#265 Slator Award, DeepL’s $5 Billion Plan, Merz Stirs EU Interpreter Debate

0:00Translators and interpreters make up 0.63 % of CLOT usage and translation makes up 4.5 % of ChatGPT usage.

0:14Hey everyone, welcome back to SlaterPod. Hi there, Esther. Hi, Florian. We're back. Quick episode this week. Next week, it's going to be a guest episode, but today we're just walking you through a bit of news. But first, this week, we had the true honor of awarding the very first Slater Award for the best thesis in MA, thesis in specialized translation, at the Zürich University of Applied Sciences, which is where I studied 20 years ago, graduated 20 years ago. Very nice. Who won? Very nice. Who won? Guy Ravnitsky won. And what did they win? What did they win? Just an honor. A set of steak knives.

1:00No. Guy Ritnitsky, one, who wrote a thesis on how large organizations manage data security and confidentiality in AI. And the award comes with 2 ,500 Swiss francs in price money. Which at this point in time must be like 3 ,000 US dollars because, you know, the Swiss franc keeps appreciating. so yeah congratulations to guy for winning that very first inaugural slater award it was a true

1:34Florian:true honor to be there it was a great uh ceremony uh happened earlier in the week was there with with other graduates from that from that program uh yeah which you know from the uh set age av so zirk university of applied sciences applied linguistics department and um yeah it's very cool to be there very cool to do the kind of on stage award ceremony because you know back in my day we didn't get any we we just got the the certificate or the you know the yeah do you do the thing with the hats where you wear the hats what's it called waterboard no no no no we don't do the hats we don't do the school um yeah no uh yeah so everybody just you know wore whatever they thought was suitable.

2:21Florian:And no, it was great. Great. Quite nicely done, actually. It was about 90 minutes, I guess. And there was a band that kind of played throughout, kind of during the breaks. So it was long enough, but also short enough, right? And then it was a bit of what we call in Switzerland, an upper row, you know, people having their white wine. Not just in Switzerland, yeah. Cool. So again, congrats to Guy. Very proud of this first edition of this award. And next year, it's actually going to be called the MA in Multilingual Communication Management because the, you know, Zerick University of Applied Sciences also has to adapt to the age of AI.

3:04And I think translation has a bit of a reputational problem out there in the market. So it's tough to get new graduates. So they restructured the program and are now calling it MA Multilingual Communication Management,

3:14Florian:which is what, in fact, a lot of the graduates will end up doing anyway. when they graduate from a translation program. So, yeah, let's move on to the next story. We have Anthropic, the company behind Claude, giving us another set of data. I think we must have spoken about this about half a year ago, right? When they... Potentially. Yeah, they had this other thing, this other data point. And now they kept updating their, what they call, economic index, the Anthropic Economic Index. And so they looked at a bunch of job categories. I think it was nearly a thousand job categories about how people are or professionals are using AI.

3:59And so among the top 40 categories was translators and interpreters. So this time they didn't look at, like we spoke about last time, like ChatGPT, when they looked at how people are using AI or ChatGPT here. Anthropic actually looked at how translators and interpreters are using Claude. Yeah, okay. So different. Okay. So that, I guess that's the key difference. Specifically limited to Claude, right? They haven't done some sort of like big survey of other LLMs or no, just limited to Claude. Okay. Yeah, they got their own, right?

4:35Florian:So we were joking before recording this podcast that this is one of the, I guess, stories or things that I want to keep my hot take. I don't want to go down the rabbit hole because it's so much information. So, you know, pardon if there's any, I don't know, lack of nuance here, but I want to keep my hot take because it's a huge piece of research from Anthropic. There's all kinds of nuance and kind of geographic borders and things like that. I mean, think how many data points they would have had to have gone through for this. So, yeah, no wonder it's super complex. It's kind of complex. Yeah. So I guess let's just go through the key stats here.

5:15So basically usage share, they're saying that translators and interpreters account for 0.63 % of all Cloud AI usage. That's a lot, right? Now, if you compare this to the data point we went through last time, which was 4.5 % of all chat GPT conversations were on translation related queries. So it's interesting. So translators and interpreters make up 0.63 % of CLOD usage, and translation makes up 4.5 % of chat CPT usage. So again, different categories there. So different methodologies. I was reminded in our piece, different methodologies, not directly comparable, but still directionally interesting.

6:02I wonder if the 0.63%, okay, so that's the number of translators and interpreters using CLAUD. How many, like, how would you extrapolate that to the entire world population? Like, surely there's not like 0.63 % of the world's population are translators and interpreters. So maybe it's kind of overrepresented on CLAUD.

6:25Florian:100 % overrepresented. Good point. Excellent point. You know, that's where we're back with, well, those are large language. model, right? So the language use case is likely to be like 10x overrepresented. Because I'm sure if you go down that list, let's do this now. Let me go down that list and see what other categories to have. Of course, now I'm bouncing into the wrong thing. Anyway, people can look this up. There's all kinds of, I mean, again, there's like a thousand job categories. So I think we might have touched on that last time. I mean, if you're like a, you know, a dredge operator or something, I mean, there's not much use right now for quad.

7:05I suppose the thing it doesn't take into account, though, is people, unlike the OpenAI study, the Claude study wouldn't show other job categories using Claude for translation. So it's only people who identify as translation interpreted. That was the last data point we covered, right, which was about half a year ago,

7:25Florian:which I think it was also in the 2 % to 3 % category. So that's the thing that you can compare. That's a bit more apples, apples. Like it looks like translation is about a two to 5 % usage of these tasks, right? But then obviously professional translators, interpreters make up a much smaller proportion of it. All right, look, we could go through some of the other points here, but like core AI tasks, translators mainly use Cloud for core language work, translation with context style, rewriting, adapting text, etc. But even some kind of admin tasks like fee discussion, requirement setting, things like that.

8:04Florian:Delegation versus augmentation. This is a tricky one. It says that about 70 % of frequent translator tasks, five out of seven, can be fully delegated to AI. While tasks like proofreading or resolving meaning ambiguities remain more human augmented. Well, that's the last mile, last meter, last foot, whatever you call it, expert in the loop. I like the bit because I did delve into that a little bit as well. And so it's sort of saying it's got three categories of how people are using CLAWD. So directive, which presumably just means, okay, CLAWD, do it. And then you take away whatever response you get out.

8:42And then they also measure the task iteration, which presumably is something to do with, you know, the humans coming back and re-prompting or saying okay do it like this but change x in terms of the tone or the style or whatever so yeah I thought that point was quite interesting to see like for example with proofreading editing and revising translated materials which was one task the level of task iteration was 42 so almost half of the people who are asking Claude to to pre-read, edit and revise are then, by the sounds of it, sort of re-prompting after the first initial output. There's a final later point that's interesting, like some of the other roles showed.

9:28Florian:So transcription, speech and text was apparently 0.07%, but then in a similar bracket would be copywriters at 1.14%, technical writers 1.48 % and creative writers at 0.76%. So yeah, basically indicating tech services, you know, clusters. AI is already very, very, very deeply integrated, of course. And then they also, just a super random anecdote, they also looked at specific geographies. So for example, they said that California shows disproportionate use for IT-related requests, digital marketing, and translation. So IT, digital marketing and translation in California, disproportionate use, and they're speculating, likely reflecting its tech sector.

10:15Florian:Yes, of course. And linguistically diverse population. Makes sense. All right. Let's abandon Claude by Claude. Again, this was the hot take. They have a ton of stuff on the internet for you to review. Hey, you know who's good at marketing or getting better at marketing? RWS. RWS. So shout out to the marketing and PR team at RWS who posted this week that they translated a trillion words in the past 12 months. RWS is language technologies, that is. And they said in the same PR that it's the same as handling the entire digital knowledge base of a Fortune 500 company every single hour. I like that.

11:00That's good. That's good PR, a trillion words. It's very good positioning, I think. well yeah in terms of situating it within the fortune 500 setting or even just using magnitude of a trillion i want to know the exact number i don't want to know that it's a trillion i want to know if it's a trillion and three million or a trillion three million four hundred and twenty five yeah you know trillion is kind of all the rage these days because nvidia is now a

11:32Florian:four trillion company, Microsoft's, you know, three and a half trillion. OpenAI just, when we're recording this on Thursday, raised at half a trillion valuation. So, you know, why not go out there and say I've translated a trillion words? I need to go and look up what a trillion is because I haven't really been paying attention. Is it a thousand billion? Yeah, it's a thousand billion. It is? Okay, fine. All right, good. You know, a trillion here, a trillion there. Soon you're talking about real money. No. All right. So that's kudos to RWS management. We rarely talk about PR, but this is a good one.

12:11Florian:Now, somebody who talked about AI translation that you wouldn't expect is Germany's chancellor, Friedrich Merz, who was visiting Spain and pushed back on the Catalan request to become an official EU language. We must have covered this before in the podcast. Yeah, we definitely did. So, yeah, I mean, he basically said, I mean, first, the prime minister of Spain, what's his name, Pedro Sanchez? I'm not sure. I just have to say an email. I think Pedro Sanchez. He said that they reaffirmed Spain's official push for Catalan, Bosk and Galician to regain recognition as EU official languages. But he also acknowledged that that requires unanimity.

12:57Florian:And then, you know, Merz, British Merz pushed back saying, basically, it's complex, it's adding complexity, and there's resource implication in the EU institution. He mentioned that he was part of the European Parliament. So he knows that translation workload that's generated by adding each new official, like a new official language. And then he suggested that in the median term, AI will replace most interpreters in the EU language service anyway. So that was the spicy quote that we got from that press conference. Yeah, and he said that he sees a scenario where everyone will be able to hear and understand any EU language in their own language directly through AI.

13:41Florian:All right. He did also caution, it's not immediate. And, you know, this is going to be a medium term thing. But that's a big thing for a German chancellor to say, you know, towards the EU institutions who employ thousands of linguists that, hey, basically AI is there in the medium term. So, you know, we also rarely cover these like AI replacement stories that are floating into our inbox every day. But this was an interesting quote to get there. Yeah, can't really ignore this. And also, I mean, if you start adding regional languages at the EU level, it kind of, I mean, where do you stop? You know, there's dozens and dozens that you could still add.

14:24Florian:So it becomes somewhat, yeah, kind of exponential there. All right. So a Google Translate rival is exploring an IPO in the United States. Guess who that could be, Esther? Who could it be? Maybe DeepL. It could maybe be DeepL because it's a Google Translate rival. And this is what I just read is the headline of a Bloomberg article that came out like two seconds before we started recording this podcast. So the headline is Google Translate rival DeepL explores an IPO in the United States. I would guess that would be at Nasdaq. What's an IPO? That's an initial public offering. What does that mean?

15:06Well, you start trading your shares on the stock market. Interestingly, that story actually was kind of originally covered way back in April by kind of a European-focused startup portal called Sifted. We might have even covered it on the podcast. I think we talked about it and we're sort of saying, oh, it could be London, it could be U.S. Good point. Yeah, so maybe now we know that it might be directionally the US, but it's been, what is it, half a year ago? And then Bloomberg obviously tends to be quite well informed, right? So somebody called Vinisi Chan and Sweta Kopinat covered this, again, just today on Thursday, October 2nd.

15:51They say that DeepL is exploring a potential initial public offering in the US, according to people familiar with the matter who don't want to be named because the discussions are private, yada, yada, yada. and they said that DeepL has held preliminary discussions with potential advisors on the listing. So that would be an investment bank, kind of IBD, an investment bank department that takes companies public. They say that first-time share sale could take place as soon as next year. And here's the kicker, that DeepL would want to seek a valuation of as much as$5 billion in the listing. You know, that's, what would that be?

16:2610X RWS or even more. Right now? Yeah, probably more. Quite the valuation. They also reminded people in the article that DeepL raised$300 million at a$2 billion valuation, basically cautioning. 2024, right? That was in early 2024, I think. Correct. And then the article also goes on to say that everything is still up in the air, including the listing venue and the timeline, etc. But if they want to go to the US, they probably have to go to NASDAQ or they will go to NASDAQ. and yeah then the article also tries to give some context saying that AI investments continue to surge with booming demand for tools that use the technology that like NVIDIA you know underlies a lot of this infrastructure we know that DeepL uses NVIDIA chips straight off the rack yeah not much more there now let's go through the hot take that will be pretty pretty big news for the language industry would it not?

17:23I mean, not least replacing whoever's left or delisted from being a publicly traded company over the past couple of years. It's nice to be able to have that sort of level of information and insight into a company when they release the statements and the updates that they are required to. Correct. Probably less in the US though, right? The sort of level of disclosure. Oh, it's a lot, especially in what's called the S1 that you have to file with the regulator before you go public. So those are really long, lengthy documents. I also remember back in the day when Lionbridge was still public, they had to do a lot of info.

18:05So there's a lot of info coming out of the U.S. We just don't know anymore because there isn't anything that we cover specifically that's listed in the U.S. So that would be pretty interesting. I mean, maybe some of the AI, not the AI, some of the interpreting companies are still listed, but yeah. So this would be, again, major news. I think it would be an exit for some of the existing shareholders that have been invested with DeepL for close to a decade now. So that would be quite nice for them to exit, right? That's probably 100x, 200x, who knows? Obviously, massive additional firepower in terms of money for DeepL, you know, really accelerating their enterprise move.

18:45You know, also a very important signal for the language AI kind of industry generally, and maybe even the broader AI space, because I haven't seen many of these kind of, some of the newer AI companies go public. Actually, very, very few, you know, unlike what's... It's too early, really. Is it though? I mean, think about, I mean, for those who remember like the late 90s, right? The whole.com, I mean, there was a ton of companies going public. Now, I mean, off the top of my head, I can't name a single AI company that has gone public. So yeah, let me just remind people of the process. First, you hire an advisor.

19:21That's what apparently DeepL has already done. Investment bank, lawyers, auditors, et cetera. Then you move and file an S1 with the SEC. If you do list in the US, that's where we would get all the juicy information from. So we should set some type of alert. Then you do a road show. You try to get people excited about buying the shares to IPO at a good price. You price it and then you start trading. So let's see. Easy. Easy as pie. Only costs, I don't know, a couple dozen million to do that. I'm not sure. Actually used to be kind of on a tangent, used to be one of the biggest client segments back in...

20:02IPO work. IPO work, tons, tons of IPO work. We translated, you know, hundreds and hundreds of IPO doc, like prospectuses, filings and stuff like that into Chinese when everybody wanted to go public in Hong Kong. So, well, maybe they're trying to drum up some of the excitement on the DeepL side. They unveiled like a marketplace for DeepL API solutions. Do we want to go into that just briefly? Yeah, I think so. DeepL marketplace, that is the launch that they've announced recently. So they say marketplace is a centralized platform. The idea being that it's a central hub for apps and integrations.

20:44The thing that stuck out to me is that not only is it a hub for apps and integrations that have been built on DeepL's technology, so not only apps and technology that is deep L proprietary, but also apps from partners. So yeah, I've got a number of them on there already. I've read up on a couple that were, a few of them were actually mentioned in the press release, I think. So they've got within that marketplace, not just sort of translation and language AI apps, but sort of classic sort of translation management technology, technology but they've got content like more like content management type text and app as well but the ones that were familiar to me were console tech phrase obviously cloudworks weglot robert gridley that we've talked about but and memoq so all accessible by via the defail marketplace and apparently there seems to be a button where you can kind of click and apply to become a partner if that's something you would like to do within the marketplace.

21:54Yeah. And in that press release, they don't refer to Translation AI or Language AI. They now call themselves Stipel, a global AI product and research company. Yeah. That's quite a change. They might. I mean, all of the coverage of the IPO stuff is still Translation AI, Translation AI. So it might take a little bit of time. Yeah, I'm trying to go broader. Google Translate rival. I mean, if that's the headline, right? Yeah, that's a tough one to get rid of. It's going to stick for a bit. Yeah, positioning. All right. And then finally, if you're a global AI product and research company, you need a chief product officer.

22:33And so they also announced that they hired Gonzalo Gallolas as their new chief product officer. And he has quite the background in tech, works for a low-code company called, let me pull this up. What was he called? OutSystems. Yeah, that's right. So 10 years at OutSystems and a couple other things. So yeah, welcome Gonzalo Gallolas to the world of language AI and language solutions and good timing. Hope you get some shares so you participate in the IPO.

23:07Florian:All right, let's move on. I do want to talk briefly, but really briefly about all these AI live speech translation launches, shall we? Yes. Where to begin? Zoom. So Zoom brings AI live speech translation in-house. Because so far they've been just basically, you were able to plug in your solution and Zoom featured it on whatever marketplace they had, right? But now they're bringing it in-house. I mean, we've been covering Zoom's efforts there on the research side for a while and now they're actually making it productized. I don't think it's available yet. I think it's going to be launched in December.

23:45Florian:December, didn't I say? Yeah. Okay. And then, I mean, we did speak about the Apple AirPod launch, then Google Translates live translation feature, the Microsoft AI interpreting API, which we spoke about last time. Then, you know, companies, even startups like Palabra AI raising money. And then, you know, Mark Zuckerberg, Meta was on stage with his kind of live translating Ray-Bans. Yeah. I didn't really follow this. I don't know if it really worked live on stage, but I must have. A couple other things actually failed, like live failed, and that got him, on the one hand, some ridicule, on the other hand, apparently some respect from developers.

24:26Florian:He's on stage and it doesn't work, so it's not like a canned presentation. Okay, but then we came across a post by a Korean news outlet that actually tested Samsung and Apple real-time translation, and they said that the barriers persist, the test reveals delays, and mistranslations of industry terms, raising questions about full language barrier removal. Well, those questions about full language barrier removal have persisted for quite some time, and I don't think they're going to go away just yet. Yeah, I think we must have linked to it in our daily newsletter. I'm not going to go through this in detail, but yeah, so big push in AI Live speech translation.

25:08Florian:I think now we're getting to the point where it's going to have an impact on LSIs going forward. So it will be interesting to see in our coverage maybe early next year what some of that impact is, how many enterprise clients or users are moving to some of these products for day to day. Yeah, I know it will be interesting to see how that sort of shakes out when you start to see some of the companies reporting revenues, for example, whether it will take some time, like a couple of years, whatever, to kind of trickle through. Especially something like the Microsoft AI interpreting API, that should be a big competitor for some of these LSIs.

25:49Florian:But we were in Korea, we're staying in Korea. There was a small deal, I think. Can you talk about that? Yeah, so we've just got a roundup of a few different acquisitions, partnerships, etc. So the first of those being Bearing Lab, based in Korea, as you said, buying Introsphere, also based in Korea. So this is news that Bearing Lab acquired Introsphere. Typically, Bearing Lab has been... So first, I mean, it's got its own AI translation, proprietary technology, but it's been very much focused on the... I think that was a combination of focus and specialized. Anyway, it's a new word. Focused on legal and patent services.