 And there we go. We got some smooth music there for our intro new welcome. It's open education week, 2024. And we have a very special session now for our opening event. It's actually our official opening is about an hour's, but this is the opening for the opening. And I'm Alan Levine from the OE Global Studios. And we got a really fantastic guests from different parts of the globe. And we're talking about how we talk language and translation and AI. And the best part about this is this all kind of came about organically through some conversations in our community space. And to me, I think that's the most exciting piece. So we're first gonna ask some people to say hello and let us know where they are. And then we'll talk a little bit about what happened here. And then we're gonna get into the meat of talking about this AI translation and issues. And who knows where we go. That's the format of these conversations that they're open. And so I wanna welcome everybody who's watching in the audience. And of course, if you have any questions or comments, please put them on stage. And I will do things like put my colleague, Ila's whoop whoop on screen. And we can talk about if you can translate whoop whoop into different languages. Or that's probably not the most relevant question. Enough of me. I'm going to just start. I'm gonna go around my screen here. So I'm gonna ask Shishumna to say hello, let us know where you are. And I have to say, I apologize because you have the latest time zone. Oh yeah, that's okay. Namaste all. Namaste Alan, namaste Paul, Stacy and Dominique Collin. Thank you so much, Alan for giving me this great opportunity. I'm from Hyderabad, Southern part of India. I'm a OER practitioner and OER enthusiast, I must say. As in when I'm an OER global platform, I started connecting with people and trying to understand and trying to unlearn certain things and relearn. And I'm trying to still learning a lot. And part of, I'm part of this conversation, I think just because of being open. That's how I would like to introduce myself and staff. That's fantastic and definitely open and the connection was made with the gentleman to my right on your screen, Collin. And you're actually on vacation right now, is this true? Yeah, so I am on vacation. I'm in the beautiful city of Granada in Spain, but I normally am based at Nantes Université in France where I run a number of things including hosting many of you at the Open Education Global Conference well already two years ago, actually in 2022. And starting collaborating and starting thinking at that point, very hard on the issues of multilingualism. And what we're going to talk about today is in a way the actual practical applications of that thinking. Excellent. And boy, I can't wait to get into this. And next, almost at the last second, I appreciate Paul Stacey coming in who's definitely been tracking and following and heavily involved with things going on. Like, you know, I guess Paul, people come to you and say like, you should be the expert on AI which is kind of, it's not the mantle you wanna wear, right? Hi, Alan, yes, I'm delighted to be part of this. Thanks for the invitation. I'm based in Vancouver for those of you that don't know me. So I guess Sishuna, I'm at the other end of the time zone. I'm at 6.30 in the morning for me here. And I'm up bright and early, but certainly the AI space has been really volatile, I would maybe describe it as this past year. And I've been as many people in the open space trying to follow it and understand the complexities of it. And I also would say that translation and the ability for AI to support translation is a very interesting phenomenon. And so I'm eager to hear more from our colleagues this morning, Alan. Excellent, thank you for joining us and getting up early. And next we'll go to the other side of Canada, zipping right over my head because I'm in the middle of Dominique. Like Sishuna introduced you and I set out an invitation and you showed up and thank you very much. Welcome Dominique. Well, thank you and bonjour to the OE global community. I met all of these people on the screen in Nantes. So thanks Colin. And I saw some of you in Edmonton late this October. So I've been following the OE community. I'm actually very happy that the OE global movement has really, I would say, embraced the world because I started at the University of Toronto in the 90s and really looking at machine voice recognition because I was a phonologist. And we did so many trial and errors and there was not a community to really support all this work. So I find that what's happening today for me at the end of almost my career is just like, at least when I'm gonna leave, it's good. It's in a good place because now there is a whole community working on this. So I am from Toronto right now, originally from France and working really a lot with Shoshuna to change the practice of educators from K-12 to post-secondary education in embracing openness. And it's a challenge, a huge challenge. Okay, so glad you could join with us. And as I said in our interviews, I think it's always important when we have a linguist in the house. And so, and also apparently we have to quote the movie, A French Connection. So I think the story starts with Colin who brought to the world this AI for teachers project which was multilingual from the outset. So can you give us the overview of how that came about, Colin? Yeah, sure. So the project itself could deserve a long time but it's important to say it's a project that is supported by the European Union and European Union is, it does matter in a way openness. So you're not allowed to produce with public money in the European Union material that then you're going to sort of just put under a closed license. So it was a natural thing to go for openness there. So the topic was AI for teachers. Sort of, we already had identified that teachers were going to be working in an AI world. And so how can we train teachers? So this was the topic and was built one thing which was a MOOC and which is also open. And this textbook which is roughly 250 pages of PDF to give you an estimate, I think there's about 70 chapters, small chapters, right, explaining what the teacher should or would want to know to move forwards. The project was multilingual from the start because there were five European countries involved. And so right from the start, we had to think out the materials in French, in English, in Italian, in Slovene and in German because not because of Germany, because of little Luxembourg, but who convinces us that we want it in German. So that was the first challenge, but what was interesting was to produce all this material and there's a textbook. And then to realize that once you put the CC license on it, you'd actually not gone far enough because there was still so much hard work to be done by anybody else who wanted to get hold of this. And so this is where, let's say, I'd probably have a pass on to Tsushima or other people saying, okay, so how do we do it? Once the people have been nice enough to produce something with a nice license. I mean, what can we do? And the reflection is, what should we have done in production mode to make sure that it's not just about the licenses, how can it be just as easy as possible to then take over, get the material, reuse it, remix and do everything that we're saying we should be doing. So let's say it's wise, it's a practical story of saying, how can we make openness work in a multilingual and multicultural setting? Yeah, I think I would like to have a word here that about this AI for teachers, it's a very interesting topic for me since I think before chat, GPT-3 has really emerged. Like I've been thinking and then writing as a your practitioner, I'm thinking that I'm trying to create a course and then trying to see what are all the available materials and all, but then it's always because of some other work, I'm just keeping this aside. And suddenly on a platform and I saw this book with, as Colin just pointed out, it's not about the license, but what we can do when we get such kind of a material in hand and the moment I saw that, oh, yes, exactly. This is what I wanted to give it to teachers because after chat, GPT-3 and a four and a lot of other tools have come up and teachers always think that the first question is, I really like in this book also it's there, will AI replace me? The teachers will, first question is that, so how well we are going to utilize that? I always support that, but technology cannot really replace the teacher cannot do much what a human can do, but still it can accelerate and it can act as a catalyst. So that's how I wanted to translate the book at the same time. I wanted to add a nativity to it and what you call is the popular ones or the popular way of using technology in this part of the globe. That's how I thought I will do it. But then when I saw this and then I got a chance that, okay, we do have India wonderful translating tools available. And I thought, okay, let me do a practical, I mean an experiment kind of a thing, let me translate with it and see how good they are the result would be. And then how much time we really cater to post edit. But the main aim for me is first thanks for that wonderful book I read and thoroughly enjoyed and while translating, I'm trying to not just translating thanks to the license that you released with, I'm trying to remix it. I'm trying to come up with a new revised version maybe very soon, but before that, I think I've given myself a time. I'm kind of crowd sourcing, you might call it. That's what Colin also said that we are trying to approach different teachers to translate it. And then post editing, I will be doing it adding some inputs from this side of the globe. Excellent, now, oh yes, please Dominic. I'd like to jump in because I like the idea of translation from the multilingual perspective, but with my educators, there is a European connection calling like we're trying in Canada to import the common European framework to actually assess students in foreign languages. So French, Spanish, German, and it is in the curriculum and they are reformed by the ministry. But what is interesting is that I find that educators and I like the idea of Shoshona crowdsourcing, they are so creative. So what they are trying to do is to look at AI providing like building a database of text for students to practice, to actually get to the different levels. But what I find interesting is that the translation there is not multilingual, it is actually proficiency-wise. So they are looking at text in the AI that could be used from a grade 12 perspective, grade 10 perspective, and how AI can actually adapt those texts. And I thought that this was brilliant. So we tried testing it with a few colleagues in Toronto and AI is amazing. I asked, can you provide me text for an Anglo students because we make the distinction in Canada between the Francophone systems and the Anglophone systems where they are learning French through different programs like French as a discipline or as immersion. And AI was capable of giving me different texts that they could actually use to practice or for teachers to use in the classroom. And what was fascinating is that AI, then I said, can you really adapt it to a lower level or a higher level? So I am bringing the multilingual, but it's beyond like, it's even like the dynamic of language across the proficiency levels, which AI can really help. Now, the problem that we encountered was of course for the oral text because it's very difficult to get AI to actually recognize we had AI translate from, and this is what I wanted to bring. The translation happens in different categories. It's multilingual, but it's across written and oral speech. And then it says, well, like you can ask AI to translate written text and actually make them sound text. But then that was terrible. I have to say it was terrible. So what I like is like Laurence Venuti who wrote a long time ago, a beautiful book called The Translator Studies. And he says, language is a repository, he actually quotes and he says, language is a repository of ancient errors and a treasury of potential truth. And I thought that this is so amazing that there is actually a place where we can improve AI and it is really from written to sounds because when AI can assist teachers, doesn't matter, can we imagine like in terms of the plurilingual aspect, moving from the written to the speech level. And it is really where the educators who know nothing about AI, that's already where their mind is. So that's why I wanted to bring it. Like we are experimenting with AI, but educators are already because of they are practitioners and they are in front of the students. It doesn't matter the level. They're already using documents to make them learn and be successful in learning languages. So we really need to engage this community because by giving us challenges like this, this is the way we can really push AI. That's what I wanted to say, like translation happens in every dimension. And almost like there's so many places to take this, but thinking too, like our understanding of what a textbook or a work is, it's fixed. And when you're talking Dominique and we've seen some of this ability when you say like present this or explain this in the manner that so-and-so at so level is like almost the text almost becomes dynamic, which is an interesting potential, but like comes with a whole back load of issues to question about, you know, whether it's explaining correctly and all that, but like the potential layer is huge. But I'm interested in this idea about rethinking what we think like a body of text is. Well, I was actually answering Sushuna because she says she's relying on the natives and on to actually give that dynamism. And I think it is key that we are working in tandem so that what we call giving the domestic angle for something domestic being a level or a language, we need to have people who know and that's where I don't think teachers are gonna disappear because we need them. This is where they are excellent. This is what they know and this is where they excel is actually being able to assess levels of proficiency and we need them to actually help us make AI responsive to that. Any thoughts, Paul, from your global perspective I mean, I just will say like it is kind of exciting, isn't it in the sense that the promise of OER has always been like, you know, we can translate it, we can adapt that, we can localize it. And now we actually starting to have some tools to expedite that whole products. And so I find that quite exciting. I always wonder when we're talking about translating a body of text, like you're saying Alan, from one language to another and wanting it to be relevant to the people who not only speak that language but potentially are from a particular part of the world. I've been starting to think a lot about whether the localized versions of language models. So you've got these big language models that are general purpose, but perhaps we also now need to think about their localization such that the translations they produce would be not only in the native language of the speakers of that region but also localized to their context. And it's sort of like the possibilities are really quite remarkable, Colin. Colin, yes. So I would sort of half disagree. I don't think, I think it's about sharing and meeting halfway in between AI and human intelligence. And that bit of the local, let's say the local adaptation is what we humans are going to be good at. I'm not sure I'm going to really want AI to do that. This is where I've come up with this textbook, with this idea of what I'm calling a half baked book. So my idea is that when we see a textbook or when we see any object on the internet, if you're not computer minded, if you're not a techie, you're going to just see the text. But if you are a little bit of a techie, there's so much absolute rubbish happening underneath. In press books, which is a great platform, we still end up with having three layers of tags. We've got the tags from XML, we've got our HTML, we've got the tags from press books, we've got the tags from WordPress. So all these layers are actually getting the way if you're trying to actually get systems to work. And my idea was to say, well, can we use AI to just get past these obstacles, which all these tags and all this technology and then give what I'm calling a half baked version, which is a machine translated version, which I really don't want it to be, you know, suit just this particular part of India or this particular part of West Vancouver or something. I mean, it's a half baked thing. You know, I don't know if you can do that. We can buy that to sort of bread in the supermarkets in France. You buy a baguette, which is half baked. And you're supposed to then do the job yourself and sort of put it in the oven and finish it off. And if you don't do it, you just, if you take it out of the fridge and eat your half baked bread, it's no good at all. So this is the concept in a way. And because, yeah, I mean, what you were saying just a minute ago is the thing is with OER is we're very good at sort of giving generalized ears and just saying, go do it. But doing it is hard work, right? And on the ground, our colleagues are facing huge problems with licensing issues, with just our issues of getting hold of a PDF and say, what do I do with this PDF? And how do I work with it? And so on was being practical and saying, no, this is how we can do it. Look, we're just bringing you an object which you can now work on and you can get your collaborative platform, which we're still missing actually to say the truth, right? So place where Sushuma's dream of saying she's got many different people to go and work on. How do you get this to work? We still haven't got that right, right? But a good platform, which would allow this half baked book or half baked, whatever OER to arrive. And then AI has done its job and certainly we don't really want it to do any more from there onwards. We want a human intelligence to work. That's the dream. Maybe that's, sorry. Go ahead Sushuma. I was gonna say to Colin, I think that for me the image that always comes back to me is the middle-aged text, where the text was very small and what was important were the notes. Because I think that we are at a stage where we need to train and educate as well at the same time as providing half baked documents or ideas. So for me, how do we do the training around? Is it making those, as I said, potential truth visible? So I think that I like your idea Colin of the half baked but how do we do the half baked and still give the tools to the people to understand how you actually translate everything like including HTML languages and all of this. So it's like the idea is providing a text but it's everything around it which is important as well for them to grasp the shift between like the letter world to the digital world which is part of the OE and the AI but a lot of people still do not know what digital means. We do a great technique, absolutely. Yes, but then I just wanted to comment on what Colin said as a technologist, right? Like he wanted that to work on a certain way to help collaborate but not exactly what a human can do, a human only can do but then technology can help us to have a platform like a collaborative way of like together we can translate or localize it the way we want it but then this is just, I know it's a novice idea I just want to present it to all the technologists over here too and I've been using I realized after three chapters I tried four tools, AI tools to translate and I found very proud to say India's Anuvaadini is the tool that I'm using it's supporting 29 Indian languages and several foreign languages already and it's called Anuvaadini I'll try to give the link to and what I realized is it's really interesting and then I thought that why don't somebody can come up with a plugin where because it takes a PDF and it translates and then it on the live we can just edit that PDF and download that way the machine is learning maybe with our editing what it translated and what it should be what as a user I want maybe that's how a machine is learning maybe I'm thinking I don't know exactly the backend and all the things about Anuvaadini but what I thought is when one can edit it why not three four people collaboratively can edit the same document and have the tool inside a press book or inside any platform that accepts a PDF so I just wanted to put it across maybe as a technologist they can think and come up with such kind of a platform because I love the way we can do in any where the tool used to be is that hypothesis we used to use earlier and now we use Google Docs to collaborate and do and see the iterations and even for the translation if that is the case in the work and not just for the translation any kind of a work that we'd be doing it would be really interesting because I just wanted to point out that Colin because you said the technology should do that part and the human should do this. I think that the comments are the human intelligence. Yeah, yeah, absolutely. But maybe curious just to jump in on about this half baked thing I'm really intrigued by that idea Colin and I'd be really keen to hear you say a few more words about what the other half of the baking is that you're expecting people not AI to do because if we're talking about half baked bread you put it in an oven and it becomes bread or even a frozen meal, right? You warm it up and now it's a meal but I think your half baked thing is it's sort of like that but I think it's a bit different and I'm just curious to hear more about your concept of it. So I do agree, right? So I've done it in, I haven't been able to do too lagoo but I produced on Friday night a Chinese half baked book, right? So I can share it with you. I mean, it's been, I had to get back to coding I hadn't been coding for a long time so to be able to get it and to say the truth, I used the chat GPT it was horrible to say the chat GPT actually worked really wonderfully well for that and it was just chat GPT, please do what you're told and I mean, it did it. So, okay, somebody's sharing. So this is on a platform with some Italian colleagues have put a press books platform there and so we're doing tests and I checked it with the Chinese people and they're saying, yeah, it's okay it doesn't work, right? Because there's too many things that are wrong in the translation and then there's a lot of things you know, when you've got a book like this you've got so much, you've got videos and images they also will probably need translation or they will need adapting. So yes, there's much more than just putting it in the oven but I also do know that this per first step which is saying here the whole structure, right? That part that you don't see if you're not a technologist all that has been respected. I believe that is very important because from here onwards it's just a question of giving an account on press books on this press books and doing it. And all I have to do now is change in chat GPT in my prompt, change Chinese to something else. I tried with Spanish or Polish and hopefully people can do it. I'm not sure they'll want to do it but I'm interested in the concept of saying does it help to do that? But it's just saying to the people, hey, just take the XML from press books and work from there onwards. That just doesn't work. It's too much work. It's too much hard work. And I'm thinking of half baked it. I'm thinking of those services where they like deliver to your door the ingredients and the instructions so you can make a meal. But I kind of want to back up a bit and talk about first how it is done. First of all, for your original book, Colin, how did you handle the translation? Did you hand out documents? And was it done in press books? And then also for people like Shishuna, I'm wondering is she going through page by page copying and putting it in? Like what are the mechanics of that process? So that's one of the reasons for which I also done the half baked book is we had to actually do this and spent most of my Christmas holidays trying to sort of get some texts that have been written by some human translator in on Microsoft Word and sort of copy paste that into press books and just then spend an enormous amount of time trying to clean up all the tags and clean up everything. So it doesn't work, you know, if you have to go through these translation human translators will say, no, no, we don't, we only work with Microsoft Word. I don't know why, but I mean it seems if it was the most wonderful way of working, which it's not. So you end up with this file that has been prepared by somebody in Microsoft Word and passing it back into press books, which is just like passing it back into HTML. I mean, because it's just as clean as that, it's just an absolute nightmare. So yes, we did it by hand and it's just hours and hours of work and my belief is that if, you know, if you had the possibility of working directly onto HTML with a nice viewer, then it should be, it should be all right. Absolutely. That's the same case as I do agree Collin that many people think in the, I don't know whether it is an ICT question, a basic ICT question is that many people are very comfortable with the Microsoft document and they would love to do it here. That's just the reason, even though I introduced, I trained them using these three tools and I told them, this is the way to do it. Even then they love, oh, just give me the document and we just do give you the document. Yeah, I know it's a lot of, Marma's effort to get into that, into Presbert. Yeah. Can I maybe suggest something here? I think that for a lot of people we're still in the live rest mode, which means it's like a page and they are words and for translation, you know that when you pay them, it's either by the page or by the word, by the way. So we are still into that mode of the letter, as I said. And Microsoft Word fills all of the need for that. Moving from the letter world to the digital world I think needs to be done at the same time as we are doing OER and AI. Like that shift has not happened and even for accreditation, I am on the executive committee for York regarding promotion from, I mean, if you want to become like an associate professor to a professor and all of these different loops you have to go and it's still the citation and it's still the book. So it has to be a press book rather than a bunch of links to H5P for example. So we're still in the book world in academia and in most of the time. So Microsoft Word, I mean, of course fills that easily. How do we make that shift? I think is a question we need to figure out as well. How do we push for no longer the assessment being the word but being another the digit or whatever it's gonna be. But it's a different unit anyway. I think this is an interesting idea. Like what is the book? And you kind of poking at this a bit earlier too, Alan, I think it's like, is the notion of a textbook still going to be a relevant notion going forward? I think that is a really big outstanding issue. And certainly people like David Wiley been asking questions like maybe it's the prompts, you know, not the book. It's the prompts that are actually the secret sauce, if you will, or where the learning happens. And I just wonder Colin, whether even the AI for T book, while it serves an immediate purpose of meeting the needs of people who are still expecting textbooks, thinking going forward whether that knowledge is actually embedded in the AI itself. And you need the AI as a kind of replacement for the textbook. I mean, it is what we're working on now. It's basically can we now feed the textbook to an AI and then have the AI just become sort of something we can have a conversation with and just talk about things. But again, it's nice to have a conversation if you know what to ask, what to talk about. And the advantage of a textbook is that you can go if you don't know anything better, you can go from page two to page three. And maybe something in between where there's the ability to have those conversations with the text, you know, it sounds like Paul's like saying like, if you have the body of work embedded into some kind of embedding of an AI that then could deal with translating that and not just drawing from everything else out there, then you have the possibility to be that localized. But Boyd language is not an easy thing. Like we're talking about the difficulties of translating from Microsoft Word to HTML which is highly structured. I mean, to get into language and nuance and meaning. And I'm curious to assume, especially for like symbolic languages, like it's gotta be almost a very manual process to do this translation and review of what AI does. Yeah, it is. And for people looking outside and say, when I say that, oh yeah, I'm taking a help of AI to translate. Or just it on a click, it does everything. That's the notion that everybody have. But then how much effort that we have to put in in terms of not just the translation but also about the structure of it and what the original document is, whether you want to match the context of it and the consistency that you want to maintain or all these things that really comes into matter. But then I realized that rather than sometimes, because the translation with me is, it's with me since a long time because I've always been into translating things from English to my language. So like I find it very easy to do from scratch than using machine. But now I'm coming from that background. That's what my feelings were. Like now when I'm looking at these tools and then trying to experiment and to learn more about what exactly machine can do, whether it can really reduce the time that we spend. In a way, yes, in a way, no. Like when it comes to like Dominic said, when it is to do domestically, make it relevant or contextualize it, maybe it is from subject point of view. When it comes to technical point of view, yes. Sometimes we feel, no, no, no, not this way. I just want to do it from scratch, from like writing freshly than looking at it and just thinking, mulling over the human brains and then trying to do from scratch. That might be easy. I don't know when machine will be that intelligent to do so. I think there are questions in the chat that are quite interesting regarding any prompts. And thank you for raising this. Like any suggestions for good prompts for translating text using open source LLMS or chat GPT. I'm sure Colin and Shoshuna may want to answer that question. I just want to actually do a plug in here. I think that we are trying to always reinvent and I would like to point an amazing text to read which is called the translation studies reader by Lawrence Venuti and he looks at translation through like the theoretical aspects from the 1960s until now and looking at because every time there is a new theoretical framework. So we had the comparative framework, then we had the culturally responsive framework and then we had the politically correct framework. And I find that sometimes reading those, the way different theorists try to actually describe translation can help us do these prompts. So for me, I let Colin and Shoshuna answer the question but what I would say, I don't know if there is a very specific prompts to do. I think that with AI you just try your prompts, see the answer and then it's like more an investigative process and there's not like an ideal prompt. I think we all are into experimental mode and we need to figure out what is that next stage in translation studies because we went and followed, of course, like linguistics, then textual linguistics and then culturally responsive and it was always with human intelligence and we haven't tested with the machine. So I think everybody needs to actually try crazy prompts and we should do a crowdsourcing exercise. What are the craziest prompts that actually revealed, as I already mentioned, potential truth and write a book about this? Yeah, I was saying here like, yes, prompt engineering is definitely the thing that's how machine learns. But then I feel this is, please correct me, Colin as a technologist, I think these are more of a statistical based methods than a rule-based ones. So once you give the prompt, first, what I observed is when I wanted to do it on a same machine, use the three language model and when I do first in one tool and then I come back to Chagdipiti or a Google translation and then it picks up few words from there. And then if I don't, if I do it from first from scratch from Google translate that the results are really bizarre. I mean, when I do from a better one and then come back here, it may be that intelligence is really good that it can crawl and search what I have done earlier there and then it picks up and comes here. So the prompts might, which worked for me may not be exactly the same way it will work with others in the other part of the globe or even for that matter in other, just in other machine, I believe, I think that there's no, this is the prompt that to get that. Maybe that might differ. Well, I'm interested in the answer by John because he says that he wrote a kind of a machine or tool that actually triggers different prompts. I'd love to see it. That's John Gondel, our technologist who's probably writing Python as we speak. He speaks Python fluently. But yeah, I think this idea that we could reuse prompts is a little bit like, I don't know what's possible. It's always changing. And that's the idea of the LLM that it's going to change too. And so you won't get the exact same response, which is okay. But we want to get the context. I wanted to ask Dominique, because you mentioned doing some of the translation of H5P activities. And I know that's an interest, I assume that too. And I've seen some of the demos that the leader of H5P has done where creating, like actually creating H5P activities because it's very structured. And so you can say like, I want this activity and it needs to be structured in this way to have like the blank and the options and the feedback, which kind of works. And so I'm thinking, would that work with if the textbook has a format or if there's a way to have it, like deal with the tags or the links, so you don't have to do it manually, like can it be structured in a way that the AI can give you that output? And so that's some of the interesting things that I've seen people do when you ask it to do programming tasks because you don't just say, do this, you have to give it a little bit of framework. Well, I think it goes back as I mentioned in the 1960s and 70s and Chomsky was behind that universal language. I mean, that body of work is still really relevant today. And yes, if you are looking at translating things that could be like really based on universals, I think that this is easy. And that's why sometimes pedagogical activities when it's really translating quizzes and things like that. And I have to be careful because just as a point and a story, the first time I talked linguistics of the University of Toronto in front of 200 students and I gave the first test. And I realized that when the language is not your first language and I have to go back to the notion of bilingualism, I thought that my questions were really clear but they were clear for a native. But they were not clear for people coming from all around the world. And this is stressful during an exam time. And so I thought that even like a simple question for a quiz on intro to linguistics would be understood by most people but I saw like the fear in the faces of my students. And because our English was not their first language then they started thinking, what does it mean? Because they didn't have that certainty that they knew. And therefore their mind was looking at same thing as prompts. Like, does it mean this or does it mean this? And for them it was key because you know, multi-choice questions you don't get the question, your answer is wrong. So we can translate things like that but again the notion of mother tongue or speaking many languages, I think we need to take that into account because prompts are easy when you go from one native to another native but in between there is that continuum of people speaking many languages because of migration and not even understanding either the first language or the second language which is translated. So there is that group as well that we need to address and ADI if we rely on correspondence of universals it works for native speakers but it doesn't work for the others because they're not a certain that they understand. And so the notion of the ADI is key. So Colin for the next language that someone asked you about like, how would you go about it? You sat down and did the Chinese. I mean, you talked about that. I mean, what would be the ideal setup to work from sort of like a systemic approach? That's what I've been trying to do. I've been trying to be systemic. I mean, I really only have to change one variable which is language and I obtain a half baked book in that language. And here is where there's a trick in what I'm trying to say is that I'm understanding through the prompting that you're being ambitious about the quality of the language result. I am not, right? I want something that is going to be easy enough for somebody, you know, collective work, collective intelligence to take over. I have got no ambition to try and get it even better. At one point I was speaking with people in Madrid who were wanting to take over the task of translating the textbooks. And I said, look, wait, I'm doing it with AI and I'll pass you this. And they said, well, we're not interested. It's going to be very boring. There's going to be nothing left. A bit what Paul was saying earlier, right? If everything is done and all I've got to do is put it in the oven, there's nothing left. And in fact, I really strongly believe that there's a tremendous amount of work left, right? In order to then make it into something interesting. And this is where we've got to convince people is that, you know, AI can only go as far as this. And we, again, we, this case, me as producer, I don't want to go and give you something good. The Chinese thing is rather fun. I mean, if you go on to it, it's fantastic. All you can see is the name of the two authors are written in our alphabet. I understand nothing else. I haven't got a clue myself what I've done there. I mean, it's, you know, it's when we're saying our students don't use chat GPT to do their homework and not understand what they're doing. That was exactly what I did there, right? I produced this thing with a lot of hard work. Let's face it. But then I sent out and just thought, you know, I might be getting some abuse here with saying, you know, how dare you, you know, this is absolutely rubbish. And people said, no, this is a great starting point. I mean, we can actually now work from this. And this is what matters. I mean, this is where, as far as I can see the concept, I'm interested, it's in this. And furthermore, it's, you know, if we can do it for a huge variety of languages, then my idea is saying, yeah, okay, you know, somebody who says, can I have it in my language? We can try and do it. I mean, there's a little bit of cost because you have to spend two, three hours of prompting with our friend at ChatGPT. The reason for which ChatGPT is that we're able to actually get it to go through the tags. I mean, we tried other things like Google Translate, which was mentioned earlier, or D-Pel, and D-Pel, even D-Pel Pro was just not very good at going through the, you know, it would take a tag and that's sort of half destroyed. And that's it, you know, if you've got HTML or XML and some of the tags just vanish during the translation process, you've got nothing. So hopefully we've got something systemic, but up to a certain point, and this was a point I tried to say at the introduction, there's so many things that we could have done better by systemically or systematically having thought it during the writing process. I'll give you an example, you know, at some point you want to show a really cool chart, right, a chart, but it's got text in English in the chart. And then you've got suddenly something new to rethink about what am I doing about this? We've got an issue with the videos. We've got videos which we tried working with a company where we filmed them in such a way as it's me doing the speaking, but you've actually got the image which we've translated into five languages. So we've got that going and we've got subtitles in five languages. But I haven't got a clue what Tsushuma's going to do with those videos in the interview. I really haven't got a clue, right? I mean, it's a shame because I think systemically I should be able to give her a proposal. I think the best proposal is just redo the videos, right? And, but it's a shame, I mean, because it's a lot of money and it's expensive to produce videos. So why is it that, so all these things we had thought in the terms of the five languages we were working on, but we didn't think much more universal. I mean, you wouldn't, you would also think, you know, I mean, how dare I even think about this? But I think it's a fascinating question is how can we do it? And at the same time, not be global, not just say, okay, there's just one way of thinking things and just makes it very boring because it's a shame if you don't have sort of bits, some cultural points of views or some amusing bits and whatever. So how do you do that? It's, I think it's really interesting questions. Well, I have a question, Colin, because this is actually even treated in the book I just mentioned, are we thinking of translation as just a communication act? Because if it is, then of course, we're gonna have to look like who is our audience like and or is translation really a social endeavor as well? Like these are important question. If it's a communication act, yes, systematicity, correspondence, universalism will work. That's part of the day I can do. What I'm hoping is that humans are going to play the second part. That's where when I'm trying to, we invented for the textbooks the notion of somebody I'm calling the guest editor, which is the person who will coordinate perhaps the work in another country. But that's exactly what you're saying, which is going much further than just the act of just technically getting it translated. That has to be done by somebody who's actually, yeah, who's completely fluent in that language and able to actually understand what the issues are, which might not even be the same issues in say, Quebec and in Paris. It's not just even language, it's much more than that. But we have to remember that we, I mean, a book in itself is worth nothing if there is not a readership. So if we don't engage that readership, we can communicate as much as we want. It's like a loop, we're just communicating with ourselves. Yeah, I was just asking here in the chat, Dominique, that like, they're doing a course in English and then moved on to French. And then we had the same issue that how do we deliver the same videos, but in French? So how are we going to do this? And we really, my over, and then we came up with a simple solution that, okay, then let's take those slides and do it translated in French and then let somebody talk in French, okay? So, but yeah, when a person is not talking, that might be, or having captions also we tried, I think, Dominique, having other language subtitles or a captions that we call it. So maybe that is already available for us, for videos. But then, as Colin said, like if you want to get entire essence of it, entirely in the language, target language, I don't know what I'm going to do, but definitely. Everything has to be adapted, I mean, including images, video, like everything has to be culturally responsive and adapted. So I think prompting the machine to go and change something for something else. So prompting, again, like prompting the machine to actually be sensitive to that difference is going to take some time. But I think that this is where we could help, not the machine to do it, but to help people on how to find things by asking the right query. Is there something about maybe putting aside our expectations for really perfection? I'm thinking, as language translation has improved, I can reply to messages in languages I don't understand, enough to get the gist of it. And I let people know that I'm not fluent in X. And that's informal communication. But I'm wondering, we have this ideal that we can get it really down pat, and we can probably get closer than we could before with some of these tools. And images now, there's the ability, when you upload an image, the text can be detected automatically. And then theoretically, it could be translated. So there is potential there. But maybe the content doesn't need to be perfect, it needs to be, as Colin says, half-baked or good enough that you can taste it. I'm not saying the teachers should consume the half-baked book. It's completely, I'm just saying it's where the collective intelligence has to take over if the people want the project and say, well, there's probably a clearer path from there onwards to able to do something out of it. So I think this sounds very similar to the promise AI has had for programming, where people are using it to program, but the expectation isn't perfect code. And it's often like start with some existing code and have it improve that code. It's sort of like the half-baked version, right? You get it so far and then the person can kind of put the finishing touches on it and make it as good as it possibly can be. But the expectation is, but even with that, what they're talking about is like a three times productivity gain for program, for example. So if we get a three times productivity gain for translation that leads to something that's not perfect, but really quite a huge step forward, I think that's quite significant. Well, I think we have known that with open source tools. Like any open source tools, you have the beginning of the code and then you were continuing according to your use of the code and making something different another tool. So I think we have already been there. And yeah, I like the idea of all that. It's just the beginning, we're just giving the foundation and then you take it from there. Which, I mean, we do that when we write a text, we're inspired by other texts and other authors. The same idea. All right, we have a few minutes left. I, let's take nominations from the studio and from the audience, like what language should AI for teachers be put in the oven next? Let's take, let's take requests. Anybody out there? Or what you're hearing Colin, what other interests are you hearing from around the world? We're preparing, well, because we're in Europe, we're certainly going to answer two queries from European countries, that's clear. We're working, it's rather fascinating. Again, this is another open education, global sort of success. Well, I had this person, I'm sorry I haven't got her name from Australia contacted me saying, oh, I'd like to translate it. So what's on earth is she going to want to translate it to? And so she said, well, to Arabic. So what? Okay, but then we've got colleagues in the African Arab countries. So I'm hoping to put up a group or to see if we can help a community. But as I said before, one thing is what Sushuma is doing is supposedly to work on community work where she's going to be sort of, let's say, piloting it. But in a sort of project where it's going to be more people in different places doing it, and not necessarily with a natural leader in the sense of somebody that everybody already knows, might be more tricky. So we're missing some sort of type of platforms to do these things. She's got those couple of languages there. But then also interestingly, and there again AI can help tremendously. We were contacted by very small languages like Maltese and Albanian. And this is rather interesting because these are countries for which the effort is rather tremendous with considering the size of the country or the size of the community speaking that language. And if AI can again produce that first shot, it does give them the opportunity of having, let's say some up-to-date resources in their language. I'd like to go back to what you said, Arlen and Tushina, the context. Because I do agree for the European context with mobility, what you need is that foundation understanding. Now in some countries, Switzerland, Belgium, Canada, Cameroon, where the goal is bilingualism or pluralism. Now the level of proficiency because you said, Arlen, like what is the machine? So I think it depends on the use of the machine. If you are in countries where the expectation is not only to understand broadly, but to train people to actually become experts in going from one language to another, then the use of AI will be different because now we need to have that tool, assist making proficient people in all of these languages. So again, I think that the context is very important. You cannot have just like an approximation when you are an expert at translating at the federal government, when there are like political debate happening, you need to actually understand what's going on. So to make those people bilingual or multilingual because you have like 10 languages in hospitals to serve the population, which is the case of Canada. 55% of the population in Canada, and I think it's about the same Paul in Vancouver, is actually born outside Canada. So we're not looking like that approximation. We need to actually engage these people and be able to actually make sure that they understand what we are saying. So again, I think that the localization that you brought, Paul, I know is an ideal, but for some countries, that ideal is quicker needs to be achieved quicker than others. I'm just contextualizing the conversation. Thank you so much. I wanna let Chishuma tell us like, what's next for her project? Cause she started this. Yeah, because as I said, when I told this in our community and many teachers from our national language Hindi and also our Anwarthini supports a bunch of texts to go in one go and come out as a Word document. That's what it is doing. I think very soon, as Colin said, it doesn't matter the quality of the translation, but then we can get it immediately. It's two languages. That's what we are thinking is Canada and Hindi. Because Canada, we are going to try another local tool developed by university professors in Central University of Hyderabad. And recently I spoke to them that they were working on a tool, these machine based translations long back. And then that machine is pretty good at the language of Canada, that is other than Southern Indian language. So we are going to try this book with that language pretty soon. And with Anwarthini, we are going to do not really concentrating on the quality, but immediately what we can get that half-paked text as Colin said in Hindi pretty soon that's gonna come out. All right, well, we look forward to hearing all the updates and the action. And I really appreciate everybody coming in and our studio audience. And unfortunately, I have to close out this conversation, but we know where to find each other. And thank you everybody here in the studio for being part of this. Thank you so much. Thank you. Bye-bye. Have a good day. Thank you.