Hacker News Reader

Comment by crazygringo | Hacker News Reader

> and I have never once seen spam or abuse in them.

That's because the spam is a problem the content creator has to deal with. According to the article, captions need to be approved by the creator before the public sees them, so you'd never know.

View on HN

I'm a content creator (fast.ai), and have dozens of multi-hour long videos with community-provided translations and transcriptions. I have never seen any spam or abuse of any form.

Removal of this feature is going to impact our international students very badly. Access to technical education was already hard enough - now it'll require English fluency too.

But as I suggested in another comment, surely other sites will provide the same crowd-sourced service, if there aren't already?

YouTube isn't removing existing subtitles or the ability to have subtitles. You'll just need to use a different platform for people to contribute them, no? And then the creator will still be able to add them to YouTube.

Ideally there won't be any negative impact on international students at all.

It still makes access harder, YouTube already had the feature built in but now you have to go to another site?

On the other hand, other sites are much more likely to build better features for the subtitling/captioning and building a community around it.

So I really do think this will be a net benefit in the long run.

The spam and abuse happens on very large channels where incentives change quite a lot. E.g. pewdiepie had to disable community contributions because of people sneaking in advertising or spam in the middle of legitimate looking subtitles.

I guess it also depends on your audience. It's no surprise to me that someone like pewdiepie attracts the sort of people who would do that.

The solution is to let individual creators turn the feature on and off, which they could already do, not nuke it for the whole site.

I wonder if they couldn't have solved that in some technical manner too. For example have the autogen transcripts first by default, but allow the opportunity for community contributions to first curate the autogen transcripts and correct any devations where the autogen is wrong. So we start with preclassified speech and correct that small fraction of the time the autogen is mistaken. I figured they could have even used those delta transcripts to improve training on. Videos with huge error rates could flag the contribution for human review, and even with those ones that faithfully correct have some human review them quickly since it may be a word or two. I usually catch at least one discrepancy every video I watch with captions that's longer than say 10 minutes. More frequently on technical subject matter too.

Reply on news.ycombinator.com