We have no mechanism for users to provide direct feedback on wrong translations.
Description
When translating from a sign language video to spoken language text, when uploading a video, we translate into spoken language.
The translation is likely to be wrong, at the moment, in most languages, in most scenarios.
Let's assume the translation here is wrong:
There should be a button on the output side, perhaps on the right side, named "Donate a sign" , like this
This button asks the user if they have the rights to the video, and if so, allows the user to suggest an alternative translation, and uploads the video to sign.mt's SignTube account with the suggested caption.
If the data was donated, the button can change to:
- maybe even linking to the SignTube video.
We then can train on this new video etc...
When the user suggests an edit, it shows:
Alternatives
Google Translate has a different solution:
easy: rate the translation. bad translations could be reviewed by trusted users / reviewers
harder: suggest an edit
Additional context
Possibly generalize this to allow donations in the spoken-to-signed direction as well, but that has more variation.
Problem
We have no mechanism for users to provide direct feedback on wrong translations.
Description
When translating from a sign language video to spoken language text, when uploading a video, we translate into spoken language. The translation is likely to be wrong, at the moment, in most languages, in most scenarios.
Let's assume the translation here is wrong:
There should be a button on the output side, perhaps on the right side, named "Donate a sign" , like this
This button asks the user if they have the rights to the video, and if so, allows the user to suggest an alternative translation, and uploads the video to sign.mt's SignTube account with the suggested caption. If the data was donated, the button can change to:
- maybe even linking to the SignTube video.
We then can train on this new video etc...
When the user suggests an edit, it shows:
Alternatives
Google Translate has a different solution:
Additional context
Possibly generalize this to allow donations in the spoken-to-signed direction as well, but that has more variation.