Hi Paul,

I have actually heard back on this from Studio Support, and they explained that this is actually expected Studio behaviour as per the following logic:

In order to simplify things, here’s what I used in my tests:

TM1

TU1

(source) This is a test – (target) This is a test

TU2

(source) This is a test – (target) This is a Test

 TM2

TU3

(source This is a test – (target) This is a test

Scenario 1

The first TM is the master memory, and the second TM has a 1% penalty applied to it. The “multiple translations penalty” is also set to 1%. In this first scenario, the results you receive are both 99% fuzzy matches and they are TU1 and TU2. TU3 is not even displayed because it is identical to TU1 and by default it has a lower value than it (because of the TM penalty).

Scenario 2

The first TM is the master memory, the second TM has a 1% penalty and the “multiple translations penalty” is set to 2%. Studio initially checks the first TM and applies the 2% penalty so TU1 and TU2 are “cached” with a 98% match. Studio then looks at the second TM and applies a 1% penalty (the TM penalty) to TU3. The “multiple translations penalty” option will not kick in here because it is redundant to compare results coming from different TM’s especially if one of them already has a TM penalty applied to it. This means that Studio now has “cached”: TU1 - 98%, TU2 - 98% and TU3 – 99%. Because TU3 and TU1 are identical, Studio will display TU2 as a context match and TU3 as a 99% fuzzy.

Scenario 3

You delete TU1 from the first TM, TM2 has a 1% penalty applied to it and the “multiple translations penalty” is set to 1%. Like in scenario number 2, the “multiple translations penalty” option will not kick in because you already have a penalty applied to one of the TM’s from the project. So the results displayed will be TU2- CM and TU3 – 99%.

To simplify things, the “multiple translations penalty” is not applied across all TM’s in a project, if you have TM’s that have a TM penalty applied to them. This makes sense as the user is not “bombarded” with a lot of TM results, some of them being absolutely identical the only difference being a lower penalty applied to them.

Because this is a “special” workflow, you should also increase the TM penalty to 2% if you would like to have similar results as before.

And I must admit, this does actually make sense when explained this way. So we are now running tests with a 2% penalty on all TMs. It still doesnl;t give us exactly what the client had hoped for, but is all looking a lot better