If I understand it correctly, it means that all reported sentences are never really hidden from contributors on project level. Everyone can hide sentence for themself, but not for other contributors.
It is a really strange decision, because it means huge amounts of data quality issues and makes many problem for contribution process.
First of all, it is really hard to sort all these reported sentences if you are a researcher.
And if you are a contributor, it means you will often meet low quality sentences, that significantly decreases chances you will stay as the constant project contributor.
So my question is if it is possible to take down some sentences from the project at all or at least hide them from getting new recordings/hide for new contributors?
Current reality is what I wrote on the other topic, but the workflow might change in a couple of months.
The main problem here is “it is widely subjective”. E.g. an edge example: define “vulgar”… Everybody will have another threshold for it.
If you look at the `reported.tsv`, you will see valid ones and invalid ones. So, before training an AI engineer’s job is to scan them and re-categorize, have a blacklist of those real valid for your model (and it might depend on your model and what you are doing). Leave the blacklisted ones out of your training.
This was the idea - you flag it for others to decide.
So my question is if it is possible to take down some sentences from the project at all or at least hide them from getting new recordings/hide for new contributors?
Of course, through an issue and/or PR. Here is an example: