 Hi, welcome to the presentation about section translation. My name is Amira Aroni. I'm the senior strategist in the language team of the Media Foundation. And this presentation will also be shared by me with Paul Genere, who is the designer and the product manager, and by Nick Goudas, who is one of our software developers. So we'll talk about section translation, which is a new feature, but we first need to understand why I'm doing this at all and what is the history of this feature. So language is a very simple thing in the Wikimedia movement. We have a lot of languages. We are one of the most multilingual sites online. And language is in a way, knowledge itself. That's how we communicate to each other in a lot of languages, but it can also be a gatekeeper. It can be a blocker because if knowledge is encoded in a language you don't speak and it's not translated to your language, it's not available in your language, then you don't have it. So just as a very simple example, which should be familiar to a lot of you, the English Wikipedia is the largest. It has way over six million articles. The second biggest is German. German has more than two million articles, even though much fewer people speed the German language than, for example, Indonesian. Indonesian is spoken by more people than German, but it has just over half a million articles. Bengali, which is also world's top 10 language, has even fewer than 100,000 articles. So that's a bit of a problem, which we are trying to address. And our way to address this since 2015 is a feature that many of you may have heard of, and that's content translation. Content translation is an extension for Wikipedia, which allows people to create articles by translating them from another language. So this is how content translation currently looks like on desktop computers. You see an article in the source language on one side of the screen, and then you see initially empty space for writing a new article in your language by translating this article from another language. What content translation gives you is you see both languages at once. It's convenient for translators to see and compare each paragraph. It also helps you adapt images and links and templates and categories and several things. So it's fairly successful. We have more than 900,000 articles created in all the languages. And we even think that it's quite high quality. I'm trying to be modest here. As a very simple measurement of quality, we can see that around 3% of articles created by content translations are deleted, which is compared to an average of 12% of new articles deleted if they are not created using content translation. So for a tool that creates articles, how many articles are deleted is a simple measure of quality. By this measurement, we think that we're pretty good. However, we also have some limitations. Content translation is far from perfect. So a big issue in content translation is that it only allows you currently to create complete new articles. However, it's only for the first revision of the article. Content translation currently doesn't help you very much in expanding or improving or correcting the article. And you can see that the length and the quality of articles is quite different from one language to another. This is for example, the article about the T-Rex in Hindi. Hindi is one of the biggest most spoken languages of the world. And you can see that it only has two paragraphs of text and an info box and four references. That's not much. Because info boxes, it's nice, but the English Wikipedia article, it has much, much more. The Russian Wikipedia is also much longer. So this is not enough. It would be nice to have a longer, more complete, better reference, better illustrated article in Hindi. And another big issue is that content translation is very much built for desktop and laptop computers, not so much for mobile phones. And we used to say that the future is mobile, but it's not the future. It's already the present. Already now we know that in all the languages, most of the reading is done on mobile phones. More in some languages, less in some others, but in pretty much all the languages, all the reading in Wikipedia is done on mobile phones. And that's why we need something new to support mobile phones better. And this is where I transfer to speak about the present and the future to the designer, Paul Giner. Thanks, Henry. And yes, Henry was mentioning some of the current limitations of content translation inspired the next phase of the project, which we call section translation, which is not as much a new tool, but more of an expansion of the capabilities of content translation. If we think of the initial scope of content translation being focused on desktop and being used for creating articles, with section translation, we want to expand those capabilities to support mobile for both creating and expanding articles, but also support that expansion of articles on desktop. Or initial scope. Here, the initial scope will be on expanding articles on mobile. And now we are also covering creating articles on mobile and that's what can be tested right now with the current implementation. But before we go there, I wanted to cover a bit on the journey of what we have done so far because the tool started in the process of being designed and researched before we started in demanding it. So there was a process of research that was led by Eli, or researcher working with the language team that started to validate that the concept of translating sections on mobile was a successful and valuable concept, making proof of concept and evaluating all that work, getting feedback with the initial implementations and finally releasing into a real Wikipedia, in this case, when value Wikipedia, an early version that we did also usability testing to learn from. Some of the details of the learning from the early concept testing was about validating the core assumptions and we realized that editors were perceiving valuable, the possibility of expanding articles by adding objections and doing that on mobile was perceived as an easy and fast activity. And we also identify areas to keep researching and learning from your agency on the right side. Some quotes about some of the participants, how they validated some of our assumptions about users that are mobile only that were previously left out of these processes. Additional research was based on the proof of concept and early feedback on the initial technical implementation and we were also getting both positive feedback and ideas to improve the tool. And more recently, after having an initial version available in the Wikipedia, we were doing usability testing on it to validate that those ideas that were testing positive were, in fact, received also positively when applied to real articles and real data. Obviously, as an early implementation, we identified aspects to improve like supporting search, better placing the sections that were published in the right place in the article and some other aspects that we have been improving since and most of them are already available in the new version. And talking about the future, we are also thinking of the next steps for section translation. An immediate one is to enable the tool on more Wikipedia's. So if this seems an interesting tool for your Wiki because there's a lot of mobile editing or mobile activity on the Wiki, you can give it a try at test.m.wikipedia.org and special content translation. It's the mobile version of content translation is where we are surfacing this new version of section translation. And yeah, if that's interesting, you can contact the team and we will consider your Wiki as one of the next ones. And we are also exploring the possibility of enabling more entry parts so that people can find the tool. Obviously remembering the previous URL is not the best way to find the tool and currently there are not many more but we will start as we improve the tool making it easier to find. And we will be expanding existing possibilities to translate like better suggestions or translating an article when you identify that it's missing on a specific language to more experimental ones where we can surface some of these missing sections and let people create them when they are missing in stop articles, for example. We're also exploring further the matching translation field. And we want to learn more about how users edit matching translation and which is the perception of quality and the need for improving it in order to support really good quality, the creation of really good quality translations based on how editors improve the initial matching translation. And we also experimented in a few languages to integrate an open source newer matching translation system to provide matching translation to underserved languages that didn't have matching translation before. This is experimental. It's not a high quality translation service right now. Because it depends on open data available, but it does an interesting thing which is closing the loop with content translation where the more translations you produce the more content is shared in this open corpus of knowledge and it helps to train better the system. And yeah, finally we have time for a demo if you were curious to see how section translation works in practice for that, Nick can show you. Thank you Pao. So now it's time for a little demo and together we'll try to translate a section and publish it to a target article. So let me share my screen and here as you can see this is the dashboard of our application this is the section translation application and here are some suggestions for the users to either create new articles or expand already existing articles by translating sections and all these suggestions are perfectly valid but I'm going to choose an article of my preference and I have to click at this button but before that I'm about to change the target article here is the language selector where user can update the source or the target language and I'm planning to select Bengali language as the target language because this is the only wiki that section translation has been deployed as of now and I'm continuing to search for my preferred article and this is the sun article so I'm clicking on it and this is a screen where the user can confirm that they are going to translate an article from this section, from this specific article and here are some information about the missing sections and the availability of this article in Bengali language. I'm going on to click on this button and this is the screen where users can select the section that they are going to translate and publish and I'm going to go with the first one course and it's the general characteristics and this is a screen where user can compare the contents between the section in English and the full article in Bengali language and as you can see here are the contents of this section and on the other hand we have the whole article in Bengali language somewhere at the bottom of this page we can check just above the first appendix title appendix section sorry this is where our new section will be published and this placeholder is here to indicate it so I'm going to go to the translate button and this is a quick tutorial users can see and with some quick information about how it works I'm just going to skip it for now I'm quite familiar with it by now so here is the actual screen where users can translate sentence by sentence the whole section and starting from the section title here is this card is where the suggestion translation is provided and we can even select a different machine translation provider like Google Yandex or even start with an empty sentence for now Google it's fine for me and I'm going to go just by apply the proposed translation for the title as you can see the untranslated sentences are highlighted with a yellow color while already translated segments are highlighted with a light blue color and here are some statistics about our translation since I didn't do any change no edits made by me so I'm going on here and try to edit this translation but I'm not really familiar with the background language so I'm just going to remove some of the content and again I can check that 47% edited by me at this point I'm going on to apply the suggested translation for a few more sentences and once I feel that I am ready to publish this section and the translation is complete I'm just clicking on the done button so this is the last screen where user actually confirms the publication and has the chance to review for the last time the contents of the section and I'm going to publish the section as it is for now and you can see that Google cannot do that because the translation contains too much unmodified text and this is something that is expected to me because machine translation is never perfect and the user should always modify it a little bit so I'm going to modify the section this is the last opportunity and after doing this I can continue with publishing it again and this time I expect it to be successful let's see there is... yeah it was successful and now we are redirecting to the target article where the new section is going to be published this is the target article here is the new section highlighted with the yellow color to indicate that this is a new and also this indicator and at the bottom of the screen here are the contents that I have translated and at the bottom of the screen we can see these two banners that indicate that the new section was added and also here is an invitation for the user to go on and translate more sections for this article I'm going to click it and we are redirecting back to section translation application and specifically to the sun article and this is again the confirm the translation screen and that would close the loop so I could go on and translate another section here so this is it I hope you enjoyed back to you Pao yeah thanks Nick if anyone is interested in learning more about section translation there is... there is a weak-medium page where you can learn about the project track the progress and basically learn more about what's going on with it and this is just one of the different projects that the language team has been working on so if you are also interested in learning more about the work of the team which part of it is in the picture but we have even some more folks joining since the picture was taken feel free to contact us at Media Weekly with Media Language Engineering and yeah thanks for your attention and we are looking forward always for feedback and questions