Thank you for being here. My name is Juan Re 10 language and I'm also one of the Hollywood section managers. It is uh great to be here and great to see you all here. And so we're gonna start off our session here. First of all, with uh our first be here, Dennis Kras the call um a little bit about him. He's an A I venture founder and CEO for Vin me dot A I, he's a co under in A I translation technologies with 10 years of experience of Blue Cloud innovator uh phd with Harvard Innovation for Economic Development Program in 2014 alumnus uh and a Y Combinator School for 2019. And his topic today is A I translation and dubbing with emotional accuracy, opportunities and obstacles. Take it away. Thank you very much. Uh First of all, I always glad to meet the uh people from uh SM BT, feel like a family and always meet again and discuss different things sometimes. Uh I meet people like uh year by year and they see our pro professional, his professional role there. So they a technology is uh really uh like every year and this is really nice. Uh And uh I this year I might also to speak and to present our deliverables in uh uh in the fields that I am working. I have been working here with this field for 10 years. Yes, from probably from the beginning of uh when the items and that started uh uh became not just the question of uh of the state for some special programs, but when it started, started uh started to play an important role in the media market as well. Um And uh they um I will speak about the uh the topic about A I translation in Dublin with emotional accuracy. Uh why people do not like uh A I translation in Dublin. Now, there are two reasons, there are some assumptions that it can be not accurate but it's getting better and better. And probably in the majority of cases, a young translation even uh better than human translation when we speak about live translation because uh there's no uh human uh human can uh can't work good during one hour or two hours. But A I can and humankind can't understand uh uh all topics like military medicine, media education, everything is a good, deep, deep understanding. But the I can and the I can uh can uh can be better and better year by year or even month by month. You can, you can see this. Uh but uh the majority of cases and we regularly try to, to make a researchers trying to make a pos and now people more and more say that not because of a, not like a I but because of a lot of emotions, this is not so cap but uh in the majority of cases, people watch because of uh uh of their mood to watch it now because of they want to get emotions. Uh uh because uh they actually, they, they know all information, they, they watched this film before, but uh they want to watch it again and get emotions but not just information. And uh we understand that this is challenge number one, probably because they are sure that the big guys like uh Google Open A I and Microsoft to people and other, other companies they will create uh and they will continue to create uh A IA I accuracy but uh translation accuracy better and better. But there are some uh fields that not so many companies uh has uh their focus. And uh the I uh accuracy of translate accuracy of emotions and transfer of emotions is uh one of the challenges um people who use the I usually uh why we use the I complex, we use the I usually uh why they use the I to translate it faster, translate it cheaper. And in in some cases, also with the translation quality, for example, uh we are not only uh comments that translate videos, voice to very recorded videos. Recently, we launched a live translation on Google Meet and Zoom. It's very new technology just we are mostly presented the Google just this summer. We are Google technical partners since 2017. And uh so uh in the majority of cases when we speak about live translations, it's not only convenient like to invite any time uh 24 hours seven to any, any language, but it's also like more accurate because uh nobody can guarantee that uh this guy can uh translate it familiar with all topics. But the A I, as I said, uh already drained quite well in many topics. And uh so this uh this, these are three requirements but why people use uh A I more and more. Uh this is uh like uh my short uh assumption like this short uh uh model uh of the development of uh uh A I uh before 2010, mostly a state financed in A I uh conservation technologies. And uh so like uh research groups uh mostly get the task, some materials sent these papers to the government and it was all, it was not fast. And the, the, the most important that when uh private money uh not in the process, this is not about marketing, this is not about captivating technologies, this is not about uh uh sexy feelings of life. This is just mechanical understanding what this guy said. Uh And to, to write reports about uh this uh this topic. Uh 2020 1020 FFFF F-15. Also uh uh the technology get became better and better. And uh uh it, it was a part time of a serious improved languages for processes and uh later uh 2015 and 2021. Uh This is a time of emergence of uh deep learning. Uh Of course, there are we focus on the situation in the United States because if we analyze other countries, these dates can be earlier or later, there are different scenarios. And by 2021 the results of private money were short and the church t of an A I project appeared and everybody uh decided, oh, we should use A I and it was the beginning of the new era. And uh everybody started to use this for many, many cases and they understand that it should be also good tool for marketing for uh for conversation when you want to uh uh not only on some uh social uh social uh understanding, social uh respect and some other else. And uh uh first of all, what is the emotional accuracy, what does it mean? Uh this is our definition of our group that emotional accuracy refer to the face for reproduction of emotional thoughts, intent and nuances. Uh not only from the text, this is very important, not only from the product of uh voice to text recognition, but also from the screen. So because different nations express uh the same as different way and uh of course, from the tone of uh of this uh speech and uh from the text itself, words that they say, and the only case we analyze all this, all the three parameters. At least we can say that this is exactly emotion uh uh that we need to uh reproduce and make, make accurate synthesis of these emotions in the new language. Uh End up. Uh Yeah, I would like to say that emotions uh is an internal part of our behavior of our everyday life. And uh we, we do not say anything without emotions. This is so rare situation when people say and do not say and not express any emotions even if uh this is not emotional speech. Uh in the majority of cases, uh it means that it should be not emotional uh speech without any angriness, sadness or happiness. But it also presents that in this case, emotional, emotional like this but not others. And what are the challenges of uh uh of the emotions? Of course, this was one of the biggest challenge that different nations express the same, similar emotions absolutely different way. And in this case, if we decide to, to make research uh uh in this uh uh field uh car research, deep research, we should analyze this cultural nuances and sometimes they are completely different. Later, I will explain, I will, I will provide you some examples. You can understand that sometimes they are absolute, absolutely opposite. Um A I technology advancement in emotional transfer. Uh First of all, uh I would like to say that uh uh if you speak about uh uh emotional uh emotions, uh it's necessary to uh emotional translation. Uh It's necessary to um uh a company and make, create something. Of course, it's a mixture of different, different technologies. It's impossible to make, to create now. Uh Just from this great uh full cycle of uh uh uh recognition uh trans of translation synthesis and uh and the voice synthesis and good timing of the video and an area of this stage. There are so many risks and that it's possible to be translated, not decorate. And in the majority of cases, we use technologies of big companies like Microsoft Open the Google and but to guarantee good accuracy at every of this accuracy, emotional accuracy or text or text conservation accuracy or anything, it's very important to uh to add something to for best, best satisfaction of our uh clients. Uh And what are the key features? Uh uh First of all, um uh that uh A I models uh create on extensive data sets. Uh uh And then the second issue that I already said that there are so many cultural adaptions and uh your algorithms should be uh should help like additional parameters of uh not only text but also additional parameters of uh uh adjustment to emotional uh based on cultural behaviors, cultural norms and cultural expressions. And uh uh also uh when we speak about real time processing, uh high accuracy is also uh there are, there are so many, many uh challenges of uh uh processing because you don't have time to correct them. And moreover, you can conversations here about latency with people in the market. And unfortunately, LA is quite, quite long period here that this is also one of the challenges that we should work to guarantee the best success and find the best satisfaction of our clients. Uh We uh uh when we work with different uh uh companies, uh we also try to make researchers and try to identify the reaction of uh uh of, of the clients of the audiences. And we usually make like a p to understand if we translate with emotions. What are the results? What, what if we translate without emotions? What are the results? And for example, this is one of our clients who make that uh foreign film in English and German language. And this is one of the feature films that we translated uh not documentary, not educational content where emotions didn't, didn't play an important role, but this is a feature film with emotions. This is, it was a film of uh product that looks like uh Instagram and uh guys presented this product to investors at the station. Nobody believed that uh Instagram can be uh can be really a big project. And of course, it, it contains a lot of emotions. And we made this film and uh the measure. And after our research, the emotion of fidelity achieved 96.8% of accuracy. And this is quite a good Rachel. Of course, VRP just developed our technology. And so in some cases, people are not satisfied with it when we transmit it into new measure. But 90 96.8% if you think about emotions, it's very important. And for example of the uh general uh uh like evaluation, if emotions are good or not, if we, and if we translate without the emotional touch, uh uh it's just 29% the so big difference uh at 29%. And probably you, you always uh you, you watched many videos that translated with A I and uh uh it didn't reflect any, any emotions not possible to understand. Sometimes if your actor uh if the character said I love you with inspiration or I love you with sarcasm, the same words, but absolutely different opposite meaning. And so the client satisfaction, client understanding can be also uh if you with big gap between satisfaction with emotions and without emotions. Uh Another uh case study was organizations in advertisement, advertisement uh market. Uh I will go faster uh in time. Uh And uh in general advertisement is a market where you need to uh create a new uh uh new video with new text, with the new ideas, how to capture new market. Uh but there are some cases where, where A I can help us with enterprising. For example, in this video of not about uh and not with emotional touch, like what's quick, let's drink beer. And uh we need to explain to different nations different meanings why we need to drink this beer. But sometimes uh uh uh A I can help with the localization of a retirement that is more like manual, for example, how to assemble expo. This is not so important. Uh Not so many cultural aspect, just do this transition, you assemble it or uh for example, why this pro uh how this project was produced for the best materials, the best process processing lines, the best quality control. So it's like advertisement but more uh more information uh for with, with focus on information than options. And in this case, uh uh even in this case, advertisement need emotions. And uh we had some research and when we made the uh cultural, some cultural adaptation with A I. Uh And uh also uh when we made the uh accurate uh transfer of the of the functions uh localized thats uh the results of uh satisfaction uh uh people who was better and people watch the uh advertisement uh 31% more. And uh uh and the detention rate uh uh the detention rate and people uh buying a product, it was 20 by 22% uh gaming industry. This is very interesting uh market for translation uh with A I because usually gamers are ready for all A I uh tools, the only one, the only two things this is played in you. Again, if you speak about uh what localization of fast games is really a challenge. And again, the challenge of emotions because in the majority of cases, people play games because to get emotions but not to get new information from these games. And uh in this case, cultural adaptation and emotional uh dialogue uh and uh different uh uh marketing nuances that are so important. And uh when we made such work, uh uh uh it was quite good results. Uh uh 20 25% of uh uh uh customer satisfaction was 25% more. Um What are the challenges in emotional translation? I already explained some of them. So I will be very fast with this. But in general, uh there are technical limitation, uh models are still not trained, very good uh with uh all these cultural nuances. And it's, it takes time to train better and better because uh this is absolutely new parameter. And uh if you work with these models, you should start from the, from the beginning very often. There are no resources that are already drained and years by years, it will be more accurate and more accurate and uh uh technical issues. Uh uh There are so many uh ways of uh misinterpretations uh uh different manipulations uh uh in different sensitive context. And this sensitive consequence can be understand, understandable for some nations and not understandable for another or other way of understanding. So there are many uh uh against so many of these cultural nuances. And this is again, absolutely new parameter that you need to train and create this new database. And uh only in case you do it long time, it will be really, really good uh uh result. And then uh and again, uh ethical guidelines to mitigate these, these challenges. And about culture, we are nuan nuances. I promise to uh to, to this, to uh provide you some interesting uh uh inter inter interesting cases that we uh we noticed when we translate from one language to another, uh we understood that uh uh in many cultures with the same emotions, the same um express a different way or absolutely opposite way. For example, happiness in Japan and happiness in Brazil, this is like two Pons uh Brazil, Brazilian people express their self like uh the way that uh uh with a lot of emotions, with uh loud speech, with humor, with uh uh with all emotions that is possible and the same sadness. If they are said with something, there are so many uh so many emotional touches. But if you translate it into Japan or from Japan, it's really, it's very, very difficult to detect this emotion somehow. Because for Japan culture, like to express emotional sadness or happiness such way, it's not, it's not, not supported by the society in case you are more like in moderation or it's you, you could be more respectful uh in this society. And sometimes it's really a challenge to understand what, what is the, what is uh what are the feelings of this character. And moreover, it's necessary to train this model, first of all, to understand this from anthropological uh uh from the view of anthropology, from the new of uh uh from the view of historical aspects, from the view of you that this country has because there is emotions also different year by year, sometimes even when some new movie appeared in the uh and it becomes very popular. Uh people start to cope with these emotions from this movie and only people who understand this culture understands this film. Uh they uh they understand what this is about or political, different political situation because nobody, when you said one year ago that people need cats or dogs, you said other things, but now understand what it is about. And so it's necessary to translate it and with the way that other people can understand of other leaders for them about uh talk uh in some non democratic, non democratic states uh then, but the general Western culture, there are some generalization but of course, it's not possible to train the model very fast. And there are some generalization, for example, uh it's necessary to, to train the model that for for Western countries, expression of emotions is the key for Eastern countries. It's like an expression of feelings. When you express emotions of weakness, sorry, expression of weakness, that you uh statement of weakness. If you express emotions and sometimes and also when you translate videos, it's translation, not only about voice, sometimes there are challenges when different gestures plays an important role. For example, our translation, our cases of translation from Bulgaria. If Bulgarian men or women say, do not say anything but make this. It may, it, it means yes and like this, it means no. And so you need to train this model and understand this absolutely different meanings or some African pronunciations for some African languages sounds like uh this is uh like a normal uh uh this is not like a special of some emotions. This is like uh like like we say ABC D and uh uh like uh some, some letters sounds like this in some African languages. And you, you should understand that this is not expression of emotions. If you say it means that some sound words really contains this kind of uh new uh sounds that never uh use uh European languages for example. And sometimes if you do like this in, for example, if you communicate with Turkish guys and they say it means sometimes no money or what are, what are you telling me about? Some uh it means some feelings but for Africans. No. And so many interesting things. If you, if you dig deeper um about the ethical consideration and biases, of course, there are so many superstitions, so many uh different situations when you need to, to, uh analyze really carefully, what, uh what does it mean in the, in new, uh uh in new culture? And very often they think about other culture. Absolutely different way that it's, uh, it's, in fact, uh, it is uh and uh it should be, of course, it should be trained with the people who live, who lived in this country as an internal part of this uh culture. Um And then other, and since uh OK, I will, I will go faster because I don't have so many, so much time uh future future development. Uh uh First of all, it's nece necessary to train, as I said, train to uh models by that and with, to enhance the generality of emotions that across more and more languages, but not generalization. Like I, like I showed you now, of course, it's necessary to work more and more on cultural adaptation and to improve the ability of emotion compared to various cultural content. And, and uh it's necessary to improve the synthesis of the voices in every language to reflect uh emotions more uh precisely and precisely. And uh uh and uh finally, what is the aim of our work? Why we work uh uh to translate with emotions, the main, the aim is translate, not like people, but to translate better than people. In this case, uh A I technology will achieve the main uh the main uh target and make the aim will be achieved to, to uh to make, to become more captivating. And if you work with, to create this emotional technology, it's very important to have in your group. Also science like psychologists and other people who can understand why people watch this video, what emotions they have and how to make it better with the A I technologies than any actor can, can make to achieve the results of and to engage the audience. Um OK, let's go further and the conclusion I uh recap uh of uh uh A I advancement and emotional accuracy and cultural adaptation. Uh uh And the future outlook emphasis of uh on the ongoing need for innovation in A I translation in D and, and uh uh of course, uh uh uh if you work in the I translation and Dubin industry, there is only one way how to satisfy clients better and how to develop your project. It's uh uh it's inevitable question. It's, it's inevitable development that you need to work with emotions and uh to develop better emotional models. Thank you for uh your understanding. So if we can have time for questions, uh please, you can ask me uh in my presentation, I tried uh I tried not to uh uh not to use any names of brands uh movies because we have some uh nd A with some partners. But if uh I can, I can uh answer your questions. And uh also I decided not to present any work today uh with uh emotional translation. But in case you would not not to make any advertisement of any, any friends or any uh videos or any films. And in case you want to and most I, I made, I made generalization just to uh to explain you about the industry development in this field. And what are the challenges? What are the best scenarios? What are the scenarios of the future development but not to focus on any fields? But in case you, we want to watch some of the river of some transformation with emotion. You can, you can reach me and I will be glad to send you or schedule a call and show you how it works. Thank you.