Note that some language pairs often creates flawless translations, like Bokmål-Nynorsk by the ml engine Apertium. For those language pairs there should not be a limit on publishing unmodified text.

See also T190279#4114812

Change 447583 had a related patch set uploaded (by Santhosh; owner: Santhosh):
[mediawiki/extensions/ContentTranslation@master] Prevent publishing translation with too much unmodified content

https://gerrit.wikimedia.org/r/447583

Screenshot from https://gerrit.wikimedia.org/r/447583

Note that this patch does not include the issue cards in tools column. Waiting for the issue card refactoring to complete before attempting that.

Now blocked by T200503: CX2: User modified content is saved as original machine translation

Change 447583 merged by Petar.petkovic:
[mediawiki/extensions/ContentTranslation@master] Prevent publishing translation with too much unmodified content

https://gerrit.wikimedia.org/r/447583

Checked in CX2 - three things seem to be missing:

(1) too much unmodified text ≠ too much unmodified machine-translated text (?)
Right now I see "Your translation cannot be published because it contains too much unmodified machine-translated text" when I tried to publish the articles that are blank (when I used the option "Don't user machine machine translation" and just click 'Add translation' to add blank paragraphs.
Also, the same warning appears when I use just source page for translation. Interestingly that the issue summary card appears blank.

(2) Threshold for unmodified content - it seems not possible to publish articles even with 50% of un-modified content and even when only4% of the article is translated and 86% of it is machine-translated. What are thresholds in cx2? Does it take into account what portion of an article is translated? According to the following spec, the specific threshold should be set:

We may want to apply the threshold only when there is enough content to avoid confusions with false positives (e.g., publishing a one sentence article with a few images).

(3) 'View issues' is not shown on the warning.

I filed T202342: CX2: "Too much unmodified content" warning issues (and added it as a subtask to this task) mostly to keep track of the issues related to the "too much unmodified content" warning as a separate ticket since there are some specific issues that may be addressed outside of the scope of this task.

I am returning this task to 'In progress' column, so @Pginer-WMF may review the specs or/and "View issues" option will be implemented.

In T190283#4517107, @Etonkovidova wrote:

Checked in CX2 - three things seem to be missing:

Thanks for checking this. Some comments below:

(1) too much unmodified text ≠ too much unmodified machine-translated text (?)
Right now I see "Your translation cannot be published because it contains too much unmodified machine-translated text" when I tried to publish the articles that are blank (when I used the option "Don't user machine machine translation" and just click 'Add translation' to add blank paragraphs.
Also, the same warning appears when I use just source page for translation. Interestingly that the issue summary card appears blank.

It makes sense to prevent publishing an empty document, but the "too much unmodified text" is not the right message to show in such cases since it can lead to confusion. We may need a specific solution for such case.

(2) Threshold for unmodified content - it seems not possible to publish articles even with 50% of un-modified content and even when only4% of the article is translated and 86% of it is machine-translated. What are thresholds in cx2? Does it take into account what portion of an article is translated? According to the following spec, the specific threshold should be set:

We may want to apply the threshold only when there is enough content to avoid confusions with false positives (e.g., publishing a one sentence article with a few images).

@santhosh may know more about the thresholds and how are they applied. I'll try to test with some examples to identify how

(3) 'View issues' is not shown on the warning.

I filed T202342: CX2: "Too much unmodified content" warning issues (and added it as a subtask to this task) mostly to keep track of the issues related to the "too much unmodified content" warning as a separate ticket since there are some specific issues that may be addressed outside of the scope of this task.

Thanks for capturing these in a ticket!

@Pginer-WMF @Etonkovidova About the false positives, we need to fine tune the calculation based on the characteristics of section. For example, we currently exclude section titles. There may be few other cases we need to identify and apply our threshold in a different way. I had created T200416: CX2: Identify section types to exclude from MT abuse test to explore such requirements.

In T190283#4541913, @santhosh wrote:

@Pginer-WMF @Etonkovidova About the false positives, we need to fine tune the calculation based on the characteristics of section. For example, we currently exclude section titles. There may be few other cases we need to identify and apply our threshold in a different way. I had created T200416: CX2: Identify section types to exclude from MT abuse test to explore such requirements.

In my experience testing in this area, the error that prevents from publishing disappears after some minimal modification of the content; and the warning does not block the publishing process and can be marked as resolved in case it was too sensitive. So my understanding is that in the current state this should not cause major problems. As you suggested, we can keep testing and proposing adjustments to polish this further in T200416

Follow up on the following issues:

• the specific warning for publishing blank translation - the warning still referring to unmodified content:

• the state of 'Publish' button when the summary issue card has not-resolved issues - it's possible to publish

• the issue summary card refers to machine translation even though no MT options were used:

• the thresholds for unmodified (blank) content for preventing publishing translation (need more testing)