Topic on Talk:Content translation

Algorithm of automatic blocking of translated article should be revised

8
Fourierkim (talkcontribs)

Dears,

I am totally agree with the idea of free and fast sharing of information of Wikipedia. I have posted some new or revised articles mostly in Korean, including a few revised articles in English and Japanese.

My starting point is, as others in this talk, usually based on automatic translation of English articles and then careful revision by myself. The automatic translation greatly helps me save time. All my translation based articles have been accepted by Wiki's human reviewers.


But today, I found that I cannot post my translated and revised article as before. Even after I checked "The issues are solved", the same blocking message "Your article cannot be posted because too many automatically translated words are included ...... "(in Korean) appears again and again. There seems to be no way to bypass or ignore the blocking.


There seems to be some intended change or (unintended) error in your algorithm.


I think "there should be a way to ignore or bypass" the automatically generated warning and/or blocking. The quality of the translation can not be determined by machine based on the percentage of modified words. The ultimate determination should be made by humans.


I greatly appreciate your fast and accurate revision of the program.


P.S. The article I am working on is, "Ed Roberts" in Korean, base on corresponding English article, Ed Roberts(Computer engineer).

Juxlos (talkcontribs)

I agree with this - ironically, Google's translation tool is good enough so that it's hard to edit more than 10% of it without being incorrect.

Pginer-WMF (talkcontribs)

Thanks for the feedback, @Fourierkim.

Currently we are adjusting these limits. The way they work is documented in this page. We want to find a balance to prevent lightly edited machine translations that read unnatural (and editors have reported about), and not getting in the way of those creating good translations. Finding this balance is not simple, and we'll need feedback from different users, so thanks for sharing your experience. I already made a proposal to make the limits less strict. Given the review and deployment cycles, the change may be effective in a couple of weeks.

In order to get a better understanding of the issue, can you provide more details on the following?

- Where you getting a warning for a paragraph that you edited enough?

- Where you getting a warning for some kind of content that makes no sense to edit (e.g., math formula)?

- Were you marking the warnings as resolved?

- Based on the documentation linked above, is there any particular adjustment that you think would be helpful in your case?


Thanks!

Juxlos (talkcontribs)

- Some paragraphs were already translated almost perfectly, and editing it to make it below 80% is just redundant work

- I marked the warnings as resolved, and am now just stuck with the overall limit

- The limit being in place in the first place is weird - if there aren't any quantitative quality enforcements for normal edits, why have them for translations?

Tuanminh01 (talkcontribs)

True, to block 1% vandalism, you limit 90% translators and their eargerness and ability to contribute.

Please reconsider.

Pginer-WMF (talkcontribs)

Thanks for the feedback, @Tuanminh01.

We are in the process of making adjustments to the translation limits. In that process we may go through iterations where at some points the limits may become a bit too strict and in other iterations the limits may become too loose. We are working on improving the limits to avoid the current issues, and feedback from editors is essential for adjusting the limits in the best way. Please, feel free to provide specific examples where the limit system is not working for you.

Currently our observations suggests that despite the problems of the current limits being too strict there were not many blocked translations. The stats page shows that last week 4676 translations were published, more than any other previous week (including those previous to the latest adjustment of the limits).

Thanks!

Juxlos (talkcontribs)

Specific example would be a translation of the “June 2019 Gulf of Oman incident” article from English to Indonesian (though someone seemed to have created an article now).


Google Translate between English and Indonesian is well-established and the machine translation is already about he same as what any human translator would provide, save a few edits which I had done. It still wouldn’t let me publish.

Tuanminh01 (talkcontribs)

@Pginer-WMF Thank you for your reply. Most of new translations come from our viwiki by a technical expert who has been bypassing the above problem by translating a lot of very short articles (<5 paragraphs).


Reply to "Algorithm of automatic blocking of translated article should be revised"