Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Do openKiwi have confident score? #103

Open
baoloc97 opened this issue Jul 2, 2021 · 1 comment
Open

Do openKiwi have confident score? #103

baoloc97 opened this issue Jul 2, 2021 · 1 comment
Assignees
Labels
question Further information is requested

Comments

@baoloc97
Copy link

baoloc97 commented Jul 2, 2021

I use openNMT for translate but its dont have confident score when translate so they suggest me use openkiwi. But i dont see confident score in openKiwi too ? Confident score it mean how many percent that my model think that sentence is predict right

@captainvera
Copy link
Contributor

Hello @baoloc97,

Sorry I took a while to respond.

Openkiwi does not produce a confidence score in that sense. When you are talking about a confidence score from a MT model, it can be thought of as the uncertainty of the model on inference time. It can be obtained with methods such as monte carlo dropout and similar.

On the other hand, what OpenKiwi does it Quality Estimation, it provides models that specialise in looking at parallel corpora and producing "quality scores" for this data. Either at word-level (with Ok/BAD tags) or at sentence level (by predicting HTER), this functions as a proxy of MT quality when you don't have a reference, with results from wmt20 metrics shared task showing a high correlation between QE and metrics.

@captainvera captainvera self-assigned this Jul 20, 2021
@captainvera captainvera added the question Further information is requested label Jul 20, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

2 participants