Multilingual content delivery at scale

Andrew Hickson 00:06 Igor, I see you've joined us her user. Great. Yeah. Awesome. For those of you who are attending the conference, again, like we've seen all the way through the day, if you have a question, please post it in the question and answers for Igor. And we will address it after his presentation. Igor, I'm going to jump out of your way, and I'll let you kick into it. Good luck. Igor Afanasyev 00:32 Thank you so much, Andrew, let me share my screen. All right. So yeah, I'm Igor working in SmartCAT is Senior Product Manager. And today we'll be talking about multilingual content delivery at scale. And before we start an announcement, the best question today, after my presentation, we'll get a book about global marketing. This is a pretty new book. And it covers lots of different aspects of creating, like, creating multilingual websites, which kind of complements the topic that we'll be discussing today. Now, why are we picking that topic? This time around? Well, we see certain trends. And we have clients that are coming to us and ask us those questions on how to implement proper, scalable process for delivering multilingual websites, blogs, etc. And they are coming from different directions. One of the recent trends that we saw is that we have lots of garment clients that are coming to us and because of like college related content, because of having a need, sometimes a legal need to deliver multilingual websites, as like, for emergency response programs, etc. They understand that their current processes are outdated, and they cannot cope with the pace at which they are creating the content. Sometimes they have to update a certain page on the website multiple times a day with some new updated information, they want to deliver that in multiple languages. We also see that software companies, they can like crack the idea of continuous localization for their products, they do fast product delivery cycles, and they enjoy using continuous localization for that, that the same time they see that their marketing content is a little bit like not on par with the same speed and ease of continuous localization that they have for software. So they are asking us if there's anything can be done to do the same kind of thing for a marking content for technical recommendation for Health Center articles, etc. And the third thing is that machine translation is getting better, not every year, but every quarter and every month, to the extent that clients are usually happy to use machine translation. And they want to pre translate everything with machine translation published their website, or specifically coupon talking about technical documentation helps into articles, they want to pre translate them using machine translation. And they buy into the idea that machine translation can give them this near real time content delivery experience, but at the same time, they're technologists and the processes they use, do not allow them to tap into that into its full potential. So how can we solve the problem of really scalable multilingual content delivery, there are two key things that I'll be demonstrating and discussing today. The first is continuous localization, a, an approach to content delivery that allows you to create the next level of automation. And then we'll be talking about more efficient machine translation and machine translation, post editing scenarios. There are more things to think about when we're talking about automating and scaling your operations. And if we have time, at the end, I will be covering three items, last items on that list as well. The first of all, let's talk about content delivery. Ultimately, this is the main like the main area where you can apply your efforts to implement in much better much scalable content delivery in multiple languages. And if we're talking about continuous localization, and that's something that I was talking about in the past in my previous presentations, the conferences. The problem with the definition of continuous localization is that that definition doesn't really exist in many companies, both from the buyer side and from the vendor side, when they're talking about the opposition, they can be talking about completely different things. That's a problem. Because if you are shopping for a translation management solution, or you're talking to different vendors, and they say, Yes, we are offering you continuous localization, because it's a common buzzword these days, you don't really know how to compare those things, how to understand that what they offer will be actually truly scalable. So I googled around and found a few kind of definitions or something that resembles the definition of continuous localization. And none of them are really great. Some of them are not real definitions at all. Like continuously position is a process that relies on automation to accelerate the sequence of steps involved in producing professional translation. It just says that continuous localization, kind of like a better automation, but it doesn't give you a definition of how to distinguish continuous localization from any sort of other automation, if you will. Some say that continues to position is a best practice for integrating translation into software development. Again, it focuses on software and just says that conditional position is something that is good. But what is continuous localization in the first place? The the third one is, is a definition. Kate Vostokova 06:15 very, I'm very sorry for interrupting but few people complaining that the volume is very low. Actually, I can hear you perfectly, but we have like several complaints in the chat already. So if you can somehow I don't know. I increase the volume. Yes, thank you so much. Igor Afanasyev 06:34 Isn't better now? Say something a little chat says yeah, chat says yes. Right. Okay, awesome. Awesome. I'm sorry, I'm having couple of microphones, I guess I was not using the right one. So the third one is, says continuous localization is a localization of software that is developed using Agile methodology. It looks like a definition is pretty precise and explains what continuous localization is. The problem is that it's really narrow minded. It makes you think that continuous realization is only about localizing software, and it prevents you from thinking out of the box. So I would like to propose my own definition, the way I believe in that the way we are thinking about it on in smart cat and the way we're trying to do processes around like when implementing things related to continuous localization. Continuous localization is a process where a source content is autonomously gathered and published for translation in its entirety. And the word translated content is autonomously published or integrated into the product without artificial delays. So I put emphasis in a couple of places here, because they can be a real good litmus test when you're thinking about continuous localization and trying to compare different automation methods and approaches and technologists. Because it will, if you can see that a continuous localization process as it is, like presented to you by like a vendor, technology vendor, etc. Does those sorts of things that are emphasized here, then you will have a process that will be really scalable, and I will try to explain this in further slides. Now, here's the sad truth about the connectors and the current state of technologies. When we're trying to make in really scalable process for creating a multilingual website, we're thinking about connector, and connector is a sort of like magic thing that can solve that scalable problem for you. It makes everything simple and smooth. And if you have a connector, you're good to go. Now, the problem is that most connectors offered by modern TMS cannot be used for continuous localization. And here's why. Now, let's look at the process. Like the typical process that you have with having a TMS and having connected to your favorite content management system. So you have a content management system. On the left, you have your translation management system on the right. In the Content Management System, you have some pages that you want to send over for translation. So what you do with the connector, you basically select those pages, and then you create a translation job, where you specify here is my list of source language and in my target languages, here's the deadline. And then you're sending that over to a TMS, just a few clicks, you know, looks pretty simple, then you wait for the translation job to be fully complete, and then you import the translations back. Now the problem with that approach is basically mimics what you used to do without connectors in the first place. It resembles the manual approach that we all know No and love or hate depending on the volumes that you're translating. So, without connector, what you would be doing is like copy pasting the content into like a Word document, sending it over by email to a vendor, then get the translations from them back, then copy, paste them into your CMS. That's the manual process. Connectors automate that to a certain extent they reduce human error, you don't need to copy paste things, you just like select things in a nice UI, and then send them over for translation. But it's still largely a somewhat automated manual process that uses the same job based approaches. And why. Why it's not scalable? Well, it works well enough, if you want to send some new content for translation. I don't know once, like every week or twice a week, but what if you want to do this every day or every hour, the problem with that you will have lots of things to do in order for this to happen. And on the TMS side, you will have lots of small fragment and jobs to handle. And those small fragment jobs are really hard to deal with on both sides. As a localization manager, if you have too many small jobs to handle, again, it becomes like logistic nightmare to handle those. If you're a linguist, and you get some small translations to complete translation jobs to complete again, you do not understand what you're actually doing. It translation job can be one document, a couple of documents can be even just a few strings that needs to be translated. So how do you deal with that? And how do you understand what you're translating. Now, if you're trying, again, if you're trying to speed it up, you will see that you have this scalability ceiling. And you cannot simply like do this fully automatically. If you want to really automate the process and not have to click any buttons to create those jobs, you want to automate that part as well. You still will be waiting for some new content to be accumulated on the CMS side so that the like New Content drop or a new job will have substantial size, if you will, that will be manageable. During our last vlog from home, back in May, we had Rebecca ray from CSC research, and she presented in your paper that they that they like assembled. That is called continuous localization work speed. So it really highlights the kind of the state of the art and state of the mind of the industry, when it comes to all things related to automation and continuous localization. And one of the things was she had an interactive poll, where all the attendees tried to answer the question, what was their number one challenge related to continuous localization. So besides just not knowing where to start, which was the first like most popular answer, the second one was balancing continuous localization. So the speed and quality produces. And then in the paper itself, Rebecca Ray says it well that several interviewees expressed frustration or reservations about how continuous localization effects linguistic quality, rapid and frequent drops of content with zero contextual information to guide linguist is typical under the modal. So, it looks exactly like what we saw here in the previous slide, right. So, as you try to implement this continuous localization in a wrong way, based on the job based approach, you will have this fragmentation, you will have much less context as if you were translating a like the entirety of the content the entire website at once. So, the quality will really suffer. And you can really scale that. So you can do the do the new content drops every 15 minutes again, you will have more problems with that. And trying to speed up things to give more lead time to linguist will actually lead to less quality, not more quality and more time to translate. So how can we deal with that? How can we make sure that we can have a proper scalable approach and process to localize websites without having those quality issues. And this is how a continuous localization process looks like. Again, going back to the definition of continuous localization that I gave before. One of the key things is that you want to synchronize all of the content that you have in your CMS for translation and make it available in New York TMS. Available for translation at all times. So here you have your CMS, you have your set of pages, or objects that you use to translate. And you set up a connector that works as a two way synchronization between your CMS and TMS. On a TMS side, instead of having multiple fragmented jobs, you have a single continuous translation project that exposes all of the strings for translation. And that connector that does this two way synchronization is responsible for synchronizing the content between CMS and TMS and publishing all the strings for translation. Now, other any connectors that use this exact approach, well, that's something that we are building a smart bet. And right now we have a few connectors that use exactly the same approach. And we encourage you to try them and see how that approach would work for you. In a few of them, like like Google Docs, or WordPress, Contentful, are really easy to integrate, to set up and test in your environment. What I want to show you here is how that looks like in the actual world, like how does the connector work? For example, with Google Docs, here, what we see is, for example, you have a Google Drive folder that you are setting up an integration with and in that folder, you have two documents. So you want to localize them using this continuous localization approach, you do not press any buttons to make them appear on the TMS side, you set up the integration once it will monitor this folder and all the sub folders and will expose all the documents for translation in the TMS. In the TMS, you see all the documents available for translation at all times, you do not change the project, the project stays there forever. And usually what needs to be it needs to be like, just leave it up to 100% at all times. But if you have something something new appearing like new document or in your string, it will appear in that project as is fully translated fully untranslated or like semi translated depends on translation memory matches. So if we open a certain document in the source in English, it will look like that. And if you open the same document on the TMS side, you will see that it has all the content, all the headings and paragraphs etc, available for translation. Here I'm providing the translation, like machine translated content for this example, so once the document is translated, the same process that synchronizes your CMS and TMS pushes changes back into your, into your content storage here, it's a Google Drive folder. And for all of the source documents, it creates a folder subfolder with the same name underscore translations and contains all the translated documents there. And if we go into a specific folder for one of the files, we'll see that it generated a Russian version of that document with full, full formatting. Now, the beauty of this approach is that this process works behind the scenes all the time. And it doesn't matter how fast or how slow it works like you can run it every day, you can run it every hour, you can run it every minute, it will still lead to the same kind of view of your docket of your documents and your content on the TMS site, you don't have this fragmentation and you have full context of what you're translating. For example, if you want to change a word over here, in a content creator goes into the document that just replaces that word with something else the next time the connector detects this change, that change will appear on the TMS side and here you will have a new segment that will need to be translated. If you are changing the type or adding some new content and new paragraph over here, again what will be available for you in a TMS you will have a fully translated document that you already translated before. And between those fully translated segments, you will have a new segment that needs to be translated. So as a linguist goes, and provides translation for all of the new stuff, they see all the past translations, they see the page in its entirety, they get all the context and they understand what they're doing. Now the same process monitors the translations, and as the translations are being edited initially, or updated afterwards, they all are synchronized into the into the final document. Again, you don't have to click any buttons integration works behind the scenes. You can work as fast as you need to. And the faster it works. Because the more lead time gives to your linguists, so you're not compromising the quality, the quality of life and like the quality of job of The Linguist because they see all the content at once. But you're giving them much more lead time to actually provide quality translations to, let's summarize all the things that we discussed about traditional job based approach to writing connectors, and the ones that use this, like true continuous localization mode, the traditional job based ones are manual by design, they work on the same principles that you use to manually submit documents for translation. And continuous localization ones are built with automation in mind. And because of that traditional job based approach doesn't scale much. It allows you to simplify certain operations, it reduces some clicks, it reduces some human error. But at the end, there's a limit on how fast you can do those, recreate those jobs because of fragmentation. The continuous localization approach really scales easily because no matter how often you run that automation, good don't have that fragmentation in the first place. So fragmentation leads to small jobs, too little context on the traditional set of things. And that leads to degrading quality. And with continuous localization, as I mentioned, linguist always work in the context of full documents, so they have no quality issues there. And lastly, the traditional job based approach is really hard to combine with on demand, post editing, and machine translation, post editing. While the continuous localization approach is really embraces this, and we'll be talking about this specific item in the next chapter. So how do you implement machine translation and efficient post editing and selective post editing? Using this continuous localization approach, there are different first of all, as I mentioned, probably before, machine translation is quite good enough these days, so that clients want to do like blanket machine translation and publish the machine trust leads with content online for certain types of content, not probably for highly visible marking, marking websites, but for technical documentation, developer documentation, health and learning articles, etc. And when it comes to traditional connectors, and traditional technologists, traditional translation management systems, it is relatively easy to implement this like blanket, like 100% ROI on machine translation process, or it is easy to implement 100% Machine Translation post editing approach, because this is how you set up your, your workflows, you just say that everything goes through machine translation, you send anything into TMS that gets like automatically translated using machine translation, you import things back, and you're done. But when it comes to scaling operations, or you mentioned that you want to have a hybrid model, you have a large website with lots of content to translate, and you want to do this machine translation. But for certain types of content, you still want to have this like post editing capability. And it turns out that the traditional job based approach are not that good in providing that sort of flexibility. Because what happens is that when you are done translating the job using machine translation, that translation job is closed, it's archived, you cannot amend that you already have like paid for that it's a done deal. Now what happens if you want to send just a specific page or a specific type of paragraph or sentence for revising, you cannot really easily do that unless you specify a like full page for translation, and then make it make a translation job with different parameters so that it goes through a different tech process. And again, if you want to do this at scale, it requires lots of manual operations from the localization manager. So it doesn't doesn't really scale that much. When you're thinking about machine translation process, you need to understand how the overall process will look like with machine translation. And specifically, as I'm talking about selective post editing, would it be possible to fix a particular translation what steps you will need to do in order to do that? We'll will that fix that you will be applying for a particular translation used in future translations. And if the selective postage is available, at all with the current type of connector and the platform that you're using, since the selective post editing is really hard to implement, in this like job based approach, as I already discussed what people what we see the clients are usually doing in their in their processes that once the translation is, machine translation is done, but they see that they need to post edit a certain string, what they usually do is they go into their CMS and edit the string there. And that becomes a much easier thing for them, instead of doing like this whole like round trip with another translation job and like sending over some instructions to the linguist that they need to pay attention to a particular sentence or whatever, they just do the post editing, right in the CMS, this feels like a natural thing to do. Really easy, really, really fast. But the problem is that this fix if this, if it's done in the CMS site, it's not applied to translation memory on the TMS site. Now the next time they want to translate a similar page or similar sentence in the context of another page, the machine translated process, like blanket machine translation will do the same kind of error for that string. And they will need to reapply the fix in the CMS over and over again, as they will be dealing with lots of content, again, not something that is really scalable. If you're talking about lots of lots of content. What's What does the what does the continuous localization process bring here in terms of improving this selective post editing capabilities? Well, previously, and I'll probably go to one of those slides. We see that with a proper continuous localization approach where you see all of the content available for you for translation at all times, you can do this blanket machine translation for all of the content that you get in the TMS and that content will be automatically synchronized and published on your website by default. But since you have everything available for translation, you can see the current state of translation for every single sentence, what you can get is, if you want to fix a typo, if you want to change some terms to something else, if you want to change the punctuation, whatever, you can just simply go to that project and your TMS find a specific sentence in the file and change that. And that's it. And that change will be automatically propagated into the final document. So you don't again, have to create translation jobs or do any additional clicks to make that relation happen. So with continuous localization implemented that way, it is possible for you to go back to any previously translated content, and amend that and improve that over time. And the important thing here is that you're not simply improving that content for that specific page, and you're synchronizing that into your CMS. But you are also providing all those amendments, all those changes into your translation memory. So the next time you see a similar sentence appearing elsewhere, it will be using a revised translation that you provided before. Now, Andrew Hickson 28:27 sorry, Figuarts. Just to jump in very quickly, we've technically reached the end of your presentation time, I have a number of questions open, but if you want, I can give me a couple of minutes just to wrap up. And then we'll jump into the to the questions because there's quite a few coming in. Igor Afanasyev 28:43 Okay, okay. Sounds good. So I'll be quick here. So this is probably my last slide anyway. So if you want to really implement this selective post editing at scale, there are certain ways to do that. And, for example, you can have a full machine translation of the entire content, but then selectively translate most important content and how do you define that most important content? There are a few ways to do that. First of all, you can implement some sort of voting widget on each of the page of your of your website, which asks people to provide feedback on the quality of translation, if and if you see that there is a signal that a particular page is translated, not in a way that people expect when they want to have a better translation, they can vote for that. And then that can be a signal for you to send a particular page for additional human review. You can use some traffic and like page popularity as a signal and if there's a certain number of visitors visiting a specific Help Center article, then you want to spend your time and your budget on doing post editing for that page, specifically, and you can do this proactively. If you know that there's some important content going up and you want to make sure you you know that it will have have loads of traffic, then you can do this as well, upfront without having for any signal. So with that, I believe we can wrap up, those other things are pretty minor. And again, there are more things for you to consider. If you're trying to scale up scaling supply chain and project management. Sometimes it's done with large LSPs that have enough capacity for you that to run any project. But you can also do this with the translation management system that allows you to tap into a marketplace of freelancers and suppliers. You want to provide more context, ideally, automatically, you want to automatically publish whatever you're translating, at least in a staging environment of your website. So you can give that staging staging website to your linguists, they can see the final result, and then go back and amend the translations. So it's Andrew Hickson 30:49 awesome. Yes, sorry. I'm gonna have to jump in because we have so many questions coming in for you. Would you mind removing this stop to stop sharing the slides for a moment? Yeah, absolutely. Katie said people want to see us. I don't understand how or why but okay. Okay. So very, very quickly. The best question will be given the book that you've nominated the language of global marketing translate your domestic strategies into international sales and profits. There was one question in there, which I will address very quickly, Yulia has asked, Will there be a video of this recording? And yes, there will be all the presentations are being recorded. And they will be distributed by Kate and the SmartCAT. Team after I assume in the next week, but yes, all of the presentations have been recorded and will be shared. So the first question that I'm going to jump to the one that's been uploaded the most is from Eliana. And she asks, Are the content updates visible for translators in the TMS, or they need to go through the whole content in order to identify the new updated bits of text and translate them accordingly? Igor Afanasyev 32:01 Well, I believe anymore in TMS whenever they there's a untranslated part of or of a document or untranslated document should highlight you have, what are the things that you need to focus on. So in SmartCAT, specifically, you can see a list of documents, you can see the overall translation progress, you can see a translation progress for each individual document, and then you can pretty much continue from where you left off, and then go through all of the countries related segments. But as you're going through them, you will be seeing them you will be just navigating, be navigated to them. But you will be seeing them in the context of the overall translated page. So you get the full context. Andrew Hickson 32:44 And the second question here, the next question is from Yulia. And she asks, how would it work for UI localization? Would it be preview based? Igor Afanasyev 32:57 So for UI localization, the you're looking into integrating in some sort of preview, and it really depends on the technology that you're using. Like, if you have a like web based software, it can be translated in the way that you can provide some like in context translation, that's that's the ideal case. If you can do that. For desktop products or for mobile products, you need to install some sort of like SDK or whatever to be able to preview and like associate specific screenshots or screens with with the, with the translation. So there are different technologies for that cool. Andrew Hickson 33:35 Maria Maria Rubio asks, Are there guidelines to quote empty and post editing jobs? Igor Afanasyev 33:45 Well, I mean, it all depends on what you're trying to achieve. And the kind of content that you're working on some content can be like, machine translated by default. And then like lightly post edited, if we're talking about some internal documentation, developer documentation, help and center articles for marketing materials, you want to do some bit of more effort before you're publishing things live. So again, there are no certain guidelines. I mean, it's it depends on your definition of quality and your desire to do things as quickly as possible to deliver things to your clients. So what are your clients want to? Andrew Hickson 34:32 Awesome. The next question, which has been recovered from Esther esecurity highest? She says, Thanks, Igor, for Google Docs, where content teams are often collaborating on different iterations of the source, with people adding comments, changing wording a few times, etc. Does this not introduce too much waste? Igor Afanasyev 34:53 That's a great question. You can set it up. You can set up a process where you will be publishing things when they are final, if you want to reduce intermediate edits, in my mind, it's actually great to have people working side by side. That's the beauty of continuous localization. And that like two way synchronization where people can work on the content, iterate over that. And then that is exposed for translation on the TMS site. Yes. If that's like completely like, like initial versions of the draft that you don't want to send for translation, you just do not publish them in that monitored folder and Google Docs. But once you think that you're done like 90%, ready, you can publish that. And with that, you are able to give linguists much more lead time for translation, which is always nice. And also, what I noticed in the past with this, this approach is that linguists can do some QA in their foot for the content. So copywriters might be, like, deeply focused on providing a proper text for like English speaking or like US market. And then they publish this like, almost, almost ready thing for translation. And then some linguists say that, by the way, this thing completely doesn't work for the audience in Spanish, for example, this just doesn't make sense. And this quick feedback loop allows to quickly like even before the content is finalized, it allows content creators to quickly iterate on the source content as well. And this is like a really, really nice, cool. Andrew Hickson 36:34 The next question we'll jump to is from Mikhail hefur. Anika from lychee. What type of projects would questions just jumped off my screen? What type of projects would be relevant for this automation? Most clients send files and don't need updates on a regular basis? Igor Afanasyev 36:52 Well, that's that's the thing. We see that trend that that our clients that are coming to us and they want to translate websites they do want to have, they want to iterate on websites, this is a website, modern days is a part of a product, you want to trade on that you want to do some A B testing on different types of content. They want to provide those translations. And this is why we are trying to come up with a solution for that for that problem. That's worse. Andrew Hickson 37:22 Yeah, I think that's very concise. There's a question here from Luis Castro. You mentioned various connectors that you recommend, what are your thoughts on BitBucket? And would you recommend it? Igor Afanasyev 37:36 I mean, like Bitbucket and any other systems that work with Kitara, like software based and it's more or less pretty much the same in terms of continuous localization, if you store your content in any version control system that is the original source where you store the content and where you can have continuous localization working with that being in Bitbucket, or GitHub or any like private git server. Doesn't matter. Andrew Hickson 38:04 Okay, cool. Again, there's a lot of questions here. So if you're, if your question hasn't been answered, do check out smarkets. Post during the week, we're going to keep going for a couple more minutes. We'll eat into your break time because there's some great questions here. Martin asks, How do you think that continuous localization fits with the current needs of many vendors of minimum rates? Igor Afanasyev 38:29 That's a good question. Because continuous localization is something that is driven by clients, they want content to be delivered fast. If a vendor works in a, in a world where every single translation job has a like minimal size, and you know, like some some fees associated to that, that doesn't fit into the proper scalable Continuous Replication process. And this becomes that limiting factor of translation based translation job based approaches. So I would just like if you're on the vendor side, I would try to explore how you can work with the TMS that you're currently working in that like true continuous localization approach, we will have a single project where you get all the like one never ending job, essentially. And you just cut off that on a regular basis, like every two weeks or every month, you're just saying that we translated that number of strings in that project. So that's the like the mafia code for that project or the set of projects. But if you are merely if you are able to do that, that you will unlock that continuous localization for your clients. Yeah. Okay, cool. Andrew Hickson 39:40 I'm gonna ask just one more question because we are eating into the break time. Well, it's actually two questions from an livid Tina. I'm going to pronounce it that way. I don't know if that's what it is. But this doesn't work with software. And also, how does it work with multiple translators working on the same project? Which of them gets paid for repetitions and 100%? Matches? Igor Afanasyev 40:06 Yeah, so first of all, contextualization originated from localizing software. This was the initial implementation and need for for development teams. So that perfectly works with software, you're storing your strings in a version control system instead of a CMS. But everything else applies there. If you just go back to my slides and stuff, CMS, look at that as a repository with resource files, that would be the lead for for software. When it comes to working. Multiple people working in parallel on a single project, it depends on the abilities of the translation management platform for smartcast. Specifically, we do allow multiple people working collaboratively online on the same project. And everybody, like whenever somebody goes into one segment, they can like look that for themselves, and they provide translation they get paid for that specific segment. So multiple people can go side by side and, like close, like all the provide the translations for all the missing missing segments in the in the file. Andrew Hickson 41:13 Okay, cool. I'm gonna speak through the last couple of questions. And again, if you want to provide more elaborate answers, we'll do it through the SmartCAT channels next week. Dolores has asked Do you believe that the linguist of the future will need more tools to localize and trance create, or technology will do most of it? And I'll give you another question. While you're thinking about that one, it just says he is asked which system do you recommend? So we'll finish with those two questions. Igor Afanasyev 41:42 I believe with the development of machine translation, most of the content will be going lightly post edited or not posted at all. But for trust creation, specifically, this is where the linguist will be focusing on. So it will give more time to provide like the baseline translation for everything. But then linguists will be focusing on most important content and will be providing like really quality translations, which in turn will help, like improve the automated translation system. As for recommendations, well, I work at SmartCAT. So that's good, isn't that I can recommend. Andrew Hickson 42:18 I was gonna say, you can just knock that out the park SmartCAT. Obviously, please don't screen capture that and put that out anywhere. I don't work for SmartCAT myself. But anyway, Igor, thank you very, very much for your presentation. There are a lot of questions that you're going to have to dive into in the chat. Would you like to announce your winner of the book? And again, the book is it is the language of global marketing translate your domestic strategies into international sales and profit? Igor Afanasyev 42:53 Yeah, I like the question about like this Google Doc connector and about how to collaboratively work on source content and and translations. So I don't remember who was the Andrew Hickson 43:07 looking behind through that negotiation? I can't see it. Kate Vostokova 43:19 No, no worries. We will find it later. No worries, no Andrew Hickson 43:22 problem. Awesome. We will get in contact as soon as we figured out who asked that question. And we will let you we will let you know that you are getting a book. Igor. Thank you very much for your time. Kate Vostokova 43:33 Esther says that it was her. Ah, yes, Andrew Hickson 43:38 Esther curio, congratulations. There is a book on its way to you