Episode Transcript
[00:00:00] Speaker A: Foreign.
[00:00:06] Speaker B: Welcome to Localization Today. This is the Multilingual Media podcast and my name is Eddie Arrieta. I'm the CEO here at Multilingual Media.
We have an amazing guest with us who is going to be talking to us about Argos Multilingual most recent announcement, and we are talking about Eric Vaught. He is the solutions and Innovations Director at Argos Multilingual. Eric, welcome.
[00:00:45] Speaker A: Hi.
[00:00:46] Speaker B: Hi, Eric.
[00:00:46] Speaker A: Good to be here.
[00:00:48] Speaker B: It's great to have you. And of course, we are talking about Argos Multilingual launching their enterprise grade AI translation platform, Argos Mosaic. Eric, it's been a couple of weeks since the launch. How is it going? Is that how you pronounce Mosaic or Mosaic?
[00:01:07] Speaker A: With Mosaic, it kind of runs smoothly at that point, but incorporating the AI quality or AIQ as part of the essence of it, that kind of captures the idea for us.
[00:01:20] Speaker B: That is fantastic. And Mosaic, it's been already implemented, it's already running. How is it going internally? What's the team saying and what are the clients saying?
[00:01:34] Speaker A: Well, I think one of the most telling experiences that we had was with a client who.
Let's back up a minute. Everybody's looking for AI solutions, right? I mean, everybody has heard about the capabilities of AI and they really want to see something happening, some, some kind of tangible outcome that delivers tangible value.
Mosaic is not a new tool. We've been building this thing for a while and it's been building on a group of individual capabilities. And as such, we were able to demo it for one of our customers who is, you know, doing what a lot of customers do, which is fishing around for ideas about what the next best thing is. And this particular customer just flat out said the demo blew their mind away. It was just really, really powerful story. We were getting a lot of requests for evaluating it for different contexts and I think the feedback has been really, really positive. Equally exciting has been the data that we're collecting on the back end, as we're seeing in essentially real time the difference between the performance of our kind of default process, which is essentially a post edit process, like we're taking machine translation and post editing it. But with Mosaic, that post edit process is demonstrably faster and the quality is, is consistently better.
I don't want to say it's 100% good in every single situation. There's been a couple of cases where things didn't work as well, but the vast majority of them have seen significant benefits in this case. And our clients are very excited about that.
[00:03:29] Speaker B: And this is really great. To hear, because you're talking also about the compound effect of having an innovative mindset within a company. Could you tell us a little bit more about the capabilities? And you're talking about quality. What are the other elements that Mosaic is including in this release?
[00:03:46] Speaker A: Well, Mosaic takes an agentic approach, so we're breaking down the problem in asking AI to handle it in different key functions. So, for example, one of the queries is to do a summary of the content of the file, and then we're doing that using kind of a core dynamic prompt structure that extracts key elements out of the document that you're trying to translate and turns them into information that another agent can use to improve it. So just imagine, in a very simplistic way, agent number one says, what are the key. What's the summary of this document? What is it about? And what is the key terminology for it? And then agent number two might look up the glossary terms that apply to these found segments and from multiple different other sources. It could be a glossary, could also be a style guide. What's exciting about this is that each of these agents can take these instructions and do something that AI is very good at, which is looking at a big picture and then pulling together the recommendations for it. So as the agents move to the human stage of the process, after multiple dynamic steps, it ends up with a set of recommendations. And what I think is super cool for our customers is it isn't just a recommendation for what to change, but it's very specific as to why it's making that recommendation. So it might say this would be more fluent, or maybe it would argue this would be more compliant. With a style guide. We recommend changing it for this reason. That helps the linguist to focus their mind on not just the what needs to be changed, but the why. And as such, we see we also measure productivity in our tool. So we can watch the how fast or and where the linguists are spending the time, and that helps us to understand better exactly how this technology is benefiting the end user. So it's not a black box for us. We have a lot of insights, and that gives us a lot of opportunities to consider an innovative and continuous improvement mindset. So it's. This is not a. A product that's just been released into the wild and just will live as a static thing. This is a dynamic product that. That continues to improve essentially all the time. I mean, we're really leveraging constant innovation in this approach.
[00:06:30] Speaker B: You've spoken about the role of data and these measurements and these Insights. Could you tell us more about what's currently driving those insights and how important it is for your team to be looking at those? And I understand, of course, in your current role, being able to look at solutions and innovations, measuring and keeping a pulse on what's going on probably makes a lot of sense. Could you tell us a little bit more about what role data is playing there and these measurements you're talking about?
[00:07:02] Speaker A: Sure. Well, number one is that there's no single metric that tells the entire story. So you might have edit distance and TER scores that both tell you slightly different things about the performance of, of what a human has transformed between an input and an output at the station that they're working on. We also consider things like speed and quality. So if you think about the correlation, the better the quality of the MT output, the less time it takes for linguists to bring whatever the MT output is and moving it to a state that the human expert considers final. So you're looking for what are the leading signals that can help us both in terms of estimating what that effort would be, as well as correlating that effort to various input parameters. It's all about the relationship between these. These parameters. So if you're, for example, trying to figure out how much a discount should be, like, how much should the mt, how much does Mosaic add, and how much does that translate into reduced cost? You have to have some metrics there that are based on several different meaningful parameters, one of which is how much work was actually done. And many of these tools kind of measure that computationally. There's also a time element, which we've recognized is not the only one. But generally, if it's easier to edit, it goes faster. We also see another metric, which is if you're. If a linguist is incented to too much to rush, and you're really kind of prioritizing speed over quality, you can have a situation.
We do know statistically that the longer a linguist spends on a given segment, the generally the higher quality will be.
This is a rule of thumb, right? If you're going fast, you'll miss things if you spend time kind of taking your time to get it right. So that's a balance, right? So you have one indicator saying speed is good, it means greater savings, and another indicator that means taking your time leads to better quality. So balancing these metrics is the key of what we're looking for. And so we collect a lot of these and we meet regularly to review the data that we're seeing, and we'll make Some, you know, a senior level decision to think about how best to move the capabilities of the tool forward as a balance between these different signals that we get.
[00:09:41] Speaker B: And of course, I wouldn't want to get too much into your secret sauce, but of course your role plays a significant role in the evolution of Mosaic. And of course these meetings and opportunities that you are having to talk about the data and the insights and take strategic decisions and bring that sort of dynamism that you've been talking about is core. Could you tell us about your current role at Argos Multilingual, how it connects with Mosaic and how does this benefit your clients and the customers?
[00:10:17] Speaker A: Well, I think if I could use one word to describe the majority of my career, it would be solutions. And I think solutions is an intersection between a business requirement, a technical requirement and an operational requirement. So we have an element that really needs to focus on what the problem the client's trying to solve is. That has to be an important part of this equation. So I think any kind of technology can be applied and misapplied in various different contexts. So we see certain types of applications in which this particular tool is a great fit and others which isn't. Another big variable is how much of a discount does this mean? So from the client perspective, how much does it really boil down to at the end of the day? Like how, how, how much difference does it make from a mechanical standpoint? The reality is that it varies quite a lot. Some cases we've seen 20% improvement, other cases 50, even 100%, even more than 100% improvement in overall output, commercial output, from the quality cost parameters that are relevant for this particular project. Then you also look at it from an operational standpoint. There are steps that need to happen and this might depend on what the technical ecosystem is that we're dealing with. We have a large number of tools on the market, large, large numbers of content repositories, data repositories. All these different systems need to work together. I think when you look at all the different parameters or the permutations of the complexity in our industry, we've built an enormously complex industry together with lots of different tools that do fragments of a solution that need to be combined together to make it something that adds value for our customers. And then lastly, the technology itself, there's a combination of tools like Mosaic and there's also all the other hidden tools that make, make this system work, from the CAT tool environment to the, the tms, the mts, all these different capabilities, terminology databases. There's lots of different pieces. So my job is to try to look at this holistically and make sure that all these pieces fit together that ultimately add up to something that the client finds benefit in. And it's. And the key here is that there's no silver bullet that really fits everything.
It's more of a bespoke kind of mindset where you look at the problem that you're trying to solve and you build a solution based on that rather than starting off with some product and trying to make that solve every problem.
[00:13:03] Speaker B: And definitely there is an element of the mindset. We've seen it in our previous coverage of the adoption of the different tools that will make life easier for clients. Speaking about that there might be buyers, those that need these tools, they are not using Mosaic or maybe they are still deciding. What would you, what would you, what would be your recommendations to the companies? Where could they start with Mosaic? How do they get in contact? Of course they can go to the Argos multilingual website. But I'm talking here high level. What are the things in terms of mindset as well as tool adoptions that they should be thinking about with their technical teams if they want to start working with you?
[00:13:48] Speaker A: Yeah, well, first thing I would recommend is something that probably applies to everybody everywhere, which is as AI proliferates in our environment, that there's a lot of things that we should restart, rethink from the ground up. So that means questioning the entire tech stack.
Is everything really adding value that you expected to add? And are you really building a tech stack that's optimized around EAI capabilities? That's one thing we learned when we were building this is that not everything that we've built in the past is still has the same amount of utility today. So number one question and really try to understand what you're really trying to achieve. That's just my solutions mindset talking. That's where you start all conversations. If you wanted to explore what Mosaic can do for you, I think the best thing to do would be to set up some time for us to show you. And I think part of the reason for that is Mosaic is not, we're not offering this as a standalone tool yet, maybe someday. But I think this is primarily a capability that we're offering to our clients.
We can offer it as a standalone function, but as real power is in enhancing the productivity of the human that is in the loop. So this isn't intended to be some a platform that, that our customers can log into independently and do projects outside of a traditional localization framework. But remember point number one, which is question everything. If you really start from the ground up, we may be able to show you some things that will re. Will shape your way of thinking about how to apply AI to. In a localization context.
[00:15:37] Speaker B: And clearly humans are still in the loop more than ever. And we're really interested, of course, in keeping that conversation. Eric, we're coming to an end of this conversation. Is there a message that you might have for those that are listening to multilingual and localization today specifically? And of course, I don't want to put you on the spot, but there's so much conversation about how humans are involved in these processes. How are humans involved in these processes? If you can enlighten us before. Before we go today.
[00:16:11] Speaker A: Well, yeah, I feel this is a very important question as we struggle with the changes in this industry.
Somebody I was talking to the other day was made an observation that humans are essentially the next word, next token engine ourselves. So I think the tendency is that we are just as capable of hallucinating and making up kind of bad information or missing key information as the computers are. But there's a difference. The human can be accountable for making sure that they do the research to really understand the root purpose of the content. They are the. They need to be the truth tellers. They need to be the substantive reality check on this whole machine that we're creating. And it's very, very tempting to start trusting that the AI, if it's very good, must be right all the time. Like, we have to resist that temptation. And the human in the loop for high visibility content is still mission critical as it's always been. Machine translation now does 99.9% of the translations, but the humans are still doing just as much translation. They're just. There's just a lot of stuff that they're not doing. It's expanded to create this whole new market. And I think AI is doing the same thing. It's creating new things for humans to work on and work with. But the human at the heart of it, I passionately believe, needs to be committed to understanding and conveying the truth that they're trying to get across in a particular piece of content, you know, whatever it is like. So don't just throw stuff into an AI and trust it. Go get the source. Go find out what the truth is yourself. This is true for everything in a misinformation era that we live in, but especially true in our industry where we count on the humans to tell us what is true.
[00:18:15] Speaker B: And it's a very reassuring thought. Thank you so much for sharing that. And I love a world where we are critical, strategically critical, to make the right decisions and to, of course, aid ourselves with the tools that are available. So thank you so much, Eric, for joining us today and letting us know more about the Mosaic ecosystem and the Argos multilingual ecosystem.
[00:18:39] Speaker A: Thank you so much, Eddie. I really appreciate the chance to chat with you about this.
[00:18:50] Speaker B: Fantastic. And thank you everyone, for listening. Today we were talking to Eric Vod, his solutions and innovations now at Argos Multilingual, directing this very important function. This was localization. Today, my name is Eric, Eddie Arrieta, CEO here at Multilingual Media. Thanks for listening. And until next time, goodbye.
