Robohelp 2020 using machine translation with links into sentences

[justem88826319]

Hi,

Using the new feature on sentences containing links doesn't give same output than the ones without links.

As example, using french language as source on those sentences :

• Texte du filigrane inséré lors de l'export du document depuis le portail documentaire. Ce texte correspond à la valeur du paramètre TEXT_FILIGRANE.
• Texte du filigrane inséré lors de l'export du document depuis le portail documentaire. Ce texte correspond à la valeur du paramètre TEXT_FILIGRANE.

Give those different sentences as english result :

• Text of the watermark inserted when the document is exported from the documentary portal. This text corresponds to the value of the TEXT_FILIGRANE setting.
• Text of the watermark inserted during the export of the document since the documentary portal. This text corresponds to the value of the setting TEXT_FILIGRANE.

As we can see, the translation without the links is using whole sentence as a bunch and is more accurate.

The one with links translating the "parts" without considering the sentence in it's whole (as we can see on the "since" that should be "from").

Is there any option to define that display text from link should be use into the sentence and not splitting it into parts for translation (we can see into the XLIFF generated file that there is different parts to translate) ?

Should be a good enhancement if it's not possible for now ?

Thanks for your help.

[Stefan-Gentz]

This is a "classic" (MT) translation problem.

Is there any option to define that display text from link should be use into the sentence and not splitting it into parts for translation (we can see into the XLIFF generated file that there is different parts to translate)?

Actually, RoboHelp does it like that already. The XLIFF RoboHelp produces is correct and according to the XLIFF Standard. The source sentence is in trans-unit/seg-source/mrk element and forms one translatable segment. The hyperlink is one g element inside the mrk element and is to be treated as an inline element by an XLIFF reader.

Any professional XLIFF-compliant CAT tool will handle this correctly (e.g. Across, MemoQ, SDL Trados Studio, XTM, Memsource, etc., to name a few).

If your MT system struggles to read the XLIFF according to the XLIFF standard, my suggestion is to first read the XLIFF files into a professional CAT tool, and then run the MT process from there.

[justem88826319]

Hi,
 
I forgot to add in my first post that i'm using the Robohelp bing provider as MT system.

I agree with you that XLIFF file is correct. As i said in my post i saw the g element for the links when there is one.

To be more accurate, my question is about "how robohelp processes the trans-unit using the MT". Seems it process each g element one by one and concatenates the result.

Using memoQ, i see that result is a bit different and more accurate (using same azure MT and keeping tags and styles on memoQ configuration).

With my previous example :

• Texte du filigrane inséré lors de l'export du document depuis le portail documentaire.

I get :

• With robohelp : Text of the watermark inserted during the export of the document since the documentary portal.
• With memoQ : Text of the watermark inserted during the export of the document from the documentary portal.

Without links, they both give the same "Text of the watermark inserted when the document is exported from the documentary portal.".

By the way, Robohelp MT is quite good so far when there isn't too much links in sentence. My thought is to avoid using another software if we can.

If you need i can join a XLIFF file.

Thanks

[justem88826319]

I did some more testing and it's working like with links for any style in a sentence (as there are g segment for them).
 
I undestand why when i watch the XLIFF file, but it's seems to be more accurate that we need to translate whole trans-unit and write it back with the style...
 
Best practices are to avoid too many links or styles into sentences?