AI, WordPress, and the Translations Nobody Saw Coming

Well, hello everybody. My name is Gemma. So nice of you guys to join us today. I'm South African, so I'm really sorry if you don't understand the accent, but just raise your hand and Dan will translate for you. [laughter] Um, so thank you so much for coming to the last session of the day. I know you've had quite a long day, but we're so glad to have you here. Today we are talking about lost in a machine translation. So we're going to take you through one of our projects um that we used a translation tool um and basically the fun the wins the not such wins um and we'll take you through that. So just to introduce you although Betsy has already done a great job. Thank you Betsy. I am Gemma. I am from South Africa. Um, just before I start, are there any avid Friends fans? Oh, thank goodness there's two of you. [laughter] >> Wow. No other Friends fans. >> Okay. >> Okay. So, do we know the scene? All right. Well, we'll explain it to you. Pivot. >> Exactly. >> We will explain it to you. Um, but basically now you know where our slide is going. So, like I said, I'm from South Africa. Um I'm a software delivery manager. I work at side which is a German WordPress agency. Um and yeah that's me >> and uh this is Orestes. Uh I come from Greece. Uh I am a senior WordPress engineer for uh site. I've been with site for uh a little over six years and uh I have been the technical lead for the standx electronics project which means that uh my team and I are basically creating clever and effective solutions for whatever problem he decides to throw at us. I'm the client. I have all the problems and they have all my solutions. I'm Dan Pelato. Um I'm originally from Chicago but I live in Michigan. So, apologies if any of you are Ohio State fans. Um, but I've been with Sandex Electronics for just over two years and I've been in uh the digital marketing space for for just over 18 years. >> All right. So, just a quick overview of what we're going to be taking you through. So, if you're a Friends fan, you will know that the series all start with these, but because there are not very many, this will make more sense. So, the one with the massive global rollout, that was what the ask and approach from our client. Um, the one where some of our translations got a little bit too literal. So, we'll take you into some surprises that we found and we had some fun with. The one where we tried open AI because that's why you're here, right? And the one with the human partner. So, what worked, what didn't work, and basically our verdict on this project. Um, and then we will open the floor to questions at the end. >> So, let me give you just a little bit of background on Standex Electronics. Standex Electronics has been around for many years. were a a large global manufacturer of electronics uh industrial B2B uh components and when I joined Standex uh we had one site one domain and we supported a few languages on subdomains and uh you know we've grown through a lot of acquisitions so we had some legacy sites and I was kind of brought in to manage a a global redesign uh migration project and now current state we still are Sandex Electronics but we've broken our product lines into three brands, Detect Edge and Grid that have uh distinct product offerings and now we are one uh one company but we are really four brands and we are launching sites across 12 different languages or sorry 12 different countries across nine languages. So we are in the process of our migration which started about a year ago. Um when we are done with it we will go from one site to 48 sites and my background is SEO SCM. So I don't know how I got in charge of a massive global redesign but uh these guys are here to help. [snorts] So um let me kind of walk through the problem if if that wasn't clear enough. It's it's a a big large scale project and as you guys may or may not be at different companies that are big or small, our team is very small and we have a big problem to get after. So, so we had to partner with a an agency to help kind of fill the gaps, right? I'm only one person and we've got a team of marketing managers in the different countries for the different brands. So, um you know, this was kind of a crawl, walk, run approach and I would say we're probably in the middle of sprinting, right? We're not quite at the run. You can go ahead and go to the next slide. So, um, as I mentioned, we have limited local support, right? So, when we came on, we have not only brand managers, but they're country specific, right? So, we have somebody in the US for detect, we've got somebody that is global for edge, we've got somebody that's in India for grid. Uh, we don't have somebody in every country in all the countries that we want to have a global presence in. We want to have a global footprint but we are only limited by the certain uh limited resources that we have with those limited resources. Our old site was manually translated. We didn't have an agency uh we weren't doing any kind of AI translation. We were saying hey our you know our Chinese colleague and our German colleague can you guys manually translate this content? Well marketing managers they're not digital specific. So they've got other tasks. They're doing trade shows. They're doing emails. They're very busy. they don't have time to translate a lot of content. So as a result, our old site was probably 80 70 to 80% translated in German, maybe 60 to 70% in in uh in Chinese and then uh Japan I think it was maybe about 50% and so much so that they actually had built their own Japan site for their global audience. Um and that was our reach German, Chinese and Japan, Japanese. And we didn't have uh representation in all the countries that we wanted to play in the countries that we are getting leads from that we're doing business in, but we didn't have that global footprint because our our team is very small and we don't have that that big global reach. And you know, most of us don't speak and read and write all the different languages. And I certainly don't speak or read any languages besides English. And they thought I was the man for the job. So uh that's uh you know a bit of the high level of our of our problem. >> So the approach we followed was our usual go-to combination. So we started off with the WordPress multicight and on top of that we used the multilingual press which is site's own multilingual solution and uh in uh this setup its brand and its language gets its own website which keeps things well separated and clean and multilingual press takes care of u uh keeping track of all the content relations between the websites to help To reduce the required manual effort from the editorial teams, we uh opted for a fairly new at the time uh feature of uh multilingual press which was automatic translations and briefly and oversimplified. What this does is to uh take a the content of a page or a post, send it over to an external service provider and it receives back translated content which then reimpports into uh the page content. Uh for those service providers, we can either have uh a machine translation engine or a large language model an LLM. There are a lot of differences between um machine translation and LLMs spanning quite a few different aspects. Uh first of all they have completely different architecture. Uh their training methods completely different and of course the original purpose they were built for again different. Of course it has its own strengths and weaknesses. For example, uh machine translation engines are super fast and have very strong literal accuracy whereas uh LLMs are great at handling instructions and carrying over meaning and intent. Small spoiler what started as a great strength didn't actually fit this project very well. So in terms of the plan, we did not plan for this translation piece to become its own project entirely. And basically that's what it what it did. We started with the simple um multi-ress well sorry multilingual press the multi-sight um and to auto translate it. So then that scaled to seven markets um and it's kind of started to really bear the weight of the tools that we were using. Um and so we looked at the languages and we thought okay well let's start somewhere. And so any guess as to which language would have been the easiest language to start with? Throw some guesses at me. >> Greeks. >> I heard it over here. Who said German? All right. >> Right. Because that's the easiest language >> naturally. Yeah, I read and write German. Sure. [laughter] I mean, I had a German exchange student one point, so I was like asking her to review our website at one point, but um it made sense for Standex, right? We have uh a secondary headquarters that's based in Germany, which is a big reason we brought on side, their German-based agency. And the tool that we were using for the translation up front was DL, which is also a German-based translation service. So, between side, deep l our local team, uh German, we should have been able to crush it, right? That that was the most obvious. It worked really well until it didn't. uh the words were translated but the let's just say the intent behind the words were uh not really translated appropriately. >> You could say that. >> So at Standex we have a philosophy called uh customer intimacy and it's this idea behind partnering with our customers coming alongside of them from design to development to deployment of custom solutions. Right? really kind of handholding all the way through the process and really building that relationship with that customer. Any guesses as to what customer intimacy translates to in German? [laughter] I am not going to say it, but thank you sir for shouting that out. Yes, it means that we were more intimate with our customers than we would maybe like to be. So you know deep l understood the the the words and the translations but unfortunately it didn't understand the context of the page that was being translated right so it was more of a little literal onetoone translation and um you know when my German colleagues saw the page they they informed me uh what it meant and we quickly had to uh >> say it with me >> penet. [laughter] So, um, as you can imagine, um, >> Oh, sorry, I missed that one. Sorry, that's my mistake. Carry on, Dan. [laughter] >> Go ahead. >> No. So, so what we did is like Dan said, his German colleague found it. The site was live. So, what we did is we picked it up, we hid the site internally. We had discussions around how what we were going to do and we had to ultimately pivot with this because this solution was not working. Um, it wasn't really cost-ffective. it wasn't it was taking a lot of time from us. Um and so we didn't really take it very lightly. So Dan has a really good relationship with Does anyone have any imaginary friends when you were a kid? I had an imaginary friend when I was a kid. Dan has an imaginary friend called Chad GPT. And so it was not difficult to convince Dan to switch over into our pivot. To be fair, I'm not the only I guarantee I'm not the only one who has named their AI agent. Anybody else have a name name for their >> You are the only one. >> Chad. No. Okay. >> I'm >> afraid you're the only one. >> Well, let's be clear here. This This wasn't just a translation problem. This was a process problem, right? We very easily could have just gone in and asked a German colleague to change the title of the page. But this was bigger because this is going to be rolled out to other countries where we don't have local language. We don't have people on the team that can review. So, we need a process that we can trust long term. So, when we roll out into France and Italy and Spain and other countries where we don't have local representation, I need to be able to trust that it's accurate, right? Again, not speaking or reading or writing any of these languages. Um, you know, I put my trust in in the technology >> Chad >> in Chad. Um so you know as you can imagine having limitations of onetoone translation you get issues like like we run into where it's not only just inaccurate from a um well I should say it's accurate from a a translation standpoint from word to word but as um one of the speakers on a previous um seminar actually talked about the cultural applicability of of AI right we need to make sure that the content that we're talking about is representative of of culture in context of the of the content right? It's not just like word for word, but it has to make sense to our customer and the native speaker. Um, not just the people that are working on the project. So, we had things to consider like the brand and language and cultural context, all these different meanings. >> From a delivery perspective as well, the automation of this just wasn't working. So, I mean, as Dan said, he doesn't read or write German or Chinese. And so, we were having Dan's teams were doing the reviews and they were having to do these manual corrections. So they were constantly waiting for these and there was they were updating substantial amounts of incorrect translations. And so when we looked at it in the bigger picture, we needed a solution that would improve the accuracy. We needed a solution that would reduce this manual effort. Um and there were a lot of factors that contributed to it. I mean there were different time zones. There were there was planning into teams. You know people had other work in in dance team that they needed to get on top of. So overall from every aspect it just was a very lengthy we were working a lot harder rather than smarter. >> So we began exploring uh a switch to open AI and uh being an LLM it offered us a way more flexible way to control the translation behavior. It allowed us to uh have project specific instructions and uh provide uh approved terminology and brand guidelines and language specific rules. So we didn't have to rely any longer on a fixed machine translation engine. >> So did we know what we were doing when we brought this solution to Dan? >> Wait, what? >> Partially [laughter] >> surprise kind of. It was one of those yes and figure it out later, say yes, figure it out later moments. We we had confidence in our teams. We had confidence in ourselves. We had confidence in in Open AI. Um and like I said, it was quite an easy sell for Dan with Open AI and it became a collaborative trial between side and between Stanx Electronics. Um we discussed the person the possibility of using Open AI with our internal teams. who researched it, we moved into it. Um, but this was the first time we tried it with a client. Um, and happy to say that it worked out really well. But given his excitement around open AI, um, it was it was very quick to get moving. Um, and what started as just a suggestion internally actually moved into a a really um, a great trial with a great outcome. >> Well, and something else to to keep in mind, too, you know, I mentioned we're a small team. So we had an existing site with one brand and then we launched three new brands that had no identities, right? So we don't have a team of copywriters on hand and it was really me and my buddy Chad that were working on building out a lot of completely brand new content, right? I've got some guiding principles and our brand tone and voice and uh but a lot of the a lot of the content on the sites was net new content that was reviewed by product managers and engineers but we had already created a uh well an intimate relationship with Chad I suppose is [laughter] so it it made sense to when she they brought up the idea of using open AI said of course this this just makes sense for us >> and all this might sound easy and in a way it kind of was. So for starters, OpenAI was already integrated into the core of multilingual press. So we didn't had to build the integration from scratch, which was kind of a nice to have when launching a new project. And it's really easy to activate it. It's basically just a checkbox setting up the API key and picking your model. And it mostly works out of the box. There is a built-in uh system prompt that controls quite a lot of uh things regarding the translation behavior like um um various uh guidelines like uh how the HTML markup of the uh blocks should be transferred and not messed up by the uh translation process and uh the security aspects and guard rails around all these uh translation behavior. So no one gets to do anything weird and uh handling formality and tone and such things. But uh by using the service uh after a while we figured out that there are some rough edges. So we had to do some fine-tuning. So we went in and configured the uh bakedin uh system prompt to have those brand guidelines and the technical terminology in place and improved cart rails and things like that. >> Yeah, it's more than just, you know, setting up a glossery of terms. I think a lot of translation services out there, agencies, um, they ask you for a glossery, especially for an electronics company, as you can imagine, there's a lot of technical terms, a lot of terms that aren't used in in every everyday language. Um, different part numbers, you know, voltages and amps and things that that are very specific to our brand. So, we had to go beyond glossery. And I think, you know, working with AI and being able to customize prompts takes that to the next level. So one of the things uh I wanted to do so we didn't have this issue again is we need to like I said improve our process. We needed to bring in some quality assurance. So we actually met with two different people uh two of our marketing managers and said all right let's take the current state here's the here's the page uh you know we took a product page a technical page um you know an about us page different types of pages and pulled them up in their current state. These are the ones that had been translated in open eye. This is the current state that's not live on staging. Then we're going to go through and we're going to take that same page, log in, and we're going to republish the page with multilingual press using OpenAI and then show the new updated publish page side by side in, you know, in a in a browser. So I was screen sharing with our our German team and our uh Japan team and our China team. And I said, "All right, read these thor through and tell me are is it better? Is it worse? Is it inappropriate?" Um, and I'm happy to say it was uh they were both astonished at how well the translation had improved, not only from a a local language standpoint, but also from a technical standpoint, right? It it it read as though um it would be understood by our customer, the engineer. So we wanted to just uh not only do like a a hallway testing comparison with our stakeholders more or less um but we evaluated it on two criteria. If you want to go to the next slide we looked at how well does the translation go from English to local language whether that be German, Chinese, Japan etc. making sure that it it reads accurately. So for us the source of truth we start with the do in English and then everything is translated out. So, how well is it? What is the readability in local language versus deep and and open AAI? But also taking the translated language and then translating it back into English to understand did it did it accurately match the intent behind the original um content, right? And DL did good. If you use DPL, you know it's it's it's very dependable. Um they do a really good job, but there were just certain things that we needed to improve upon. So they were graded well. They in terms of readability, they were still an A high B, but OpenAI was better. It was a higher higher grade. Same thing with the semantic match back into English. If I had to put this in a, you know, a school grading system, right, a deep got an A. In a lot of places, that's good enough. But with OpenAI, it was an A+. Now, neither of them I want to call out is 100%. Right? We know this is AI. There isn't anything that AI has done yet that is 100%. So keep that in mind. But it was dramatically improved where I now have trust in the process, trust in the team, trust in the output of switching to AI that we can launch sites in languages that we don't have local representation and feel 97% confident that you know it's going to read well for our customers, our engineers and our stakeholders. So, we found some more favorites and as Dan mentioned this, there were also some some caveats to using Open AI. Um, so we're going to take you through just a couple of them. The first one is the Japanese. So, Japanese got the tone wrong. So, we sent it over to Dan's colleague in Japan and when he read through it, he kind of said to us, "This is a little bit more like surfer beach bum tone than I would like." Now, I'm from a surf town, so I was like, "Well, what's the problem?" Um but I suppose when you are a B2B engineering website um you really don't want to you want to avoid that. So our fix I mean it was as simple as a prompt enhancement um which we did. >> Yeah. Uh that one was not a favorite but definitely worth remembering. So uh one day uh we received a message from Dan saying like look the blog page is in Hiddi and that would be nice but for the Indian website we're using Indian English. So the blog page had title content slug all translated into HI which of course broke navigation broke internal links it caused a ton of issue. So we reverted back to the last working version and then again went into the prompt and added more guard rails so it respects the language and local and then we got to the French version and um so standex electronics has uh has magnets that come in packaging. They're enclosed into housing. We have both uh magnets that are in housing and are not in housing and those are referred to as bear magnets. And uh the good news is it was translated properly. There was no issues. There was no concerns from the French team. But as you can imagine when I go back and I check that how well it translates back into English, it translated bare magnets into naked magnets. And of course you could just imagine the fear that came over me having just gone through what we did. And I had to make sure that this was in fact not inappropriate. It was uh accurate. So that uh it was correct, but it was I don't know if I'd say it's funny. It was a little concerning at first. But we we were we were safe. >> So we had we looked at this and we had some cost comparisons and some other open AI wins which Dan will take you through. Um but it we we saw a lot of benefits to this collaborative trial and we thought it would definitely be helpful and knowledge that we would love to share with you. Um we definitely took it into a few of our projects. Um and even though it's off a little bit slightly off just the translation of open AI Dan, why don't you take us through some of the quick wins we found? Yeah, I mean we were evaluating costs, right? We're switching vendors. So, one of the questions that our management team is going to have is is, hey, what's what's the what's the investment difference here? Because we had an expectation going into this of what cost would look like. And with Deep, there's there's monthly subscription services, there costs for so many characters, and going from one site to 48 sites, hundreds of pages for each brand, you can imagine this is going to get pretty expensive pretty quick. Um, and it wasn't astronomical, but it it came with a cost. And then when we looked at OpenAI, and we know OpenAI and all the AI tools like to like to charge you by tokens, I said, "Well, let's, you know, throw some money at it and see what ends up happening." And it it literally I mean, we were pushing content every single day. And it was costing less than a dollar a day to the point where with the Japan example, instead of going page by page and retroactively fixing each page when there's hundreds of pages, I said to Side, I said, "Can we just delete it? It wasn't live yet. And just redo it all because it had been translated in DL. And now that we had OpenAI, can we re-ransate the whole site? And Arrestus was very nervous. He said, "I don't know. That might cost a lot." I said, "A let's do it. It's I'm not going to go page by page. That'll take too long." It was less than 30 bucks for the whole No, two sites because we did the electronic site and the detect site. >> That was a no-brainer. I said this was fast. It was very inexpensive. And most importantly, it was scalable. Right. This is something that we can then take and roll out to all the different other countries very quickly, very cost-five. >> And apart from like the cost part of it, which of course is important, there was another win. So one of the problems that uh uh Dan came to us with was that they are a team of limited resources and we tried to keep the uh amount of time required for manual things to the bare minimum and day-to-day operations of the uh current setup. showed that for things like publishing content across all languages, we were now looking at 15 to 30 minutes depending on content length. While in previous setup with manual effort, that would have been hours maybe even days. And similarly for cloning a whole website and re-ransating the uh whole thing we were looking at maybe a few minutes for the cloning process and then a couple of hours for the whole translation in the previous setup. Again that would have been days maybe weeks. So there was a huge reduction in the required time. And a pro too that I don't think you guys knew is if you do partner with an agency, it works to your advantage if you're US-based to hire a European agency because you can give them something to do at the end of their day and when you wake up the next morning, everything's magically done and taken care of. So that was also just that's just us. >> That's right. Yes. Yes. >> Yeah. But yeah, it it cut our time in in half. Like I mentioned, not only a small team, but they have other things to do. They're not sitting around waiting for for content to review. So, you know, being very respectful of their time and their priorities was was a win as well. >> Yeah. So, I mean, from a from a software delivery perspective, it really flattened our project timeline. Um, and I just wanted to be clear like AI, it didn't solve our translation problem by any [clears throat] means. But what it did do is it really removed that bottleneck of the, you know, the whole project. I mean, it was the operational scaffolding around translating the sites, not necessarily the translation itself. So our overall verdict, open AI translations are really strong, but human review is actually essential. So if you came to this conference and you're looking for answers wondering whether AI will steal your job, anyone in manual translations here, then I think you'll be fine. But that said, it I think AI does still need a human partner. I think that's been one thing that we've >> kind of throughout the entire entirety of our project. Um, and the real surprise for us, which I suppose is really nothing to do with the the coding and the actual day-to-day, but for us, it really built quite a uniquely dynamic team going through all of these trials. Well, that's my >> dare I say we became friends. >> Sure. [laughter] But yeah, I mean, every every one of these presentations, there's a lot of AI presentations this this week. I I'm pretty sure the takeaway is is trust but verify, right? AI is great and it's getting better. I mean, even when we started this this evolution to where we are now, translation every week, every month is always improved with every prompt enhancement, but it still requires intervention. You can trust it, but verify that it is accurate. If anyone's familiar with that famous line from Ronald Reagan. >> Yeah. And there will be certain things that are not going to work out of the box. And I think sometimes we come in hoping that these things will work out the box and we can just have it done and put it implemented and it'll just be okay. But it's actually okay for it not to work out the box. Human connection is such a vital part of our day. Um, and I really think like without it we wouldn't have found the dynamic that we have within our team. I mean we used the friends theme because in our weekly calls it probably takes us a good 20 minutes to get to work. >> [laughter] >> Um we really I mean as you can imagine with all the fun examples so Dan always comes with something new and exciting. Um but yeah it's been great to work in this team. So what would we do differently if we were to look back at this project? So when we asked each this question to each of us my first initial answer was well nothing. I mean it was great. We found loves we found friends. Um we have a successful project. But actually I think from my perspective I would have trusted our curiosity more. I think we get into a position where our day-to-day keeps us in a tunnel very focused perspective. Um and you you get into a habit of it's always been done this way and even though open AI could help you and it could really enhance your your dayto-day. I think when when things have been done a certain way forever. And I think if we had been more curious going into it, we probably would have solved the German um model that changed your business model. And uh in Greek we have a phrase it goes by it roughly translates to if I only I had known earlier and uh yeah if I was to do this again I would definitely go with open AAI or a different LLM from the get-go. This would have saved us a lot of time and would have allowed the team to do more things more quickly. >> Yeah. And one of my one of my changes that I would have done differently, we actually implemented and and pushed live this week. I would have changed and kept all of the URL slugs in English. Uh we thought an advantage to having, you know, local language sites in every country was having a translated URL. and until one of our colleagues brought it to our attention and say, "Hey, these characters, these Asian characters especially, don't work with a lot of our marketing automation tools and marketing platforms. It's causing problems. We can't track things." And come to find out when we did some competitive research for a B2B space that's a global company, including our competitors, all the URLs stayed in English. And a lot of us don't look at URLs, right? We're we're we're navigating through organic search or we're clicking on a link in an email. So, it really is more or less irrelevant to what it is actually written. And then we can gain a lot more um tracking and visibility into some of our platforms. So all of our sites now are on English URLs. Wish we would have started that from the beginning. >> Yeah. >> Thank you. >> And for my friends friends fans, can we we do this together? Anyone? [laughter] Okay. Thank you guys. That was really interesting. I hope that uh we have some questions. Um there are microphones on both sides. I invite everyone to come up and ask questions while we still have these lovely folks in the room. >> Yeah. What was the tool that you used in the first place that gave you the the bad intent that it translated literally? >> It was um a translation service called Deep L. It was a a German-based tool that is I think commonly used for translating things like um well I don't know if they do I'm sure they do web content but they do a lot of uh cataloges or or paper print translations. Um but that's what we were using before. >> Okay. Um, why? Um, I'm doing stuff in Spanish and the Spanish is complicated because the Spain Spanish is one thing. Mexican Spanish is something else. Latin American, I'm from Puerto Rico, which is different from Cuban Spanish. And I've been using Google Translate, um, which has been doing pretty good, but I still have to proofread it. Um, >> the verification process. Yes, >> exactly. Um, [sighs] [gasps] what my my question is is is what do you recommend? I don't know if you know, like if I'm going to translate some some pages, not all of my one of my websites, um, should I just keep it in generic Spanish, business, or, you know, >> why you take that one? That's a very good question and I really believe that it boils down to your audience and your intent. But uh if you want to support different audiences that are very localized, I could go do different localized versions. So supporting Mexican, Spanish, and Puerto Rican Spanish and whatever. And uh by having an LLM integration into this whole translation process, you can pass all the required instructions. So you can ensure at a certain level that the cultural dent and the um specific local rules are in place and the content carries over dent in a correct way. >> Okay, thanks. Great question. >> Hi, my name is Mark. Uh, I have been using Multilingual Press for at uh, my company for the last six months. I think we ran into your team last year at Word Camp US and uh, so started using it. We started with uh, two languages, Spanish and Vietnamese. And we also started with DL. Spanish seemed to work pretty well. And again, I know that that's deep and not side, but uh and then we had almost exactly what you uh described as the beach bum style for >> I thought you were gonna say it was the other one. No. Okay. >> No, no, no. Uh for Vietnamese. So, uh as soon as I get back, I'll be tasked with uh three more languages. Japanese um and I'm blanking on the other two, but German might have been one of them. Um so, you've given me a lot to think about with the use of open AI. So, we're probably gonna have to to switch at some point, but um so I'm wondering um so as we went through this process, we discovered some areas that other plugins seem to handle, but uh multilingual press didn't. So, one thing in particular was like your global strings that are in the theme, uh things like read more buttons and such like that. Is there a plan to integrate that at some point? Uh, is is that a separate plugin? >> That's a great question. Unfortunately, I don't have the answer. What we usually do, uh, it's to provide translations with the theme, with the plugins. Uh, but that's a great question. I'll make sure to pass it on to the multilingual press team and reach out. >> Sure thing. And I don't know if uh just to speak on that too, I don't I don't know if you're working with site or not, but there's obviously an advantage to working with the agency that developed the the plug-in and there's a lot of custom functionality and custom modules, content blocks that we created for ST or side created for standex that wasn't part of MLP and you know coming at this project with the understanding that everything was going to be translated and when certain things weren't we brought it back to the team and I I I think it's fair to say MLP has evolved and has improved. So, I don't know if there's different versions of it that are using, but I would definitely reach out to the side team. They've been super helpful in and making sure that elements like like buttons and different custom functionalities are are translated are included in in the MLP product. >> Yeah, for sure. And so, based on what I just said, I also wanted to point out that it was extremely easy for us to build our solution on top of your product. uh a couple prompts to Claude and and it tied in and it tied in with the DBL integration that was already set up. So, it was easy for us to figure out >> because the plugin is great. Um yeah, thank you. >> Thank you. >> Awesome.