Flock 2026 How Fedora Translators Community Raised The Level

Hello. [laughter] That's a large audience. Uh okay. Uh I don't have to be anxious about uh yeah uh the content and everything. Uh hello, I'm Ben. I do the talking uh in uh Weblite. Uh I care about the people, about the community uh and commercial customers, which includes uh Federa and Red Hat. Uh okay, this will be quick because uh it's not that many people uh here. [snorts] Uh let's go through uh a few questions. Uh who apparently all of you are interested in translations, right? Uh that's why you're here. Okay. Uh and who of you translate who contributes with translations or raise your Okay. >> [snorts] >> uh and who of you contributes to uh the translation tools by code or uh local internationalize the uh the project um and who's um is and uh does any of you has English as a first language? Were you born with in English-sp speakaking country? Okay, that's why you're here apparently. Um, and who knows what Zanata was or still is? >> Okay. Okay. >> I was I was not sure if what to what to say if it's is or was. Come in. >> Got it. You're the best. [laughter] >> Uh, a voice from the other side. Um, okay. Uh, and did anybody of you work on Zanata? Okay, great. That's that's great to know. Maybe we'll have some uh some insights. Um, so and who of from all of you works for Red Hat or worked from Red for Red Hat? Okay, great to know. Uh so uh that's uh curiosity of why do we translate and why do we localize uh what are the differences and everything. Um and yeah why do we localize? We localize to make the software more accessible to make it yours even if you can speak other languages. uh for other people. We localize uh to make it even possible to use uh the software uh and to keep languages alive. Uh for who is important question as well uh because well we usually don't do the work. We don't do contribution uh to open source software just for ourselves. We see something more uh there. So we can uh localize the software for kids. We can localize federa uh tools for our relatives that don't have the knowledge of uh other languages to welcome new people in um emerging uh economies uh or in countries uh that are uh now adopting more and more technologies to the world of open source which is welcoming and diverse. Um and uh we can also do it for us uh because we want to feel home when we are using our own tech. Uh as um most of you have friends here, you probably didn't experience Yeah, I expected a little bit larger audience. So I could like find the one person that's here alone from all over the world from all all over the world being the one person that speaks their language here and the whole day they are unable to speak their native language uh because they're here everyone speaking speaking English and everybody's a little bit like well yeah uh you can navigate but you don't feel at home you you don't feel that warm so we uh localize uh because of that to feel comfortable while using our tech. Then we have Zanata. I will be happy for your inputs. Uh from uh the feeling what Zanata was uh there was not enough uh it was not enough uh to deliver quality translations. It was not enough to make it comfortable and not complicated process for everyone involved. be it developers, be it translators, uh there were other tools and uh scripts to synchronize everything. It was too complicated and it was not uh being developed and maintained enough for some time. Yeah, there is an input. Jean >> Baptist Zanetta was a good tool to translate, but it was a terrible tool to collaborate as a team. It was very difficult to know who does what and they had software engineers. The UI was okay but there were no collaboration features. Yeah, thank you very much. Uh so yeah, it was not enough for everyone involved. That shouldn't be the case. And it was also not a core business of Red Hat which was proved by uh a layoffs by uh some layoffs that were like okay part of the people that were working on this we no longer need them. Uh and Red had this big company saying we no longer take this as a priority and then the community of local uh localization contributors where uh get to a point like okay what should we do? So it was not the end of localization of Fedora because yeah uh it still um is localized and I would say uh in better way and that's because of a migration that happened uh thanks to Jean Baptist and other community members in 2019 uh from uh Zanata to weblate um it was a demanding task. It was I would say performed well uh because the community was well informed uh because the community members like Jean Baptist Rafael uh and others uh did the work to inform everyone to migrate everything to set up everything at the new place uh and uh uh that's the large part and it was also supported by Red Hat financially eventually which is good. Um, okay. What's what's Weblade? Weblade is a localization platform. Uh, that's what we do. But we doesn't have to be the core team of Weblade. Uh, it can be anyone uh all around the world. Uh, Weblite is a Libra software from its core from the start. There are no differences uh between how you use Weblite. You have the same set of features. You have the same code base. uh and thanks to being uh web light being developed and supported and liked I would say uh by the community all around the world uh it's used by the communities uh of Federa open soc libra office many other large or uh uh a little bit smaller projects uh and uh yeah that's about what you need to know at the moment about web light if you don't use it already it's a nice collaborative tool that you don't have to care about you don't have to uh look after it. You don't have to install it. You open a website as a translator. It works. As a developer, you connect it to your git and that's it. You don't have to do the work. Weblade does the work for you and you do just your part of the work. Developer does the development uh looks after the code, maintain the repository, devel um and the uh translators do their part of work. Uh okay. So uh what changed uh I promised I will uh I would uh share some numbers. I will leave that to uh Jean Baptist later uh because he has way nicer grabs u and I didn't want to steal his uh uh his work or um his focus. Um but um yeah there are 1,822 users on translators uh on translate federa which is quite a large community. Not everyone is uh the same level at the same level of activity. Uh somebody's uh more uh somebody's contributing more somebody less. uh but yeah it's a reasonable number to uh get uh the translations to 100% but we are not there. Uh yesterday uh a group investigation uh realized that there are 50 million changes to the strings in the last since 2019 since uh Federa translations and all the projects that changed uh uh uh or joined uh on this instance uh started using weblate as the translation platform. It's quite a lot of work. Uh and projects like Anaconda uh Anaconda um core parts of uh Federa uh are translated to like uh 95 used languages. Uh so yeah, can it be better? Yes, of course. The work is definitely uh not finished. Uh and that's what it gets uh what gets us to the failover uh is not enough. If the uh if the software is not translated fully and your user sees um a u English string, it can get lost totally. Not everybody uh well everybody here uh understands English but it's uh not uh done. So it's uh not a work complete. Uh we need more people to uh be interested in translations to contribute to join the efforts uh to see the reason why we do it. Um and uh uh I would say uh yeah I can describe it on this room that uh community of translator translation contributors is uh getting old a little bit uh and we are not in large numbers on the conferences. From the talks I have uh when providing services of uh of weblade support to communities, I can see that the community is diverse that we have young uh uh young people but usually contributing to one or two projects not being um not being uh interested in the whole idea of uh open source uh localization. Uh would AI or machine translation in general be the answer? Well, no. It's not not enough definitely not yet. Uh but it can help. It can be a help to uh translators that would be like uh that shouldn't be scared like okay I have to do a large part of work. It can be uh just uh help that uh I will have less amount of manual work that will I will just polish to nice translations that will be humane that will be for everyone. uh and um yeah that's uh gets us to a few challenges uh that there is not enough focus on this uh that uh many translations that appear uh okay I did my work to the translator if you check the statistics if you check uh failing checks and uh other things uh other helps in weblate you will realize that the work is definitely not complete even though uh that it was in previous versions or so. Uh so it needs focus. It's needed it needs more work. Uh and uh yeah um maybe you can discuss a little bit uh what can be done uh in this matter if we uh for Federa need more support from uh from Red Hat if if we need to shout out more about uh how nice is uh uh are some parts of this work. how um rewarding it feels when you uh can uh give someone of your relatives or someone totally unknown an old laptop with uh unknown system for them and it just works for them. It speaks their language uh because it's prepared to be accessible and everything. Uh okay. Uh that's the uh end of my talk. This is one of the stickers you can get from Mel and me here or a lovely t-shirt. Uh I'm Ben. Uh I do talking uh with Weblade. Uh I usually uh am uh better at it. Uh but uh I'm not in my at my 100% uh uh recently. But what I can promise is that I'm really good in answering questions and providing information you want. So you can ask now, you can ask later. uh we can connect for uh a call and uh yeah if you have anyone who's interested in localizing Fedora or anything else if you have suggestions how weblate could be more important tool uh more uh comfortable tool for everyone uh involved. Yeah. Uh please uh share those suggestions uh our way. So if you have questions, [applause] thank you. I have one question. You mentioned AI is not good enough yet for translation. Uh do you see any uh initiatives or or progress in this area? >> Yeah, maybe. Yeah, I can. uh the progress is definite uh definitely there I would say uh maybe in general um how quality translations are provided I think that's moving uh really fast uh with some of the models uh the issue is that uh so far it's hard to tell any model that this one part is logical block of something um you can provide those uh language uh large language models uh with glossery uh you have maintained uh in weblate uh you can provide uh a lot of other context but uh still it's not it's not perfect uh it's moving forward I would say I don't know how well it will uh be in uh in a few months or years uh at the moment it's great to uh that uh many of the models are imp um uh integrated with weblate. So you can get the translations and then finish them uh as a human. So you provide human translations and not just one by one uh grammarly correct translation that doesn't feel uh real in in the UI. Then >> uh Mikuel has more to say. He >> maybe I will comment a little bit on that. uh we have done quite some work to better integrate with this model. So weblate now sends way by much context to get more accurate translation. So it's improving on both on the model sides and on the blade side. So yeah, you can definitely use it to get okayish looking translations, but it's still it cannot beat the human doing the translations, at least so far, because you know more about the application than the model does. and you know the typical user which all of that can be somehow fed into the model but it's complicated to transfer all your knowledge in into the text input for the model. So yeah it you can improve how it translates by providing it additional context. We have inputs for that in weblite but still it's something that you would like to review because the models sometimes just don't get it. Sometimes they allinate the word because they don't have enough coverage for some languages. So they produce words that look like in that language but they are actually not existing. But this varies per language. For widely used languages like German and f or French, it will probably be better than for some minor language in from Africa. Uh I have experience with Czech language and mostly it good but sometimes you get the words as I describe this is non-existing word and yeah you need to catch that manually. I would uh like the the hope I have in uh implementing uh uh more evolved uh machine translation models is that uh those will be used for uh maybe some would uh say boring part of uh translating string by string. it they those will get better and provide better translations but it doesn't mean that sorry uh that the whole localization work is done because currently we are in a phase that's even if we would get to 100% of translations we are not uh in a perfect uh fa uh perfect level uh because that way we would get uh to localize or translated apps that are many times missing plurals or handling plurals but are uh not using gendered language uh are not using uh are reducing uh because the source language is usually English >> [snorts] >> uh and um for example Czech, French uh and uh many other languages uh are more complex uh you uh you can have denominations you can you are talking formal informal uh and all those nice things that you are currently not having. You have a reduced part how software talks to you. Sometimes with games it gets more evolved because the game uh you have a character there so the game talks to you some way. But if you open a laptop, it's it talks the same to all of us and to all Federa users. There is no difference uh if you use German uh or German uh if it's uh who does uh uh use the laptop or desktop at the moment. Uh so once we have uh the uh once we have the help of really well performant uh machine translation or AI powered models, we could go to uh raise the bar again and start translating uh these uh start making it not just simple translation but actually uh deeper localization uh that will suit more people. Jean Baptist has something to to add to that. >> No, it's a new question. >> Okay, go on. >> I would like the web tool to provide some insight about the health of the community. Each time there's a flock event, they there are some graphics year after year. How many contributors are there? how many downloads and I would like to have this tool to help us to understand in the federal community there is 1,800 translators in the translation platform. How many of them are active? What does it mean to be active? Is it growing over time or is it going down? Which languages are growing? Which ones are going down? And I would love to have something that is allowing us to understand our community so that we can try actions to strengthen it and improve over time. I don't know how know if you have the some numbers but I would love to to have some and to be able to compare with other communities not to compare to be the best but from other communities what do they do to make it work and to grow because ultimately we always want to be more people more contributors. So currently the only thing and Mika will add more uh from a technical perspective but uh currently what is possible to reach on any instance of web is SL stats and you have some basic outlook uh but yeah it's not very clickable. you cannot select uh the part. This dashboard would probably be uh something that will come out of uh uh plans we started based on a UIX research we did uh uh together with our co-workers. Um and uh but when and uh we are not sure but what's sure uh what's possible is that uh the more input you and other people will give about this what what it should provide uh the better we can then provide it uh and develop it uh the so it suits your needs and wishes. Yes, Mel. So, >> yeah, we do collect some of the historical data. So, some kind of that is technically possible, but as Ben already did that point, we need to know what community is interested in because like we don't exactly know what you would like to measure on the community health. So, writing it down is a good starting point. Basically uh so far uh weblate statistics is that the mal created first version of weblate in 2012 then it evolved uh and more and more people wanted some statistics and everybody wanted a different statistic to to look for. So every time we added a different statistics that's counted. So we have a lot of numbers uh and you can download it as raw data. you can visualize it somehow. Uh but not everything is in the reports. We uh weblate collects even more data. Uh that's possible to work with but how to interpret it uh if to export it into uh some graphana or some other uh tools or uh if we should build something uh tightly integrated. That's the that's the discussion and it will be great if we can define it together. >> Just let me add that the next talk might provide more insight into metrics like in five minutes. [laughter] >> Okay. So we are on time which is great. Thank you for being here and yeah uh keep doing the work. Keep uh uh making uh localizing uh software interesting and enjoy other talks. >> [applause]