Topic on Talk:Content translation

The analyser of the percentage of the unmodified text is completely faulty, is a mess.

7
Grintescu (talkcontribs)

The analyser of the percentage of the unmodified text is completely faulty, is a mess.

Hans Haase (talkcontribs)

Thanks, but when completed, it displays anything near 100 %.

Pginer-WMF (talkcontribs)

Hi @Grintescu,

Can you provide an example of the translation or paragraph that was not working for you?

Without more details about the problem it is hard to investigate it.

thanks!

Grintescu (talkcontribs)

How can I provide an example? You should enter my account to see how I can't transfer, on a Wikipedia page, the translation of the English article "Bias". The automatic answer is: "Your translation contains a total of 85%, or 80% or ... of unmodified text.

Grintescu (talkcontribs)

I'm sure the mathematical principle of the the analyser of the percentage of unmodified text is wrong. For God sake, 90% unmodified text doesn't mean automatically wrong translation. I can give you a list of 2500 commun words in English and Romanian (for example: Eng. medical = Rom. medical, Eng. lunatic = Rom. lunatic, merit = merit and so on. Did ever think that your Content Translation has a percentage of right translation of about 80% and that the remaining of 20% is the work of the translator? Voila how the analyser is giving the 80-85% of unmodified text!

Pginer-WMF (talkcontribs)

Hi @Grintescu,

Content translation applies the limits as described in this page. Currently there is an error in the way those are communicated, but the limits are applied as described in the page. So the problem with your translation seems to be that there were 10 or more paragraphs each with 95% of Machine translation (provided that you marked the warnings as resolved to let the tool know it was reviewed, if not you may get the error for 10 paragraphs with 80% or more of unmodified machine translation).

Unmodified content is not a guarantee of the translation being wrong, but it is an indicator that we can adjust to the needs of each language. It is also possible that we are counting modifications that are not expected to be modified (such as math formulas), that's why I asked for details on the kind for content for which you were getting warnings.

We can adjust the limits. Please, check the page linked above and feel free to propose any suggestion that you think can help to prevent translations abusing of machine translation, while not interfere with those translators making a good job.

Thanks!


Grintescu (talkcontribs)

I checked. Thanks a lot.

Reply to "The analyser of the percentage of the unmodified text is completely faulty, is a mess."