Topic on Talk:Content translation

I have been trying to publish an Indonesian article translated from English to no avail. Wikipedia refuses to publish my article because 68% of the text in my current translated article is unmodified. The thing is I have actually edited the article quite extensively. I'm not sure how I can bring the percentage of unmodified text down to zero. The Wikipedia's auto-translate feature actually did a wonderful job despite many errors, which I HAVE fixed. Editing it further in order to bring the percentage down would be the same as writing an entirely different article! That's not the purpose of translating existing content, right? And, also what is the point of offering auto-translate, which I suppose can be useful to help expedite translation, when in the end I have to re-write everything because of the silly restriction?!

Thanks for your feedback. Content Translation has a system to encourage the review of the initial machine translation. The limits can be adjusted according to the needs of the community. For the particular case of Indonesian, the limits were made more strict as a response to the publication of low quality translation. Currently for Indonesian, the system prevents publishing when the overall unmodified content is higher than 40%. You can check the whole discussion in this ticket.

We are very happy to update the limits making them more or less strict to find a better balance, but we need to make sure we are not moving too much in the other direction. Based on your experience, which would be the percentage you consider excessive for unmodified machine translation?

If you can share any other information about the article you were translating, that would be also very useful. Since there is some types of content such as math formulas that are counted as unmodified but they should not since it is not expected that users change them in the translation.

Thanks for your feedback, it is really important to improve the tool.

Thanks for the response. I am currently translating an article on "Williams syndrome" (Indonesian translation: Sindrom Williams). It's a rare genetic condition. What I'm doing is part of an effort of spreading William syndrome awareness in Indonesia. I have a family member with the condition. However, Wikipedia's content translation restrictions are really slowing down my endeavor.

Yesterday, I decided to delete my initial translation, and start over. I tried to translate everything manually, but I couldn't find a way to start from scratch without having to click on that stupid "+Add Translation" thing. The tool is not really stupid actually, because like I said, it does a great job translating most of the content, but a human has to edit it. Of course. I wanted to just do it all manually, but I couldn't start typing without being forced to generate auto translation! When I deleted everything, and did my own translating, I still wasn't allowed to publish because apparently my translation resembles the translation originally generated by the tool. Duh! Really. This auto translation tool is driving me NUTS! What a terrible waste of time! I spent the whole day translating only two sections of the article.

I eventually was able to find a loophole and publish my translated article without having to re-translate everything. I edited what needed to be edited. It was a lot of work, nevertheless. It was very daunting since most of the work was not really translating, but trying to fool an artificial intelligence. It was really silly. After having been able to publish two sections, Wikipedia started to scrutinize my work again. I wasn't even allowed to publish my own translation. Right now, I'm so frustrated I'm thinking to just give up. This is a serious waste of time.

Yes, I get it. The world doesn't trust Indonesia and I do understand that there is a lot of frauds out there. However, 40% is way too strict. I am an experienced translator and I have no idea how I can do that. In my experience so far, editing the auto translation and bringing unmodified text down to 75% is a LOT of work already. Often, there is just nothing more left to modify. I often get stuck at 85%, and then after applying a certain trick, I get stuck at 46%. All the time. To make matters worse, like I said above, even rewriting the whole auto translation with my own, doesn't grant a pass. This is truly frustrating.

I'm not too sure what would be the best threshold for Indonesian content translation (maybe 80% or 75%?), but setting the bar too high will only further discourage people from contributing translation.

It would have been faster to translate manually rather than using +Add Translation. If only Wikipedia let me do that...

Update: I was stuck at 46% for the longest time, but then after deleting a few translated sections and then adding this (Wikipedia does not allow me to translate this article. You should be ashamed, Wikipedia.) suddenly I was able to publish... Why is that?

Thanks for the details, @Cindyutama.

Based on your feedback and the conversations in the previous ticket, we are in the process of making the threshold for Indonesian less strict. As a result, it would be possible to publish articles with up to 70% of unmodified contents. Given the review and deployment cycles, the change should be effective in about a week.

Once the system is updated, please let us know how well it works for you.

Adjusting a system like this requires of some iteration to find the balance where we prevent wasted efforts from both translators that are prevented to publish a good translation, and reviewers that have to check low quality translations. Sharing your experience is really very useful for us to improve the system.

Thanks!!