John Narins

John Narins said:

I'm guessing that updating a TM for upLift does NOT create new TM entries for fragments.

You guessed wrong.  This is exactly what it does and is the main reason why the size of TM increases substantially when you do this.

John Narins said:

all the parameters I can set are not for the TM, they're for the project, so apparently I can set them (e.g., the minimum size/# of words/fragment) per project, which means that the upgrading works with any setting, right?

These setting relate to how much noise you want to see with the number of returned results.  They have nothing to do with the UpLift process itself.

John Narins said:

If I add new TUs to an upgraded TM, the algorithm for finding fragments in the TM as a whole is automatically expanded to account for the new TUs, right?

Right.

John Narins

John Narins said:

I'm guessing that updating a TM for upLift does NOT create new TM entries for fragments.

You guessed wrong.  This is exactly what it does and is the main reason why the size of TM increases substantially when you do this.

John Narins said:

all the parameters I can set are not for the TM, they're for the project, so apparently I can set them (e.g., the minimum size/# of words/fragment) per project, which means that the upgrading works with any setting, right?

These setting relate to how much noise you want to see with the number of returned results.  They have nothing to do with the UpLift process itself.

John Narins said:

If I add new TUs to an upgraded TM, the algorithm for finding fragments in the TM as a whole is automatically expanded to account for the new TUs, right?

Right.

Okay, so the upLift process extracts all the fragments its algorithm tells it MIGHT be wanted, and the per-project settings tells it how many of those to suggest. Got it.

How do the answers to 1 and 3 go together? In other words, if the TM update DOES add fragment entries, then... it calculates them again with each new TU? This is the reason for my incorrect guess, I think you can see what I mean.

To put it another way: how does an updated TM get new fragments as new TUs are added? Because, as I understand it, fragments come from repetitions, so a new TU by itself can't tell us whether there's a fragment worth keeping there, the algorithm is surely looking at the new TU in the context of of the whole TM, right? And then the next TU added gets analyzed taking into account the whole TM, including the previous just-added TU... how does this work?

Thanks much!

John Narins

I suggest you digest the content in this thread, and ask any questions you still have afterwards in that thread:

https://community.sdl.com/product-groups/translationproductivity/f/studio/9412/what-is-the-technology-behind-uplift

Thanks, I read through that... but it's still not clear, so I asked a question there. Thanks for your help!