have something like that set.
That being said, at least we know how to make the workaround work now
Go back and forth and see if it breaks?
I thought you were going to complain that I changed the meaning in the process of making it more reliable to translate
I didn’t notice any difference on the deepl website when I copied both sentences together from OJE there with the new version.
I’m possibly doing it wrong. Am I copying the two English sentences and translating to German to compare it to the OJE version when I have the site language set to German?
Edit: nevermind - now I understand
It was the old version that was doing that. That could be it
Yeah I messed up further experimentation by actually fixing the Position in question. Now the text that is there is the one that translates correctly.
Of course, someone who is keen enough could create a new “test” position in some discreet corner and put the troublesome desecription in there to play with
TBH, I’m not sure it’s worth pursuing, other than for curiosity’s sake.
1 Like
I didn’t dare to. White is the only one black can prevent from surrounding, so that’s okay. But yes, previously it was Black’s corner, now it doesn’t say that anymore, the players could share the corner.
I took a look at some other languages I partially understand and it seems that with the old phrasing “Black” wasn’t translated in some cases (it was just putting out “Black” again) because of the upper case B. But with “White” in the same sentence, that seems to have improved.
At the beginning of the sentence it uses “Zwart” (similar to German “schwarz” which means “black”) but later in the sentence it uses “Black” instead.
1 Like
Yay, I’m scary 
I do think the old wording was awkward, and that will lead to translation problems. It was awkward because we wouldn’t say that in English like that. The original author was clearly trying to avoid a gendered pronoun… this is the hassle we face. As soon as you express something more naturally, it’s more likely to be translateable (barring idioms of course).
And it’s why I think that “reword till it works” is actually a sensible workaround for now.
And I was assuming the AI had a clear understanding of the original text, as it did produce correct but easily misinterpreted German output, because of how words that end on something that sounds like ‘s’ (like “schwarz”) are treated differently in German when it comes to genitive case. And while that did play it’s part, it seems that the AI has trouble understanding the original English input after all. So my assumption was wrong.
I’m probably a bit late, but I tried something (only one move, actually) with Italian.
This translation seems different, and I’m not sure every Italian would understand it
Also “punto stella” (litterally correct) is quite funny, we wouldn’t use it. And I don’t understand why the title is “4-4 punto” and not “punto 4-4”.
I can’t scroll the original text without resizing the page, by the way
What’s the problem, the “AKA”?
Not only, also “come usare la pietra punto stella” or “una giocata a punto stella”. Actually it’s probably understandable, but surely wrong.
And the translation on the website is better? Or just different?
Oh, the German translation is terrible too.