Hi Elisa Wenceslá 

The documentation around this is missing indeed!  I'll risk a brief explanation for now:

Unknown said:

Fragment words (whole TU)

These are the number of whole TU fragments found.  So where a complete TU is found within a longer segment which would make it a fuzzy match or lower in the analysis results.

Unknown said:

Fragment words (TU fragment)

The number of Fragment words found.  The number of times relevant fragments appear in a segment.  I think these fragment figures will provide an indication of how useful a TM could be even if there are no relevant fuzzy or 100% matches purely because of the productivity gains with autosuggest.  So it's useful if you have multiple TMs as you may find one is more relevant for a particular domain than another; and it might help with estimating time for a job because poor leverage generally and no fragment matching means you really are on your own, whereas poor leverage generally and high fragment matching means you can still expect to get some help from your TM while you work.

Unknown said:

AdaptiveMT Impact

This is related to the "AdaptiveMT with learnings".  The percentage is an estimate of the reduced post-editing required for the number of segments in that row.  So it takes the original baseline MT, looks at the changes due to the Adaptive engine and makes it's assessment.  This is intended to show you how much use your are getting from your Adaptive engine in a particular text.  Could be valuable if you have several Adaptive engines in use and want to see which one is potentially going to give you the best results for your project.

Unknown said:

AdaptiveMT Baseline

This is just telling you how many words/segments etc. have been handled with machine translation based on the baseline engine.

Unknown said:

AdaptiveMT with Learnings

I explained this above, but will reiterate that I see this as an indication of how much value you might get from your Adaptive engine.  With the right domain and the right amount of training the theory is that this percentage should increase.  The higher it is the more confidence you might be able to have in a reduced post-editing effort.  So it's designed to help you assess your effort which is part of what the analysis is all about.

Unknown said:

New/AT (why are these together in the same category?)

I agree this could be a little confusing, but AT is really "Automated Translation" and in this context I believe it relates to anything that was automatically handled by the Studio TM, but not Machine Translation.  So number only segments for example could be draft with AT status and are in effect new translations for you as they were not confirmed through being in your TM.  I believe this is the idea anyway.

Daniel Brockmann might be able to clarify if I've gone wrong somewhere.

Regards

Paul

Hi Elisa Wenceslá 

The documentation around this is missing indeed!  I'll risk a brief explanation for now:

Unknown said:

Fragment words (whole TU)

These are the number of whole TU fragments found.  So where a complete TU is found within a longer segment which would make it a fuzzy match or lower in the analysis results.

Unknown said:

Fragment words (TU fragment)

The number of Fragment words found.  The number of times relevant fragments appear in a segment.  I think these fragment figures will provide an indication of how useful a TM could be even if there are no relevant fuzzy or 100% matches purely because of the productivity gains with autosuggest.  So it's useful if you have multiple TMs as you may find one is more relevant for a particular domain than another; and it might help with estimating time for a job because poor leverage generally and no fragment matching means you really are on your own, whereas poor leverage generally and high fragment matching means you can still expect to get some help from your TM while you work.

Unknown said:

AdaptiveMT Impact

This is related to the "AdaptiveMT with learnings".  The percentage is an estimate of the reduced post-editing required for the number of segments in that row.  So it takes the original baseline MT, looks at the changes due to the Adaptive engine and makes it's assessment.  This is intended to show you how much use your are getting from your Adaptive engine in a particular text.  Could be valuable if you have several Adaptive engines in use and want to see which one is potentially going to give you the best results for your project.

Unknown said:

AdaptiveMT Baseline

This is just telling you how many words/segments etc. have been handled with machine translation based on the baseline engine.

Unknown said:

AdaptiveMT with Learnings

I explained this above, but will reiterate that I see this as an indication of how much value you might get from your Adaptive engine.  With the right domain and the right amount of training the theory is that this percentage should increase.  The higher it is the more confidence you might be able to have in a reduced post-editing effort.  So it's designed to help you assess your effort which is part of what the analysis is all about.

Unknown said:

New/AT (why are these together in the same category?)

I agree this could be a little confusing, but AT is really "Automated Translation" and in this context I believe it relates to anything that was automatically handled by the Studio TM, but not Machine Translation.  So number only segments for example could be draft with AT status and are in effect new translations for you as they were not confirmed through being in your TM.  I believe this is the idea anyway.

Daniel Brockmann might be able to clarify if I've gone wrong somewhere.

Regards

Paul

Unknown said:

I understand that any category related to MT will provide results only when working with MT engines. Otherwise, it will just appear as empty, right?

Correct.

Unknown said:

And I guess there's no way to remove those categories from the analysis...

Also correct as far as I know.

Thanks very much, Daniel!

Kind regards,

Well, the thing is that the term "baseline" is established terminus technicus in MT world, so using anything else would be actually much more confusing... And the term really means 'a starting level', not distinguishing between adaptive and non-adaptive engine.