WEBVTT 00:00.000 --> 00:17.280 I want to start the way it's saying that I was not supposed to be doing this talk, we were 00:17.280 --> 00:22.540 supposed to be having Anna here from the accessibility team, which is an awesome team at 00:22.540 --> 00:26.140 Mozilla, but unfortunately her visa got rejected. 00:26.140 --> 00:33.820 So big boo for whoever rejected her visa. 00:33.820 --> 00:39.700 So then I took the time to come here to talk all of you about something else. 00:39.700 --> 00:45.980 I came from the middle of nowhere where I lived to the beautiful Brussels and can't wait 00:45.980 --> 00:48.980 to give this great talk to you. 00:49.980 --> 00:55.980 All right, let's move on and see what are we actually going to be talking about. 00:55.980 --> 01:00.780 Quick intro about localization, what are we actually doing to localize how to read 01:00.780 --> 01:06.820 Mozilla, then Mozilla language portal, that's something new, that's going to be the main 01:06.820 --> 01:15.860 the core part of this talk, then a little bit about bird box questions and then we go home. 01:15.860 --> 01:23.500 So it's really hard to say in just a few words what localization is and how we do it at 01:23.500 --> 01:24.500 Mozilla. 01:24.500 --> 01:33.500 So I thought I'd ask someone very intelligent and the answer that came up was very poetic 01:33.500 --> 01:39.540 and I decided to dig in a little bit, like what do you actually mean by that? 01:39.540 --> 01:45.980 And then the famous poet from the East actually clarified it. 01:45.980 --> 01:50.700 And I think it's a pretty good statement. 01:50.700 --> 01:55.580 The whole point of localization is to bring the open web to way more people than if we don't 01:55.580 --> 02:00.540 do localization and we can prove that with numbers, there are way more Firefox users 02:00.540 --> 02:07.100 that use translate at localize versions of Firefox than English Firefox. 02:07.100 --> 02:09.940 So localization matters. 02:09.940 --> 02:17.580 And for a brief moment, let's dive into some stats. 02:17.580 --> 02:23.020 We don't only localize Firefox, at Mozilla, there's a bunch of projects going on. 02:23.020 --> 02:30.100 You've heard about many of them today and there's a bunch of others more. 02:30.100 --> 02:36.980 Usually in pontoon, in our platform, to localize, you're going to see around 30 projects 02:36.980 --> 02:40.740 that are being worked on, sometimes the number is a little bit higher, sometimes a little 02:40.740 --> 02:43.220 bit lower. 02:43.220 --> 02:50.220 We have a total of 365 and a half double check, that number, 365 teams. 02:50.220 --> 02:55.180 So locales, the which we translate over, locales a little bit different than a language 02:55.180 --> 02:59.340 because we have several regional variants of some languages like Spanish. 02:59.380 --> 03:04.780 We have a separate localization of Mexican Spanish, Spanish, Spanish, and so forth. 03:04.780 --> 03:09.780 But 369 locales, it's a pretty huge number. 03:09.780 --> 03:16.900 I don't know if there's a lot of software or even out of software industry projects that 03:16.900 --> 03:23.260 have so many translations or like book translations or whatever. 03:23.300 --> 03:32.980 We have about 1,000 plus volunteers who in their free time make sure that Firefox Thunder 03:32.980 --> 03:39.860 Bird, AMO, MDN, Sumo is available for speakers of their language. 03:39.860 --> 03:44.900 They're awesome and I'm happy that I see some of them here and I'm happy that some 03:44.900 --> 03:48.700 of us are following us as I just read from chat. 03:48.700 --> 03:49.700 Hi. 03:53.700 --> 03:57.900 To give you an idea of the amounts, we're talking about a half a million translations 03:57.900 --> 03:59.900 submitted per year. 03:59.900 --> 04:05.260 The numbers, of course, varies a little bit year by year, but just to give you an idea. 04:05.260 --> 04:09.420 And a translation, a strength could be something very simple, even like one letter, 04:09.420 --> 04:12.860 but it would also be like a whole paragraph. 04:12.860 --> 04:17.220 Sadly, I don't have a number to give you the exact words. 04:17.460 --> 04:22.660 Okay, so what is the language portal? 04:22.660 --> 04:26.100 It's something that doesn't exist yet. 04:26.100 --> 04:28.260 Actually, I'm looking at you in Ludo. 04:28.260 --> 04:35.420 For those of you who didn't come to the contributor to meet up yesterday, I think it was 04:35.420 --> 04:39.620 you or Pascal who said something about that, we should be doing more things in the open 04:39.620 --> 04:44.140 to get the feedback, that's what we're trying to do here. 04:44.140 --> 04:53.620 We're in a very early stage of developing the Mozilla language portal, and it's a simple 04:53.620 --> 05:01.900 micro-absite, targeted at localization and translation experts globally from various industries 05:01.900 --> 05:04.300 various projects. 05:04.300 --> 05:08.900 And the basic idea is to share stuff that we do at Mozilla, our translations, or best 05:08.900 --> 05:17.060 practices, even tools with them, and make them available for their use. 05:17.060 --> 05:22.140 For example, I know that some of you are probably not familiar with the term translation 05:22.140 --> 05:23.140 memory. 05:23.140 --> 05:28.620 We'll basically collect all the translations that have ever ended up in any project that 05:28.620 --> 05:34.500 we translate, either today, either in the past, it's been like a project that doesn't exist 05:34.500 --> 05:37.380 anymore, and we store them. 05:37.380 --> 05:45.620 And that the base is called translation memory, and it's useful to Chrome localizers, 05:45.620 --> 05:50.100 open office localizers, Linux, digital localizers, everyone basically. 05:50.100 --> 05:55.580 And language portal is going to have simple search, which is going to allow people to search 05:55.580 --> 06:01.220 through it, don't lose these translations and incorporate them into their tools, as well 06:01.220 --> 06:06.900 as leverage this content through the API. 06:06.900 --> 06:11.940 Similarly, through the exact same means, we're going to make it available for people to 06:11.940 --> 06:16.260 use our glossary, so collection of terms. 06:16.260 --> 06:22.020 We have a list of terms that are frequently translated and part of translation strings, 06:22.020 --> 06:27.780 with their definitions, examples of use, and all of them are translated to maybe not 06:27.780 --> 06:32.580 360-nile accounts, but to most of them. 06:32.580 --> 06:36.860 That's going to be available similarly as translation memory. 06:36.940 --> 06:44.460 In addition to that, we have a process, the way we do localization, which is not necessarily 06:44.460 --> 06:51.820 specific for a Mozilla, there's no strong reason why some start up or some add-on developer 06:51.820 --> 06:57.100 who wants to have their piece of software localized, wouldn't be able to benefit from that. 06:57.100 --> 07:03.100 So we're going to make the relevant bits of our documentation available to people. 07:03.180 --> 07:08.780 We have a block on which we not only write about stuff that's going on at Mozilla, but it's 07:08.780 --> 07:14.060 potentially interesting for the wider industry, so we're going to make this thing available 07:14.060 --> 07:15.500 on the language portal as well. 07:15.500 --> 07:22.140 I also have style guys, which will probably come in the later states, not in the initial 07:22.140 --> 07:29.820 version of the language portal, but basically documents through which localization teams 07:29.900 --> 07:32.780 define the standards and rules under which they localize. 07:35.420 --> 07:40.780 We also have some language technology that we develop in-house. I've already mentioned 07:40.780 --> 07:47.900 pontoon, which is our TMS system, but the one that I want to highlight, I know that this URL 07:47.900 --> 07:52.700 is probably not very visible, and I didn't put a cure coat on. 07:53.020 --> 07:57.020 That's still half done, right? 07:57.020 --> 08:03.020 Yeah. The Moz Outen and Library is, I think, something that has a potential to become an 08:03.020 --> 08:13.420 industry's leading library to parse and serialize virtually anti-translation format into a 08:13.420 --> 08:19.500 data model that is mimicking the new standard message format too, you might have heard of, 08:20.460 --> 08:26.060 and as such, allows a very rich representation of virtually anti-massage stored in any 08:26.060 --> 08:33.180 format. This is actively worked on, we're going to be adding other features, checks soon, 08:34.300 --> 08:39.420 and I'm pretty sure there's a potential for lots of users for that. 08:44.300 --> 08:48.460 This is just a sneak peak of how the home page is going to look like or might look like, 08:49.180 --> 08:52.860 there's another URL you don't see at the bottom where this is being developed. 08:53.900 --> 08:59.900 Feel free to provide suggestions, file box, when we're so far that you can actually find bugs, 09:01.260 --> 09:03.740 as well as provide patches. You accept those as well. 09:08.060 --> 09:14.540 And now that I'm at the very end, I just want to co-out bird box, which is the framework 09:14.540 --> 09:23.740 that we use for building, a Mozilla language portal. It's a framework developed by our 09:23.740 --> 09:29.980 Moz Miao team, Mozilla Marketing and Operations, I think, is what the acronym stands for, 09:31.020 --> 09:37.020 specifically Steve Jalim, who allowed me to mention his name on stage. Bird box is basically 09:37.980 --> 09:43.020 helping developers to build the microsites like the one that I just discussed. 09:43.980 --> 09:50.460 With Mozilla branding, so it's leveraging Mozilla protocol, the design system used by 09:50.460 --> 09:58.220 Mozilla.org and some other Mozilla assets, and it allows developers to then hand over 09:58.780 --> 10:03.740 content creation to non-developers, because it's built on Vactel CMS, 10:05.260 --> 10:09.900 which basically allows anyone who can use Facebook to also add it to these websites. 10:13.100 --> 10:17.100 All right, that's it for me. Thank you so much. Thank you very much. 10:23.820 --> 10:26.060 Do we have questions, etc. 10:28.060 --> 10:36.620 Will the speed almost like crowd in? Do I see like a crowd in copy coming up from Mozilla? 10:37.420 --> 10:42.060 No, the language portal is not going to be like a crowd in, but we already have a similar 10:42.060 --> 10:47.500 tool like crowd in, called pontoon. It's available on pontoon.mozilla.org. That's where 10:47.500 --> 10:51.820 translations come in. That's where people see a set of projects that will localize, they can 10:51.820 --> 10:58.620 enjoy their teams, they see the string, they can actually translate strings. So that's sort of like 10:58.620 --> 11:06.380 our translation management system. It's specifically targeted at people that are translating 11:06.460 --> 11:13.100 Mozilla software. Mozilla language portal contains lots of or most of the translations or all 11:13.100 --> 11:17.420 of the translations that are submitted through pontoon, but makes them available in a more 11:18.780 --> 11:25.100 let's say user-friendly fashion. Then if you go directly into pontoon and try to find 11:25.100 --> 11:29.420 where to download some of these files that you can't really download them in the right form, 11:29.900 --> 11:36.700 you can't really search through them through pontoon. So it's basically repackaging that content 11:36.700 --> 11:42.300 and making it available separately to everybody, not just to Mozilla localizers. 11:44.380 --> 11:45.340 More questions? 11:45.340 --> 12:00.540 Thank you. Sorry, I, I, I, I, it's very simple question. Did you say that the main 12:00.540 --> 12:06.140 audience of this part is going to be developers or who else? I didn't catch it. 12:06.140 --> 12:11.340 The main audience of Mozilla language portal are going to be localization professionals, 12:11.420 --> 12:18.300 translators, from any project, from any, um, software projects they localize. 12:18.300 --> 12:23.340 So they, they lean into this to localizers, open office localizers, Mozilla localizers, 12:23.340 --> 12:28.940 Microsoft localizers, uh, translators that may be not even translators, um, 12:28.940 --> 12:34.620 software. Maybe they would find some of the terms, uh, that we translate useful in translation 12:34.620 --> 12:42.060 of whatever they translate. So it's targeting translators. Uh, the part where I might have confused 12:42.060 --> 12:49.260 you, uh, was the documentation, the best practices. So if you want to, if you have a project 12:49.260 --> 12:54.540 that you would like to get localized, then yes, we're basically also targeting developers with 12:54.540 --> 13:00.300 the documentation of processes, best practices, how to write the code such that it's 13:00.380 --> 13:05.180 localizable. We have these documents available today, but they're a little bit harder to find, 13:05.900 --> 13:12.540 and we think that if we put them on this, uh, hopefully more visible, uh, place, uh, more people are 13:12.540 --> 13:23.420 actually going to benefit from them. Do I have more questions? I have a question. Um, 13:23.500 --> 13:31.260 today, if I want to translate something, I can use a RIGI to do the translation, um, 13:31.260 --> 13:37.500 how, what's the value proposition that is different than me just using a RIGI to do my 13:37.500 --> 13:43.660 translation, because I'm just too busy to do the work, right? Um, it's the quality basically. 13:44.860 --> 13:51.660 Um, at least, at least today, it's very, very, okay. The quality is, and a very fine answer. 13:51.740 --> 14:03.180 Okay. At least for me. No, for me something, some, I, I contribute a bit in the past. 14:03.180 --> 14:07.900 For me, the most important thing is that we're not translating, we're localizing. So some 14:07.900 --> 14:13.820 of the AI will translate, word by word, or the sentence, we're in the culture that's why you have 14:13.820 --> 14:17.980 different versions like different locales, and we don't have just only languages, because we 14:18.060 --> 14:21.900 look alike, something to your local culture, expedience, or things like this. 14:36.380 --> 14:43.980 Yeah, so I have a similar question. Um, there's this trend of using machine translations 14:43.980 --> 14:50.860 for everything in software these days, and obviously the data set that Mozilla is intending to publish 14:50.860 --> 14:57.740 through the language portal will be of much higher quality, but how can you, let's say, um, 15:00.140 --> 15:05.420 motivate users to use the language portal instead of machine translations. 15:06.780 --> 15:13.740 Right. Um, that's a fair question. We're, um, we're already, um, giving the ability to localizers 15:13.820 --> 15:19.340 to opt in, um, to what we call a pre-translation, which means that when new strings are made 15:19.340 --> 15:25.180 available, they already see them as translated. They're marked as still meaning work, but they 15:25.180 --> 15:30.460 actually end up in the product, if we don't act, if they don't act upon them in real time. And the 15:30.460 --> 15:36.460 way that works is it's basically using machine translation that's trained with our existing data, 15:36.460 --> 15:40.860 which is going to be available on Mozilla language portal. Some localizers don't like that. 15:41.500 --> 15:47.340 They say, and we trust them, that it doesn't work for the local. There's 369 locales. 15:48.860 --> 15:53.500 I doubt that the quality is the same for all the languages. I know that for mine, for Slovenian, 15:53.500 --> 15:59.500 which is spoken by two million people, it's unbelievably good. I know that there's, there's not 15:59.500 --> 16:06.700 so much material available in my language probably, um, but, um, but it's way better, uh, than, 16:07.020 --> 16:13.020 than going to generic machine translation engine. So, uh, I think these questions, the answers 16:13.020 --> 16:18.860 to the question that Ludo asked might have a different answer in a different point in time in the future. 16:22.300 --> 16:29.820 There's a question in the chat. Is there any LLM use case involved in any part of the work flow? 16:29.820 --> 16:37.260 Also, where can I learn more about this? I'm, I'm not sure I got the question. Learn more 16:37.260 --> 16:47.100 of more about the language portal. The, uh, in LLM use case involved in any parts of the work flow? 16:47.900 --> 16:57.660 Okay. And where can this person learn more on this? Okay. So, I think one part is what I explained 16:57.740 --> 17:04.540 in my answer to the previous question, uh, but maybe the user is hinting at using that, 17:04.540 --> 17:10.620 using these translations also to train these engines, but the answer to that is that I think, 17:10.620 --> 17:17.660 even though the numbers look high, uh, and I know AI experts, but I think in terms of training material, 17:17.660 --> 17:25.180 these are really, really low numbers. Like, is there any AI involved? No, he says, in building that, 17:25.500 --> 17:40.540 no. I'm not sure I answered that question. Was that? Yeah, he's hiding. Other more questions? 17:46.140 --> 17:50.780 Thank you very much for attending the material room. We're going to clean up and close. 17:50.780 --> 17:57.100 Stickers are only available downstairs now. Okay, thank you. No more upstairs. Thank you very much.