A New Era of Language Tech with Britta Aagaard and Jochen Hummel

Season 4 Episode 1: A New Era in Language Tech with Britta Aagaard and Jochen Hummel *Lo-fi funk music begins to play (NEXT Season 4 Theme Music* Zelena Khan: Welcome to NEXT brought to you by ContentLabs. I'm Zelena Khan your host, and I’m so excited to bring you Season 4 with a brand-new set of guests to educate, motivate, and hopefully inspire you in and out of the workplace. This season, we are talking to industry experts from localization to artificial intelligence, to the importance of perseverance. Let’s dive right in. *Lo-fi funk music continues playing* Zelena Khan: I've had many lives before coming to TransPerfect. I worked in the music industry, tech startups, and I didn't know anything about the language industry. After 9 years at TransPerfect, I can honestly say I'm still learning something new every day. Combined with my love of tech, I've been able to watch this industry evolve, and our next guests have been pivotal in that. We're here with industry veterans, Britta Aagaard, CBO of Semantix, and Jochen Hummel, CEO of Coreon and ESTeam. We touch upon everything from machine translation, CAT tools, global business, and the future of language. You're in for a great episode. I know I was excited to have this conversation. I'm so honored to have you both on today. Our past shows have touched on language in different industries, but none of our episodes have fully focused on the translation space. And I mean, working at TransPerfect for many years, there is so much I don't know. So I am personally excited about this conversation. If you're in the industry, you're in for a treat to listen to these two experts, and if you come with a curious mind, we hope to share some definitions and explanations along the way. I think we'll just jump right in and introduce our first guest, Britta. Britta hello! Britta Aagaard: Hello. Happy to be here. Zelena Khan: Oh, we're honored to have you. Can you tell the listeners a little bit about who you are? Britta Aagaard: Yeah, so I am currently heading Semantix. Semantix is a language service and solution provider in the Nordics, basically the biggest in the Nordic region. And we have joined the TransPerfect family of companies a little a year ago. And we are now trying, of course, to on that great platform to show what Semantix is all about and at least grow, continue to grow our business. And what is mostly exciting about being in this new context is there is so much technology-driven, which is also the, I would say the fundamentals of what TransPerfect is doing today. So that is really exciting. Apart from that now we are saying we are moving into the translation space. My background is a little bit weird. I come from the scientific side from academia. I've been studying and lecturing philosophy for many years before I joined this industry. However, language has always been the core of it. The interest has always been beyond translation because language is so much more than what we can do only with translation. So I think that is very much what we would touch upon today. Zelena Khan: Yeah. Jochen welcome. Jochen Hummel: Nice to be here. Thank you. Zelena Khan: Can you tell the listeners a little bit about yourself? Jochen Hummel: Yeah, sure. I'm currently CEO of two companies ESTeam and Coreon. Coreon deals with multilingual knowledge management, ESTeam is a language technology company. And I'm a serial IT entrepreneur doing this already for a few decades. Most people in the language industry know me as the founder of Trados, I found Trados. I'm also a software developer by background. So I developed the flagship product and established translation memory as the standard way of doing translations and sold the business about almost 20 years ago now. And I live in Berlin, have three kids, and am very happy to be here and share my thoughts about where the translation industry is going because a lot has changed since Trados. But not really in the daily work of translators there, I think things have been a bit stable, a bit too stable. Try to talk about here in this session about what the future has for us. Zelena Khan: So, how I ended up at TransPerfect was just very . . . didn't think I was going to be here. It's opened my eyes to so many things, even nine years later. How did each of you get involved in the language industry? Britta Aagaard: I mean, I sometimes I think that I kind of ended here and many people say they ended up here for, for some reason. But when I think back there's a clear red threat. I have always been fascinated by language as the means for humans to understand. So I mean, for me, language is inherently human in the sense of that, we cannot understand anything in the world without having a concept of it. And we can also not communicate. We cannot connect with anybody else without language essentially. And this is as, as far as I can think back has been my passion in some way. And for our business for our customers is how language is really creating meaning. And as we talked about also knowledge for that as I said, we are not only in the translation or localization business, but we are actually enabling a business in a way that it's a basis for decision making its growth, its prosperity for our public sectors. It's, I mean, it's a basis of democracy. I mean, I think, yeah, very, very big words, but I think essentially that is a driver that has fascinated me in throughout all those years. Jochen Hummel: I grew up in Stuttgart and that's where IBM had its German headquarter. IBM was one of the first companies, if not the first company localizing their products, products on a big scale. And they were looking for freelancers who had, let's say some basic English knowledge and could deal with software. And I was not a computer kid, but let's say a computer teen or tween. Spent some time in the U.S., so I thought, yeah, I can do that. And that's, IBM was, at that time, a bit like Xerox, very innovative. They had all this cool stuff like XML and laser printers and screens, whatever, they had all the cool stuff. And, they were using the grandfather I would say of translation memory systems, but they always had it in a way that it was way too expensive too complex, I mean, not really marketable. And I saw this and I understood the concept and said, "Hmm, with the PC now and with the development tools available, why don't I develop such a tool in a way that it actually can be used by freelance translators?" And that's how Trados started. Zelena Khan: The rest is just history. I think before we get into any of the topics, how do you two know each other? Britta Aagaard: Yeah, I think, I don't know, Jochen if you have the same experience, but my first experience is that we actually met at a conference an industry conference where I first heard Jochen give actually a speech about the language industry and he already moved on and moved beyond translation, so to speak after selling Trados. And then we started working together because there's something that has always stuck with me from the first time basically Jochen that I spoke to you is that you had this, that what you're doing now it's more for analysts, you said. And, what you were interested in is to elevate strings to meaning, so that we go from language or from a word to actually meaning in context. Britta Aagaard: And this is something that Jochen is working with today. And this has stuck with me from the very beginning. And then I mean, this industry is rather small. So then at some point, Jochen actually contacted me for an opportunity to work together on a very large contract for a European institution where the task was to both have the software part, but also the human part. And even if we were . . . I think we punched up our weight at that time to pull this off, but we managed to win this very big contract. And then we have been working together ever since. Jochen Hummel: So we've worked now how long, almost 8 years, very closely together. And one thing I understood a bit too late in my career, I should have done this or did this much earlier is that I was always a tech guy and said, "I am selling technology and services done by basic customers." Zelena Khan: It's great that you both have been able to collaborate and work on multiple contracts and projects together. What are some of the things your clients look for or ask for when they come to both of you for solutions? Jochen Hummel: Selling technology is getting more and more difficult because technology moves fast, it's complex, and many companies have a hard time buying technology because to suspect something out already requires certain know how which for many companies and customers is hard to get. So what customers actually are looking for is a solution where services is always a part of. And so I'm very glad to have Britta and her organization, a partner where we can, as a consultant, working together, provide a full solution and play, let's say our technical, but also organizational strengths and for providing complete solutions end to end solutions for our customers. Zelena Khan: So, Britta, you brought up Semantix. Jochen, you brought up Coreon. How has Semantix translation and interpretation services intertwined with Coreon? Britta Aagaard: I think as we just mentioned, we are working on, I would say contracts, but also projects together. And then what we are working very closely on is to somehow also work on, let's say, new business models or delivery models as such. So that is, that is the basis of what we are doing together. So, it's not only servicing a specific customer need but also always work on the improvement and the innovation of what we can do together because this is what we clearly see also for the future of our industry, that we, we want to go beyond that. There's something that we can unlock in the way that we work together. Zelena Khan: I know CAT tools play a large role in machine translation and other translation technologies have emerged from this. But for our listeners who are not that familiar with the language industry, can you please give a definition of what CAT tools are? Jochen Hummel: Stands for computer-assisted translation and they're CAT tools. It's a product category, which I think I helped to define and establish. And CAT tools computer-assisted translation is positioned against machine translation. So on machine translation, a machine translates. So English text in, German text out. In computer-assisted translation, there's a human who is translating but is assisted by a piece of software. And in most cases, this is done through, by using a translation memory and a translation memory is a database of sentences, which already have been translated. So English sentence, German sentence, English sentence German sentence. And whenever there is a sentence which already has occurred before, which of course with documentation happens very often then that sentence is translated automatically. And if not so maybe then there's a similar sentence where just the word has been changed, a so-called fuzzy match. Well, then that's retrieved. Jochen Hummel: And then it's much easier, of course, to change than this word, this one word, then doing it from scratch. And that approach computer-assisted translation working with a translation memory plus in the supportive role, a terminology base, that has been the model for the last 25 years, Trados being the first product, and the products who came later more or less followed that model. So that model has never been really challenged. And that model also created the business model of the industry. So the industry works with the unit word as the main unit. And then you have a certain price for word, which is a hundred percent match or a fuzzy match or a no match. So that's the currency of the industry, the price metrics how the whole industry has worked. Zelena Khan: So you mentioned this model's been used for 25 years. A lot has changed since then. Jochen Hummel: Computers are many magnitudes of orders, more powerful, and certain advances have been made in AI and other areas. So that I think it's about time to rethink this whole model to rethink that process because translation memory has also have introduced quite a few problems, quite a few challenges in terms of data management in order to keep them clean. Also, these segments they're taken out of context. So even a 100% match needs to be reviewed. Text changes, terminology changes. These changes of course are then not reflected in the translation memory. So, there are quite a few issues which translation memory has introduced to the process. And at the same time machine translation has become better and better and has seen such a jump in quality that by now it's about time, and I've been saying this already for a couple years to rethink these processes and maybe CAT has seen its finish and should be sunsetted and replaced by a different way of doing things. Zelena Khan: Ooh. Jochen Hummel: *Laugh* Oh yeah. Zelena Khan: So as a big believer in technology driving this industry, do you think translation will eventually solely depend on machines or do you see the need for always some sort of human involvement? Jochen Hummel: What I'm talking about here is in my opinion, kind of a disruption and I don't like that word very much because it is pretty aggressive and hyped and overused. Machine translation claims to have achieved human parity. And most people, of course, would challenge that and say, “No, no, no how can a machine be as good as a human,” but they asked the wrong question. They define human parity as being as good as a human, which a machine in most cases still is not. But the thing is the reality in production is that when you outsource translations over a long supply chain you only get a certain quality also quality you are willing to wait for or to pay for, right? So the question is not whether a human can be as good or machine translation as good as a human. The question is more whether it's, as let's say, mediocre as what you typically get when you outsource it to a supply chain. And if you answer that question with, yes, if you define human parity like this, well then, everything changes. And that's the disruptive in this technology. It's not only a tech and the workflow roles would change also business models will change. And that's the hardest and the most let's say, what has the biggest effect because implementing new technologies is rather easy for companies changing business models most companies fail to do. Britta Aagaard: Yeah. And I want to add to that I mean, also the complexity has increased even further because also content types have changed a lot. We see very different types of contents and channels today. And with that also, I would say the definition of quality where we might have a definition of human quality that one translator translates one word into another, but today businesses have many other requirements. Sometimes it's time to market that where then maybe machine translation would actually fit that purpose. And other times it could be that it should actually not be a literal translation, but it should be maybe more creative tech. So then it's all about the customer experience and making sure that the brand voice is preserved. And all of these let's say more new access and complexities are actually not able to be produced in that very stiff CAT tool that Jochen once created. *Britta and Jochen laugh* Zelena Khan: There's a misconception that NMT is augmented. For those that don't know NMT means neural machine translation. The way it works is that the program's neural network encodes and decodes source text. Very heavy. Can you both explain how language service providers are pitching NMT to their clients? Jochen Hummel: Yeah. So this concept of augmented translation, that's what the industry is putting forward. And there's a nice graphic which puts the linguist literally, figuratively in the center. And then the linguist taps into terminology databases into a translation memory, into machine translation, into enriched content, into project management. And then there's this linguist in the center who uses all these data sources to be a more efficient and better knowledge worker. And I think this picture has many problems. And the main problem is some of these strategies, it's all about recycling. Also, machine translation is about recycling previously made translations. They overlap, they compete with each other and also the clients. So, the software interface, the linguists is then using to tap into all these sources becomes very, very complex. If you look at a modern CAT tool, it looks like a developer studio from a software developer, you need a huge screen, you have all kinds of windows open. So very complex. Jochen Hummel: So what do you think how many people in the world can handle that kind of complexity? A couple 10,000 translators, but that's about it. They're very productive, but they do this all day long. The recycling strategies are overlapping and you are very limited because there are only so many linguists who can deal with this complexity. We believe that it would be much smarter to actually not put the linguist into the center, but take the linguist out of, at least out of doing words. And instead have a workflow where you first recycle content, not with the translation memory, but by recycling larger chunks of content, which are in context and therefore safe recycling, and can be locked. The rest you machine translate. Then you try to measure the quality. And if the quality is, as Britta mentioned for certain types of texts is fit for purpose, you're done. Jochen Hummel: No linguist required. You're done. If not, well, then there will be certain segments which do not match this criteria. Well, then what you want to do is to look at this and not at everything, and to look at this, but then also have the system telling you why you should look at this. Maybe there's a wrong term, or maybe a named entity has missed, or maybe something just looks fishy because the source sentence was 20 words, the starting sentence is five words, whatever. But then you are instead of going through the whole text and have all these windows, instead, you are like reviewing a contract, a red-line contract, you jump from issue to issue. System tells you why, and then you're done. Now, of course, the question, when I tell this in a conference, people always stare at me and say, "Well, what does it mean? Jochen Hummel: Am I out of job now? Do you tell me I can go home?" Well, no, because in order to do all this I mean, AI needs to be trained. Workflows need to be built. Multilingual knowledge needs to be created. Data and repositories need to be curated. So there are new roles for linguists and the nice thing is these roles require much more knowledge and have a multiplier because you can only do so many words. Even with Trados you can only do so many words a day. But if you set up a system to work for you and process words, then you will multiply your knowledge. You train your knowledge, you keep your knowledge, you put it into a system, which is, I would say arguably much more fun, but also much more productive. So, there are new roles for linguists, which are, let's say much more interesting, have a much greater, much higher value and therefore also should be much better paid. Britta Aagaard: And to add to that again, from a quality perspective, we believe that this is actually the way to ensure and even enhance quality because it is all about also feedback loops in this and measuring throughout. So what machine translation in many instances still is, and also new machine translation, it's still a black box. If you just put everything into that, you don't know really what's coming out of it. But if you have a process like we describe, then you will have a constant feedback loop and also understand what the quality is and where the issues are. Zelena Khan: You brought up products and I immediately thought of the retail industry and, you know, the blow it took in the last couple of years. How has the pandemic impacted your industries? Jochen Hummel: Good question. I mean, of course, it was driving eCommerce for a while. Zelena Khan: Mm-hmm, yeah. Jochen Hummel: So I've just a while ago seen a presentation where somebody said that the pandemic actually accelerated eCommerce by five years or so. And oh, I thought that sounds cool, but recently I read it's kind of falling back into its old path, so advanced five years and back three years. So, I don't know whether the pandemic really had a big or lasting effect on these kinds of use cases. Britta Aagaard: Yeah, I agree. I mean, in general, in our industry, we say that I mean, we have been lucky. I also want to highlight as Jochen also stressed is that I think the biggest change is more from an almost societal level that digitalization per se is something that nobody can deny anymore. And of course in some markets more than others. So while I'm in the Nordics, we have a very high degree of digitalization in the whole society. Whereas Jochen in Germany that lacks a bit. *Jochen laughs* And that *Britta laughs*, that I believe is for the good of everyone because it requires new thinking. And I think that we will see businesses need to rethink their delivery models as well. Zelena Khan: Jochen you mentioned linguists, you know, possibly getting overwhelmed and now they're no longer at the center of you know, the process. Can we, and, you know, we threw around quality and quantity. Is there a world where we're going to have both now with these new advancements? Jochen Hummel: Yes, of course. I mean, when we talk about these new things, new ways of doing things, I mean, mind you, there are companies who are introducing translation memory today 25 or 30 years after it has been established. So there's this talk about the future is already here, but it's very unevenly distributed, and of course, that will continue like that. Britta Aagaard: And we can see that already also, and especially, and I think that is extremely exciting for us and for the industry as such that, I mean, we now also in our context, we speak a lot about AI. We also speak about data and we speak about the data that is language data, basically. So here is really, as Jochen just said, here is an opportunity to position the value that is created through language in a completely new way from being a service provider that is paid by the word. And always as little as possible we can now with combining AI and machine learning in a more innovative way, we can actually move on to a strategic level where this becomes a data point for decision making in businesses. And I think for us as an industry, that is probably the most exciting perspective we have had for decades. Zelena Khan: Yeah. I think artificial intelligence and its impact is amazing. Are there other ways that you feel like artificial intelligence will play a role in localization in the language industry? Jochen Hummel: Well, I would rather phrase the question differently. I would rather phrase the question, like how will the data, which we generate, I always talk about linguistic assets. So the linguistic assets we are generating, how will they, can be, they be deployed and how will they drive other use cases than localization? Because mind you the localization industry they're still following a model. I call it the broadcast model where they think there is a content owner, a brand owner, and they want to talk to their customers. And of course, these customers speak different languages. And so we are helping them to talk to their global customer base. So, take the tech line of RWS, take global content and ideas further, or TransPerfect to deliver our clients' messages all over the globe; we localize, transform your content and data to capture business worldwide. Jochen Hummel: It's always the idea that we are helping clients to deliver their message, their content, to broadcast it globally. But I think that that model is outdated because what do we do as a customer or as a citizen in the 21st century? Actually, what we do is we are talking back and we are supposed to talk back. We're invited to talk back. So we're not supposed to call somebody or get an agent if we have a problem. No, we get a form that we type in our support problem, or we have to talk to a chatbot. And if we don't get the problem resolved, then we go on social media and complain and whine. Zelena Khan: Yes, we do. *Zelena laughs* Jochen Hummel: So we are talking back. So it's not about broadcasting content anymore. It's about communication and communication goes both ways. And mind you, if a company has a corporate language in most cases, English, then they keep everything in English. And only on the fringes of the interface to the customers they translate, which is often even done by a subsidiary or by a trading partner, reseller or so. But when you communicate well, then it happens in all languages. It doesn't help that your business, your company language is English, and the one who mastered that process better. So, the one who has a better communication with its customer base, who understands also competition and the market, local markets better, that company will win. Britta Aagaard: And another example on that is not only are the customers talking back, but they are also searching for your products or your services in their own language. Zelena Khan: Yes. Britta Aagaard: And if they don't find that if it's just a string that you can search for and hope to be lucky that you will be found, no, they search in their particular way in their language. And also in their cultural context. Jochen Hummel: That's a very good point, Britta. And when you are a customer in a huge market, like the American market, of course, you don't feel it, but now imagine you live in the Czech Republic or in Malta, or Estonia, and you search for a product, and you enter the search string and check. Well immediately, you’re caught not only in a language silo but in a local market. There might be a company in Germany selling the same product or other products at a cheaper price or a better product you will never find it. So even a company like Amazon hasn't figured that one out, as soon as you enter the string in a certain language, you are restricted to the offerings which exist in this country. So simple, innocent product search already is basically dysfunctional even on the biggest website. So a very good example, of how language limits your possibilities to find the best products at the best price. Zelena Khan: I've really enjoyed listening to your opinions. What's your hope for the language industry? I mean, the language industry and linguists, if you want to go ahead and get into that too. Jochen Hummel: I believe AI will change everything. It's a total game changer, and everybody would agree and AI, how does AI work? Well by being trained with human data and half of that data is textual and textual data is always multilingual. And where do you find this high-quality data, and where do you find people who can deal with that data and where do you find people who can test this data and assess the quality of what the AI produces? Well, these are the linguists and this data is created in translation, localization, and documentation. So if linguists are smart enough to understand all this, and that's not only true for linguists, it's also true for LSPs, for the companies who are moving in that space. If they understand what kind of assets they're actually owning and what kind of workforces they're actually having and what they could contribute to, then I think they will move at the center of the AI wave. So, if people understand this in a smart and you know understand the multilingual dimension of it, then we will all move from the fringes into the core of the biggest revolution mankind probably has ever seen. Britta Aagaard: We touch about it throughout the conversation and also as Jochen just said that we have this unique opportunity to position language, not as an afterthought or something that is basically a cost driver but actually a revenue driver. That from a, let's say business point of view. But I also see that it will create even more opportunities for . . . you mentioned yourself, I mean diversity, inclusion, because, in the end, that is as I mentioned, for me, the essence also of democracy, only if you can speak only if you can be understood, you can also be part of that global community and the global society. And I think that we can, and we should position ourselves exactly in that intersection. Zelena Khan: I am so honored. This was amazing. I want to do, we have one more thing. We want to do a little bit of a word association. I'll say a word and then each of you can say the first word that comes to mind, right? And I think it'll be very juicy if you don't say the same word. So, we’ll go. The first word is language. Jochen Hummel: System. Britta Aagaard: Meaning Zelena Khan: Technology. Jochen Hummel: Innovation. Britta Aagaard: Fun. Zelena Khan: Linguists. *10-second pause* Your face Jochen I know you have something there! Zelena Khan: I'm not sure if you're frozen or you're really just thinking. Jochen Hummel: Pool. Britta Aagaard: Maybe I should say, hero. Zelena Khan: Translation. Jochen Hummel: Machine. Zelena Khan: It's got to be the first word, Britta. Britta Aagaard: I always say, don't talk about the T word *Britta, Jochen, and Zelena laugh*. Zelena Khan: I could take that as an answer. Britta Aagaard: Good. Zelena Khan: Clients. Jochen Hummel: Success. Britta Aagaard: Global. Zelena Khan: Localization. Jochen Hummel: Industry. Britta Aagaard: Workflow. Zelena Khan: The last one is knowledge. Jochen Hummel: System. Britta Aagaard: Meaning. Zelena Khan: This was fun. So if people would like to connect with you both, how can they do so? Britta Aagaard: Yeah, I mean, find us on our website, Twitter. Jochen Hummel: Yeah. And of course also LinkedIn, but also Twitter, and typically used by clear names. So very easy to find. Zelena Khan: Like I said, thank you so much. This was amazing. Britta Aagaard: Thank You. Jochen Hummel: Thank you. It was fun indeed. Zelena Khan: Language is knowledge. It's at the core of everything we do. I mean, we wouldn't be able to connect with one another without it. The work is meaningful. We're able to deliver content to clients all over the world, in every industry possible. With technology as a disruptor, the industry's evolving and it's going to continue to evolve. As customers' needs, change and grow. I'm excited to see where it's all heading. I encourage you to connect with Britta and Jochen on LinkedIn and Twitter to continue the conversation and stay up to date as I'm pretty sure these two will be a part of the changes to come. Before I go, I'll leave you with a quote by Rita Mae Brown: "Language is the roadmap of a culture. It tells you where its people come from and where they're going." Until next time. If you made it to the end of this episode, thank you for listening! Make sure to hit the like button and follow us wherever you get your podcasts. If you have a question, comment, or suggestion, we’d love to hear from you. Email us at [email protected], or visit our website at nextpodcast.transperfect.com. Support for this podcast comes from TransPerfect, a family of companies providing language services and technology solutions for global businesses.