JF Delannoy 

JF Delannoy said:

I want to retranslate from (another) TM a text where segments (some, or all of them) already have a confirmed translation.

Just pre-translate with this option... or a configuration to suit your purposes against these batch processing options:

JF Delannoy said:

Ctrl-T is slowed by the obligatory display, each time, of the Translation Results panel (which I do not need - I am confident about the TM - Can this be deactivated?

If you are applying a translation without even looking at the results first then why don't you simply pre-translate the entire project before you start with the pre-translate batch task and then just disable the lookup:

Like this you can close the TM Results window and it will never come back.  Concordance will still work for you, but the TM Results window won't.

Thanks.

As a test, I edited a few target segments to dummy text (aaa, bbb, ccc),
set the matching threshold to 100%
disabled Lookup,
and ran Pre-translate again.

I expected the empty segments to be populate, but they remained blank.

(NOK)

JF Delannoy 

JF Delannoy said:

disabled Lookup,

Well, you can't have it both ways... if you disable it then it won't lookup!  If it's enabled by default then every project you create will always run the pre-translate as you see fit and you then disable it once the project is ready to start work.  If you work with many TMs, and intend to constantly switch between enabling and disabling (although your confidence in your TMs suggests you won't need to do this) then maybe consider working with project templates.  Then you can use the ApplyProjectTemplate app to quickly replace your settings without having to go in and check, uncheck to make these changes.

Or, do what Jesús Prieto suggested here -  RE: how can I completely hide the Translation Results Window? - which might be a better solution anyway.

JF Delannoy 

Also... you presumably did use this option?

It is an unfrequent need; I don't mind changing the settings. Still, preparing ad hoc templates may simplify it.

My test works now, having chosen the option Always Overwrite.

Thanks.

(NB I don't situate the application of the criterion "Overwrite... if a better match is found. Translating EN Territory by FN Territoire seems a better match than FR aaa as edit distance goes.)

JF Delannoy 

JF Delannoy said:

I don't situate the application of the criterion "Overwrite... if a better match is found. Translating EN Territory by FN Territoire seems a better match than FR aaa as edit distance goes.

Who said edit distance has anything to do with this?  What was the match value in the TM and what was the match value of the new one in the TM, also what was the status of the segment?  You're testing so you may well have "fooled" the system!

I did not find a rigorous description of the match score algorithm.

This page,
www.linkedin.com/.../
mentions three criteria for Fuzzy matching scoring:
- words in common
- word order
- tags (embedded codes).

Is there some morphology as well? (number, gender, verb tense, compound words as in DE, affixes as in JP...)

Is there a more technical/linguistic account somewhere?


JF

I did not find a rigorous description of the match score algorithm.

This page,
www.linkedin.com/.../
mentions three criteria for Fuzzy matching scoring:
- words in common
- word order
- tags (embedded codes).

Is there some morphology as well? (number, gender, verb tense, compound words as in DE, affixes as in JP...)

Is there a more technical/linguistic account somewhere?


JF

 JF Delannoy 

JF Delannoy said:

Is there a more technical/linguistic account somewhere?

Only the product help: https://docs.rws.com/980998/548323/trados-studio-2022/specifying-batch-settings--pre-translate-files... and it's not particularly technical:

• Overwrite existing translation if better match found, so that Trados Studio replaces the existing translation only if it finds a better translation in the TM. When you use this option, Trados Studio does not replace translations in segments which are set to Translated.

JF Delannoy said:

Is there some morphology as well? (number, gender, verb tense, compound words as in DE, affixes as in JP...)

No.  Studio doesn't perform any morphological processing for TM matching.  It works in a very similar manner to the old Trados Workbench. 

On this observation you have made:

JF Delannoy said:

Translating EN Territory by FN Territoire seems a better match than FR aaa

It sounds as though you're expecting Studio to make a judgement on the actual translation, which it doesn't do at all.  It is only comparing source to come up with the match value, and it will compare source and target of the previous segment to help distinguish a Context Match.  It should also pay attention to the Document Structure Information... but I think this was partially broken a long time ago... nobody noticed... and it never got priority to fix.  So a few examples:

A 95% match in your current file will be replaced by a 96% match with a TU in your new TM... because the match is higher.

A 100% match in your current file will be replaced by a CM match with a TU in your new TM... both are 100% matches but the CM has the same context.

A 82% match in your current file (where the TU was saved translating a list item) should be replaced by a 82% match with a TU in your new TM (where the TU was saved translating a heading)... as long as the current segment in your file is also a heading (document structure information).

A 90% confirmed translation will not be replaced by a 91% match in your new TM if the current segment has a translated status.

It's nothing more complex than this sort of comparison for the decision over whether a match is replaced or not.  The quality/suitability of the translation doesn't come into it at all.  To do that we'd have to rely on machine translation to tell us if it's better.

The no-linguistics approach was a disincentive when I started considering MT.
(I do research in linguistcs).
Maybe a tad of morphology would help.
Semantics would be another story.