Machine translation scoring behaviour and LLM-based automatic suggestion services #21527
Unanswered
michael-smt
asked this question in
Q&A
Replies: 1 comment 2 replies
|
The scoring for translation memory is based on the match. I think some external services (MyMemory) also utilize this. For all others the scoring is a static hint to order the result. This mostly comes from time when we could somehow rank the machine translation services based on our experience, but it is not really useful with LLMs. There are existing issues that would cover ordering: |
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
I'm trying to wrap my head around how the scoring works with machine translations, specifically these two sentences in the priority of machine translation and translation memory services docs:
Additionally, if I'm reading the code right, for LLM-based services it seems to be that:
So what happens if you let multiple LLM-based services bulk translate at once? Which one "wins"?
All reactions