#263 SlatorCon Recap, Cohere’s Big AI Translation Launch, TransPerfect Buys Unbabel

0:00Esther Bond:Command Translate, it's their enterprise AI translation offering. And you know, they have a lot of firepower that just raised half a billion dollars at a six or$7 billion valuation.

0:14Hey Esther, we're back from California. How are you?

0:17Florian Faes:Jet lagged. Jet lagged. Jet lagged, but good. Yes, very good. Had a great time. Brain not functioning. Brain functioning 70%. 70%. Must record podcast still. Must say words. Must say words. All right, let's start with words. Amazing turnout.

0:35Esther Bond:Let me rattle off a few stats from the conference. SlaterCon Silicon Valley, that is, where we all gathered last week, including much of the team. Yeah, as I said, amazing turnout. 200 plus participants actually there on the day. Very hard to pull off in the Bay Area, I can tell you that. 25 speakers and panelists. We had 12 panelist presentations and short speeches. So basically this was the biggest SlatorCon ever in terms of actually the agenda. We had 11 partners and exhibitors, so that means we sold out our exhibition space because we're not doing this in like a giant exhibition hall, but basically, you know, we have hotels that have an amazing main conference kind of space, but then also adjacent, we try to find a very good space for people to put up their booths and we managed to do that and sold out on it.

1:25Esther Bond:So very, very cool. We had lots of advanced technology people and products featured at the conference. Very strong localization buyer and user turnout as well as you'd expect in this part of the world, right? That's where a lot of localization of technology products is happening. And then of course, many of the key industry decision makers, many CEOs, many owners, everybody there. So pretty cool. I love the weather as always, super nice. California, Northern California, a little foggy in the morning, but then usually cleared up in the afternoon. Fog and sunny.

1:57Florian Faes:Yeah, we need to bring sunglasses and a bit of a jacket for the evenings.

2:02Esther Bond:For the evenings, yeah. Because we had this wonderful garden area. I don't know, what would you call that? It's...

2:07Florian Faes:Yeah, almost like a courtyard, but grassy. Yeah, you had the pool, it had the courtyard, you know, where people could network outside, which is kind of unusual as well for a conference that you can actually, you know, step outside, have some... Yeah, I always have high hopes for London whenever we go in May. I think there's been the odd year where we've had great weather in May for SlateCon London, but more often than not, it's not outside weather. Yeah, this one was great though. I mean, here in or there in Menlo Park. Definitely.

2:38Esther Bond:All right. So key takeaways. So we started the conference, gave our view on the industry and spoke about the language solutions integrator and language technology platform dichotomy we launched, right? It got some interesting pushback, got some interesting questions. We're going to cover this as well. But then we moved on to Brian Murphy, CEO of Smartling. He gave a very, very good presentation kind of moving from macro to specifics and very kind of driven by data as well. He also did put out a warning, like, don't try this at home, basically. Like, I like that. Don't try to do your language project at home, of course, as he should being the CEO of SmartLang.

3:24Esther Bond:A builder. Yeah, so there's a lot of hidden costs, lack of expertise, and some people are now coming back having tried to do all kinds of stuff with LLMs. And then I moderated a panel with AWS, Microsoft, and Google Cloud. That was a bit of an experiment because we invited people that don't have a lot of, I guess, touch points with specifically localization and translation, but they were very much in, I mean, they're in charge of certain parts of those, obviously, super foundational, very important platforms with AWS, Microsoft and Google Cloud. They kind of gave us a view on what you should and shouldn't try to build.

4:05Esther Bond:So essentially, somebody came up to me after the conference said his takeaway was, well, don't try and compete with us. And I partially agree, but I think it was obviously much more nuanced. They were also saying, well, you have your proprietary data, you have your defensible niche, you have your vertical that you really know. So, you know, maybe not try to compete with us on the foundation model level, but, you know, do obviously build on what we have to offer.

4:33Florian Faes:Yeah. I mean, that seems to be their message, I think, generally, isn't it? Those foundational platforms are there for people to build on top of. So, it's definitely there to kind of serve as a basis for a lot of what, you know, the startups and others are trying to build at the moment. There was one company and I'm trying to segue this in a meaningful way. So, just yesterday, Oracle had like a 25 % pop in their stock price and, you know, Larry Ellis and the founder got like$70 billion richer because they're also kind of part of this kind of infrastructure boom now. alongside the NVIDIAs and the Microsofts and the Googles, et cetera.

5:15And so, I was chatting to ChatGPT to just brush up my knowledge on Oracle.

5:21Esther Bond:I'm like, what does Oracle actually do? And then I asked it, so make this a little bit more specific in terms of the language industry and then it actually told me, it being ChatGPT, that one customer of Oracle is Cohere. So Cahir is running some of their models on Oracle. And Cahir gave, so I'm not, who knows, maybe Chachupiti hallucinated, but I didn't verify this, but it sounds very interesting. So we had a really fascinating presentation by Kelly Marquisio from Cahir. So she's the multilingual lead. And yeah, they had a great launch last week that we fortunately covered. Maybe you can tell us a bit more about that.

6:06Florian Faes:Yeah, I mean, obviously it was great timing, I think, for Kelly because she was saying, well, you know, now that we have launched Command Translate, now I can actually talk about it at the conference. And I think had it been a week earlier, she may not have been able to share very much. But yeah, she's talking obviously a lot about how the multilingual LLMs are built themselves, sharing obviously how multilingualism is pretty central to Cohear's overall process. So yeah, I mean, it was really good timing, I think, in terms of having had that launch just, what, like days before or something. Yeah, I mean, I can't imagine if it had been days after.

6:49Florian Faes:Think about it. She has to go on stage, talk to a room full of people from the localization translation language industry, and she has this thing in the back like, Well, we're going to launch our major... Maybe they brought the launch forward just to make sure that it was on time for SlateCon. And then, yeah. So I'm very glad that they were able to launch this. So yeah, Command Translate, it's called. It's their enterprise AI translation offering.

7:19Esther Bond:And they have a lot of firepower that just raised half a billion dollars at a six or seven billion dollar valuation. So yeah, it's a long way. They've come a long way since we had Nick Frost, their co-founder on the podcast. And like, I think it was fall of 2022. So 22.

7:37Florian Faes:Okay.

7:38Esther Bond:Just about when, I think it was literally weeks before ChatGPT launched or maybe on the day or something. Yeah.

7:45Florian Faes:So again, very timely.

7:47Esther Bond:Very, very timely. Yeah. Cahir seems to be the timing is good. I'd love to have him back in London in 2026. Great company. And yeah, Kelly, you know, phenomenal speaker. I mean, that presentation really well kind of managed between like technical aspects, but then also kind of took us lay people along for the ride. So very good. Same for the AI dubbing panel. We had RWS, Adapt, Global LipDub and Panjaya on there. And we'll cover this, I don't want to summarize it too much here, but it was just a very lively discussion, very interesting. Basically, that technology has also come a long, long, long way just in the past 12 to 18 months and basically, I guess one of the key messages is there's just so much more that you can do now that you just weren't able to do 12 months ago.

8:41Esther Bond:So you're kind of really expanding the market and you're not like taking away necessarily just yet, right? So rapid adoption there. Yeah, one of my favorite was Scott's presentation with Scott Sukhoff from UC Health. For those of you who don't know, UC Health is the University of Colorado Health. It's a not-for-profit healthcare system headquartered in Aurora, Colorado in the United States and he really kind of broke up the whole AI discussion in a sense because we had a lot of AI, a lot of tech and then you have this huge healthcare system that obviously needs lots of language services and language technology.

9:25And he really highlighted the human involvement. I'm never going to forget that slide where I think there was a sign interpreter lying on her back and sign interpreting to somebody who had a medical procedure done. So yeah, that's frontier sign language interpreting.

9:46Florian Faes:Exactly. I think it's things that you don't really think about until, you know, unless you're in that specific situation yourself as somebody, you know, who doesn't require a sign language interpreter. I could not imagine having a sign language interpreter just sort of lying on the floor or any interpreter, you know, lying on the floor as part of the job. But I guess the point was the person who required the interpreting was lying sort of belly down or stomach down on a table, on like an operating table or something or an examination table or something. and so therefore yeah that was the only way that or that was the best way for the person to be able to to see so yeah i really liked that i think there are a lot of kind of slides um or like images and photos and sort of human life real life stories um showing like about some of the impact that of the work that they do on the communities the interpreting operations that scott runs is like i don't know it's like a mid maybe small to mid-sized european country would run that.

10:45It's just huge. I mean, 150 languages a year, almost a million minutes

10:50Esther Bond:a month in terms of remote interpretation, 6 ,000 in-person encounters per month. They actually run a fair bit of it through in-house resources. So they have like over 100 staff, nine languages. They have 1 ,300 interpreter iPads. You get the picture, right? So incredibly sophisticated and large operation. I keep saying this, but sometimes the European mind just doesn't comprehend those very large US hospitals, the scale of these US hospital systems. Yeah, it's country-sized. All right, then we had a panel moderated by Anna Windemar, head of research on AI agents in localization, kind of very frontier discussion there as the new strategic layer in localization that's appearing there with these AI agents.

11:40Esther Bond:They're making workflow and data decisions. It's quite early stage, but they're moving to more complex tasks now. So that was an interesting discussion as well. And then we had chief scientist, co-founder of Sync Labs, Sync, the lip sync company, Prajwal Renu Kanand, who I did not know was one of the leading authorities on just lip-syncing generally, right? And so he was able to quote from his own papers quite extensively, which was super interesting. That helps, yeah. Yeah. Gave us a kind of history and state of the art of AI lip-syncing, presentation, you know, challenges like adding expressive, multilingual speech, et cetera, and also shared a bunch of kind of historic demos, like how it was two years ago, how it was one year ago.

12:28Esther Bond:And then also, I think he showed something that they're still about to launch in a couple of weeks.

12:35Florian Faes:It was seeing the progression was nice, I think, going from whatever was the sort of state of the art kind of two years ago to current day. And that was good.

12:44Esther Bond:Correct, yeah. And then he had this paper, I think it's called A Lip Sync Expert is All You Need for Speech-to-Lip Generation in the Wild. And I think that was cited, at least on Google Scholar, it's cited like over a thousand times. And you see that it's being cited more and more. So it's a nice curve. It would have actually published in 2022. So it seems like this was really one of the foundational papers on this technology. So yeah, very, very interesting. And then you tell us a bit more about the investor panel that you closed the conference with.

13:15Florian Faes:That's right. Yeah, we ended the day. So I moderated a panel with investors from Bloomberg Capital and SK Telecom Ventures. So, I mean, it was a sort of, I think, always a fast paced discussion. But looking at sort of investing in the AI era, obviously kind of bringing it into the context of language AI. We talked about some of the defensible areas within language AI, some tips for founders and builders of technology. So yeah, some really great insights and sort of just good to see that perspective on everything that's happening in the language industry. And then we went to have a nice little networking session there, out there, some food.

14:01And then the next day, we got a quick tour of, that was not part of the SlaterCon experience. That was just for us. An add-on, yeah. Yeah, an add-on for us. We got a quick tour of the Meta campus, which is super interesting.

14:17Esther Bond:Not quite country size, but also huge. Yeah. I didn't know, I got to be honest, I didn't know it was so large that the kind of the office space was so large and just behind the hotel. I knew there were certain like Meta offices, but I didn't know it was the kind of the giant campus building.

14:34Florian Faes:I mean, you were talking about hard to comprehend, you know, my office kind of contains four desks and this is, I can't even imagine how many times more the Meta Campus would have been. That was a great tour. Took a couple of pictures and then yeah, headed back to Europe. You headed back to London where we're all going to be heading again in 2026. Specifically, let me plug this right now.

14:59Esther Bond:Save the date, Thursday, 21st May, Friday, 22nd May, 2026. We are back in London, so do save the date. This one we're going to sell out again, so you want to make sure you get the date. Book early and if you want to be part of it, become a partner, become an exhibitor, speak, let us know, write us an email or comment on the YouTube below. No, you don't have to do that. Cool. All right. And while we are out hosting conferences, we are not neglecting our research customers and subscribers because we have another research report just out. It's called the Slater ProGuide Translation AI 2025 version.

15:47Esther Bond:And it is a new edition and it's focusing on the shift from task level AI to outcome driven language AI solutions. So capturing kind of the much more production level nature of translation AI now. So it's 15 practical applications of AI in translation for translation and localization. Quite bite-sized, nice for you to absorb. I think this is, what is it, the fourth? No, the third edition. Third? I think it's the third. I think it's the third. Yeah. Third edition of this particular piece. Yeah, very excited that we're able to launch this. We had the AI dubbing report about eight weeks ago and I boomed this and we're going to launch a bunch more stuff before the end of the year.

16:33Now, I was thinking about if we should cover this. Should we talk about it just very briefly, this Apertus launch in Switzerland? I mean, it is. Yeah, let me just go through briefly, give them a shout out, being the local patriot I am here from Zurich. So the ETH Zurich, the Federal Technical University here in Zurich and the Swiss National Supercomputing Center, I didn't even know that thing existed. They have released Apertus, which is Switzerland's first large scale open multilingual language model and they're calling it a milestone in generative AI for transparency and diversity.

17:13Esther Bond:So it was apparently trained on 50 trillion tokens across more than a thousand languages. 40 % of the data is non-English. And it includes many languages that they say have so far been underrepresented in LLM, such as, drumroll, Swiss German, Romance, and many others. For the very regular listeners of this podcast, you do probably remember that I keep talking about Swiss German and Romance, if that was something that many people care about. But I do. I find it We care by proxy, Florian. We care because you care. Well, there are maybe 5 million Swiss German speakers and 50 ,000 Romance speakers.

17:57Esther Bond:But the Swiss German thing is, I mean, it's in a sense a small language and it would be so weird if I could speak dialect to a computer. I got to tell you. It would be exceptionally strange. strange and this is apparently coming because if these guys did it, I'm sure there will be some way for you to interact with a computer in Swiss German or your iPhone. That would be just so odd. Yeah, and I'm looking forward to the moment because it's trying. So if you prompt ChatTuber to speak to you in Swiss German, it's trying.

18:30Florian Faes:Okay. It tries-ish, but it just doesn't really work. But yeah. So let me just close on this. Apertus they say serves as a building block for developers and organizations for future applications such as chatbots, translation systems or educational tools. So they specifically highlight the translation aspect of this particular LLM.