I think that’s it. Yeah, we’ll see you tonight at dinner. Thanks for having us.

Yeah.

Not me. I’m new five years go. Dees is the guy.

That’s exactly how we do it.

No.

We hope that we can continue to work together not with just our values but on actual goals in terms of this language and cultural preservation. Thanks for having us today. We hope that we can continue to work with you as we move forward.

It’s very much appreciated. This sort of work. This is how we tend to think about things. We try to be very intentional about how we’re working so that others can get involved. It’s a very hard process, I imagine, at the level that you’re working, this is a big ...

We’ll do that. I can’t imagine there’s any problems but we’d love to see how that works. Sounds great. Thank you, by the way, for your, is it radical transparency?

I have not.

It sounds like a no-brainer that we would like to do it, even though pure symbolic, would you need support from us? How do these things work?

...because that’s the statement we’re making.

We obviously have to do it in all the other indigenous languages...

We’d love to make ourselves available to that if you were to help set up the connections. We can provide an application. If there’s even changes to the product itself that would help support the education effort, we would love to talk about those things.

Yes, of course.

Great.

Yes, we’re all CC0.

Yes, please. If I could wear a pin that said that.

We will use this. People will be speaking non-words and common words maybe.

For instance, there can be special ribbons around mirroring that data, which we have worked on in the past. I would say this is a conversation that we can continue to have and figure out if there are ways that we can satisfy both the direction that we are taking...

It’s something we can continue to talk about. We would love to. It sounds like we’re values-aligned, at least, on how to do these. There are ways to work together, even if we don’t agree on the license.

Our hypothesis is most people don’t have a legal support for these sorts of things.

If you’re the size of Apple, it’s not so hard, because you can have a whole...

We are continually talking about this with partners. At some point, we may change and say we want to support attribution at some point. For now, it was a great starting point to say let’s open this data pure public and let’s see if we can measure the impact, first ...

We think that CC0 can actually unlock a lot of potential for data sets. We think that you can still get the things you want from the attribution maybe without having the actual attribution license because this reproducibility. This is our theory. This is what we’re going off of.

We see CC0 as a way of, we still get a attribution. We still have ways of measuring our impact in the academia world, but it’s also way of having the non-academia world, the commercial world have the most impact there. That’s what we care about.

It could prevent somebody from making a doctor personal voice assistant from using our data and make it a good thing because they would have to adhere to our license, which is a public, which may be like attribute Mozilla and maybe that doesn’t meet their business standards for some ...

If their paper’s reproducible, they have to quote to their source, so we can get attribution from the academic realm. For reading speech technology in the commercial realm, we don’t necessarily need the attribution, right, because what if these products are very privacy-centric products?

Exactly, proof of stake, right? What we found with Common Voice is that at least...Maybe there’s stuff we don’t know about, but in general, we know about two papers that were written with Common Voice data. They will all always print this anyway because they want their paper to be ...

It seems like people choose attribution because they want to, first of all, they want to be recognized for this hard work that they did. They want to be able to incite it when their work is used, and they want to be able to get grant money when their ...

We think CC0 or public domain is this way of solving it by making it easy. Just have it be open and use it however you want. We think that this attribution piece, we’ve done some research as to why people choose attribution, at least in university.

If I was a university student and I wanted to make a technology that is commercial, let’s say, then I’m basically stuck or it’d be very hard for me understand the technology I’m creating actually adheres to these licenses. Because this license, when you’re talking about remixing data, it comes ...

It’s hard to, first of all, remix the data set and adhere to the license. It makes it prohibitively expensive, let’s say, in people time. Second of all, to try to get people to the public domain, you have to be able to get in front of people. Luckily, we ...

We instead tried to get them to go for CC0. First, we had to talk to Sam, and then we had to talk to the owner of his derivative data sets. There was this woman in University of Illinois who had this other data set. We had to try to ...

This is hard to do and it prevents what we think is innovation or prevents people from remixing it easily because they have to try to understand this license, how these different licenses apply to different parts of the data set and preserve that dependency.

For instance, we could change Common Voice to support various forms of attribution, but then we have this very complex structure of attribution that we have to follow. Anybody who remixes our data set has to also trace those sources of data and include that in their license.

Attribution 4.0. Also, Share-Alike. He said, "I’d be open to publishing CC0, but we’re based on these other data sets, specifically this one called Flickr 30k. This is an attribution as well. If they were to open-source it, we can open source for ours."

This is a couple hundred thousand sentences. This is, in itself, a mix of some other data sets. We went to the person who is curating the data sets, Sam Brown. He’s at NYU now. We said, "Can we use your data set in Common Voice? The license doesn’t work ...

I can explain a little bit about what that means. Let me give you an example. Common Voice is a remix of many other data sets because we have to have people speaking sentences. A lot of times, those come from other data sets. There’s one data set in particular ...

This is a continual discussion, so there aren’t any right answers that we have hard and fast decided on yet. Our experience of attribution so far with creating Common Voice data set is that attribution, at its core, creates this sort of license baggage.

We feel very strongly about public domain for data and we think that data in pure public domain has the potential to have the most impact.

In fact, the tools that we’re building that’s exactly, as you said, and Kelly called this out. We see it as a pipeline with documentation for these communities to build it up based on how quickly they want to move. I see a lot of overlap there. What I’d love ...

It’s a research project, right, like the distance between...

A big effort.

How many languages is...?

Sounds great.

We would consider Taiwan and its various languages part of the community. That’s how we see it. As Dees said, those are longer-term things that we would love to continue to talk about. Shorter term, lets help each other build this data set if we can all agree on public ...

The things that I mentioned, those are immediate. We could start working on this today, basically. But as our relationship develops, if we decide that our visions are aligned and our goals are aligned, we would love to talk about how we could evolve Common Voice and Deep Speech, the ...

As Dees said, longer term, we would love to explore maybe tools for helping teach languages. Part of our work in the next year is gamification of our websites. If there’s a use case around accent training, spoken language training, or learning to understand spoken language, we would certainly work ...

These Latin-based indigenous languages would be great, as well to get people reading and creating data sets on Common Voice itself. If you are open to creating these databases for these indigenous languages with an open license and public license, we could help you get that spoken and create a ...

That’s another thing that we would love to help with. Of course, we would love to link to your data set, in general, and just say, "Hey, here’s a great Mandarin, traditional Chinese corpus that you can use." Those are two things. The third thing is, in general, we are ...