Ehsan Abidi ashtiani

Fragment Matching works on the basis of the alignment process being able to align fragments of source and target texts with a degree of surety.  The process isn't as clever as a real person and doesn't always know with enough certainty whether or not a particular word or phrase has a matching translation.  The larger the TM, the more the subject matter is the same, the better the fragment matching... potentially.

So you won't always get suggestions for everything.

How big is your TM?  What settings are you using?  Have you got this checked in your project settings?

Thank you for answering, helpfull as usual.

My knowledge about TU Fragment has increased from when I ask this question. Now I know a sentence should be repeated in multiple TUs for the Trados`s AI to find it`s translation.

In second message, my question`s subject has changed to Whole TU. Whole TU has less settings and is active by default. Why the existing whole TU doesn`t shown in the Fragment Matches window or when I type the first letter of translation (Autosuggest)?

Ehsan Abidi ashtiani

but this would require a fragment and not a whole TU because you are asking it to match part of a sentence, not a whole TU. So I would say the same argument applies.

In according to my last picture, you are saying, sentence:

Clinical trials include only

which is in Translation Unit number 3 is not a whole TU?

Ehsan Abidi ashtiani

In your second screenshot it's a whole TU.  In your first screenshot, which is the one you are expecting fragment matches for it's not.  It will only be picked up if the fragment matching exercise on your TM has recognised that particular fragment within the sentence:

"In some labs result, Clinical trials include only, is written on the results sheet"

If you want to send your TM to me and your source file I'll be happy to take a look?

Thank you for answer.

You see what I see in my second screenshot, but my first screenshot is not the translation memory; it`s the editor.

I know it’s in the editor.

You say a segment in the editor should be filled by the sentence from begin to end!

So it`s name should be "Whole Segment" not "Whole TU". If both sides are whole, so it would be 100% match.

No. I’m saying that the match you are expecting is a fragment of the source. The match may be a whole TU but it’s still matching a fragment of the source.  If the software was unable to match this fragment with your whole TU then it won’t pick it up.

Why don’t you provide a screenshot of all your settings? Maybe there is scope to search deeper?

Also, how large is your TM?

Paul said:

Why don’t you provide a screenshot of all your settings?

Because the Whole TU should work without special setting. as this blog say:

"Whole TU unit recall is the only feature that will work without upgrading your TMs"

My initial question was some expected sentence wouldn`t find by the software. In process of solving my problem, I simplifed it to Whole TU only.

hear is this project settings:

Ehsan Abidi ashtiani

Thanks for providing more details.  I can't see anything wrong with your settings, so I would have just gone back to my original explanation to you:

Paul said:

Fragment Matching works on the basis of the alignment process being able to align fragments of source and target texts with a degree of surety.  The process isn't as clever as a real person and doesn't always know with enough certainty whether or not a particular word or phrase has a matching translation.  The larger the TM, the more the subject matter is the same, the better the fragment matching... potentially.

So you won't always get suggestions for everything.

When it comes to matching this would include whole TUs.

However, I spent some time yesterday creating a TM from a TMX with around 80k TUs and then testing various different scenarios with source files where every segment was either a 100% TM match or a sentence containing the same TM match in the previous segment.  I have been able to reproduce a few things that I cannot explain, so I've taken this up with the development team and will report back when I know more.

Thank you for your patience and for providing these details.

Ehsan Abidi ashtiani

Thanks for providing more details.  I can't see anything wrong with your settings, so I would have just gone back to my original explanation to you:

Paul said:

Fragment Matching works on the basis of the alignment process being able to align fragments of source and target texts with a degree of surety.  The process isn't as clever as a real person and doesn't always know with enough certainty whether or not a particular word or phrase has a matching translation.  The larger the TM, the more the subject matter is the same, the better the fragment matching... potentially.

So you won't always get suggestions for everything.

When it comes to matching this would include whole TUs.

However, I spent some time yesterday creating a TM from a TMX with around 80k TUs and then testing various different scenarios with source files where every segment was either a 100% TM match or a sentence containing the same TM match in the previous segment.  I have been able to reproduce a few things that I cannot explain, so I've taken this up with the development team and will report back when I know more.

Thank you for your patience and for providing these details.