Episode Transcript
[00:00:02] Speaker A: Welcome to Localization Today, the show where we explore how language, technology and strategy come together to drive global impact. We explore how professionals harness AI data and strategy to turn content into a competitive advantage across cultures and markets. My name is Eddie Arrieta. I'm the CEO here at Multilingual Media. And today we're joined by Craig Stewart, Director of AI Research at freyj. Craig is currently one of the main architects of Comet, the neural metric that's become the gold standard for machine translation quality. He holds a master's degree in Language technologies from Carnegie Mellon University and an MA in World Languages. Craig, welcome to Localization Today.
[00:00:54] Speaker B: Great to be here.
[00:00:55] Speaker A: No, it's fantastic. Thank you so much for having this conversation and, and a very relevant time as multilingual. We get to go to many events, talk to many professionals, and we are in a very particular, let's say, turning point in humanity where artificial intelligence is involved in so many conversations and in our industry is making significant progress. Things are changing and it is great to have you here as someone who directs AI research to talk about this. Craig, to begin with, there's been an explosion of digital content that's reshaping global expansion strategies. What do you see out there in terms of North Stars for companies? Is it volume? Is it qualities? Is it a mixture? What are you seeing?
[00:01:42] Speaker B: Yeah, I think it's a bit of both. Right. I think we'll talk a little bit about volume and quality. Quality is my biggest, my most favorite thing to talk about because it's extremely deep and complex.
[00:01:53] Speaker C: Right.
[00:01:53] Speaker B: And so I think I prefer to talk about quality a little bit. Volume is the more boring one to talk about because of course volume is important and that comes from the opportunity that language models have provided.
[00:02:04] Speaker C: Right.
[00:02:04] Speaker B: The idea that now all of a sudden you can generate content at significant scale and go really, really far and you can do that multilingually as well. I think that's where localization is finding itself.
[00:02:16] Speaker C: Right.
[00:02:16] Speaker B: That you can use things like large language models to generate content in multiple languages. The challenge is that evaluating that and understanding that you have created content that not only is linguistically accurate, but also resonates downstream is extremely difficult. So that quality piece as a North Star metric I think is definitely relevant, but it's much, much more complex than I think people realize.
[00:02:40] Speaker A: And this North Star of quality, which, as you rightly so mentioned, technology is allowing us to really, really look into it's shaping then to a large extent what content it's prioritized and the content that's prioritized have a really way, a particular way in which they look.
What does content prioritization look like when relevance and resonance take precedence over output? Just like you're mentioning it right now.
[00:03:10] Speaker B: Yeah, so let's tie into the first question first. Of course, like, resonance is a big thing, right? That's what I'm talking about. The depth of quality is a lot of it is really. So it goes a lot further than just linguistic quality. You're talking about resonance and the impact it has downstream. There's a lot of. For some markets, there's particular cultural nuance that has to be built in. And that part, that's one dimension of that quality North Star metric is the degree to which it resonates within a target culture. And that I think is particularly important.
How.
[00:03:44] Speaker A: And just for me to understand a little bit what we're talking about here, you're talking about the impact it has downstream.
[00:03:51] Speaker B: How does.
[00:03:52] Speaker A: How does that impact. How does that impact measure right now? And how does it inform the quality models and structures that you are, that you are working with?
[00:04:04] Speaker B: So it depends. It's different in different places. Right. There's a number of depends what type of content you're putting out. Like if you're putting out a website, it would be engagement in that website. Things like click through rate, bounce rate from the website. Maybe you're running marketing campaigns through emails and you're looking at conversion rates from those emails. It already all boils down to the idea that you want to capture somebody's attention, keep that attention and cause action from that attention after the fact as well. And also for returning customers, you want people to come back to your platform.
So there's a huge variety of metrics that are a little beyond what we've been doing in localization at the moment. And one of the things I'm keen on pushing and investing time in, particularly in the technology that we're building at phrase, is thinking about how we can go beyond that linguistic quality standpoint.
[00:04:54] Speaker C: Right.
[00:04:54] Speaker B: How can we go beyond metrics like Comet that are purely based on linguistic quality and concerned with measuring linguistic quality and how we can move beyond that and into the realm of things like those engagement metrics, conversion rates and that kind of thing?
There's some significant complexity in doing that.
A lot of the challenges that we find at the moment in developing the technology for it is that many of our direct customers, the localization teams that are using the technology, don't always have access to those types of metrics downstream. And it's quite difficult depending on the size of customer organizations, it's quite difficult to find customers who are really closely tied into that content workflow.
You have some teams who are.
And maybe smaller organizations where they talk to each other more regularly, who are more closely tied into the content workflow, have a much easier job of getting at those downstream metrics and understanding the outcomes and end goals and to what degree the technology is impacting that. For many larger organizations, it's a much more complex picture. And we find we have customers, for example, that are really struggling to understand who to speak to, where the content metrics are measured and collected, and how they can tie that back to things like language data and linguistic performance in their own world.
[00:06:17] Speaker A: Very interesting to look at quality as this proxy for or off the bottom lines, depending on what those bottom lines are. And it really gets me thinking, and I'm very curious, Craig, to know what your definition is of quality in the context of this conversation. How do you define quality for yourself and for the team that surrounds you?
[00:06:40] Speaker B: Yeah, so my definition, I think, is evolving significantly over even just the last six months.
[00:06:46] Speaker C: Right.
[00:06:47] Speaker B: And it comes down to it has been for a while, very closely tied into what we can measure and what we can measure at scale.
There's sort of two sides to it. One of them is purely objective. So the quality of content, or at least multilingual content, especially if it's translated content, lies in things like linguistic accuracy.
And you can measure, now, with tools like Comet and some of the tools we're developing in house at Phrase as well, things like QPS and Auto lqa, you can measure linguistic quality at scale reasonably well. What you can't get to is fitness for purpose. And that's the kind of the subjective element, the subjective side of quality, of content quality. That's really, really hard to drive at. So what is the intent of a particular piece of content? What is the degree to which it serves its purpose in that market? Like, are you actually resonating with the market that you're talking to? Does it have the cultural nuance that it needs? Is it reflecting your style and brand? Those are all things that you can't measure very easily with generic quality tools that we have at the moment. And those are very core to the definition of quality as a whole.
[00:08:01] Speaker C: Right.
[00:08:01] Speaker B: It's not just the objective expectation that the linguistic, the translation, for example, is linguistically accurate, but the degree to which it matches the purpose for which it was intended.
[00:08:12] Speaker A: Craig. And this is very exciting because there seems that there's a lot of work to be done. There's Like a lot of innovation still ahead and that like company engagements, like companies engage, actually going through these exercises, those that actually do it properly are the ones that are going to have a significant competitive advantage in the future.
And these tools provide all of that with all of these level of innovation. And as things are moving ahead, where do most global content teams, you think, still run into bottlenecks or misalignment? And I think we could guess from some of the things you've already told us.
[00:08:49] Speaker B: Yeah, I mean, a lot of it ties back to the volume question, right. A lot of it is scale.
I see a lot of customers and people that I talk to in situations where they're being pushed in two directions. One is to scale as quickly as they possibly can, right? To push out as much content as they possibly can to as many markets as possible. I think there is a perception mostly from leadership that, you know, the advent of LLMs and the generative models means that, you know, we are suddenly able to at a bunch of cost and scale very quickly. And that puts a tremendous pressure on content teams to churn out as much content, as valuable content as possible.
And the challenge comes, of course, when you do that multilingually across markets.
Many content teams are trying to do that in a personalized way and missing the mark where they need the localization team to tell them how to adapt content in a particular fashion to a particular market and struggling to do that at scale. So I think that's volume is one of those bottlenecks. One of the pressures that we see. The second of the two pressures that I would highlight, I think, is the pressure to adopt AI in particular, of course, at the executive level, there's a huge focus on cost saving as a primary driver.
[00:10:02] Speaker C: Right.
[00:10:03] Speaker B: We understand that all of this newfangled AI technology provides a tremendous opportunity to save costs. And as a result, there's a top down pressure to apply technology broadly as much as possible with a little bit of naivety behind it.
[00:10:19] Speaker C: Right.
[00:10:20] Speaker B: It's with the assumption that it can do everything you think you want it to do. And one of the things that we face in localization at the moment, of course, is that understanding that that pressure comes with an expectation that it can deliver equally in all languages, which is absolutely not the case.
It's a very, very difficult thing to do, especially with that broad definition of quality, which bleeds into other expectations beyond just linguistic grammatical quality.
[00:10:48] Speaker A: And Craig, how is AI helping? Can it really help manage both the scale of production, the nuance of personalized experiences? What is it providing its worth.
[00:10:58] Speaker B: So there's a couple of areas that we can highlight there.
[00:11:00] Speaker C: Right.
[00:11:00] Speaker B: One of them, we've been doing things like intelligent routing based on linguistic quality for some time and that's definitely a key piece.
[00:11:08] Speaker C: Right.
[00:11:09] Speaker B: For lower impact content, we've been very successful with customers churning out large volumes of content for a particular market, for a particular use case that is of low risk of low impact, relying heavily on machine translation with a small bit of routing to humans for post editing and review based on some kind of technology like qps quality estimation systems that are focused on that linguistic side.
Now with LLMs, we're finding it's easier, not very easy yet, but it's getting easier to focus on some of those subjective expectations like fitness for purpose. You can have an LLM judge, for example, look at content and just say what do you think? Do you think it's hitting the mark in this particular market? And we're experimenting a little bit with that.
But that relies on a tremendous amount of infrastructure and asset management around it.
[00:12:00] Speaker C: Right.
[00:12:00] Speaker B: Is that you can't just go to ChatGPT and prompt it to say, as a judge to say, what do you think about this content? It relies on inherent knowledge of what it is that you're trying to achieve and things like your brand and style and the assets that surround it. It's incredibly complex to get a system together that can make the best use of your assets and understand exactly what it is that you were setting out to achieve in the first place.
And that's an area that we're seeing some movement I think at the moment. We're doing some work at phrase in profi. We're trying to understand those subjective expectations from customers and trying to draw a line under what it is that they actually want out of the content. Not just the linguistic quality aspect, but the things like the resonance and the engagement and the impact in the brand and all of those things capture those and train technology and build technology around that to try to reflect to what degree we've met that expectation.
[00:12:57] Speaker A: Thank you, Craig. For that context. I think one of the things that we like and see multilingual that's very exciting is, is that we are right now in an era where all excuses for proper global content strategy have almost been eliminated.
Pretty much any organization of any size could have the tools to not only go from ideation to proper brainstorming to generating potential content that can have some sort of presence somewhere. And it feels that we're all almost going in the direction of actually getting to hear from our Audiences like almost like real life. What is it that they are saying, thinking, and how can we engage with that so that we inform our strategies, so that we give them more of what makes sense for the ecosystem that we're building? That it's true for multilingual.
And yet with all of that volume, there is a lot of noise, there are a lot of indicators. And then initially here in multilingual we're like, okay, we need to get to this level of volume because we've never had the volume. So now having the volume is like, okay, let's first get there, then let's strategize. And now let's start measuring things from your experience, your perspective, which performance signals could cut through the noise to reveal the real audience engagement and if our content is being effective or not.
[00:14:32] Speaker B: I think we've hit on some of those already in terms of like engagement metrics and things like that. The degree to which the customers are coming in, engaging with the content itself and sticking around and buying the product and then coming back as well. I mean, I can give specific examples. I mean, to be clear, I work in the AI side of things. I'm a little bit removed from the content metrics themselves, but I have a keen interest in sort of making the transition and understanding what that looks like. And I think that there's a lot of scope in exploring things like engagement metrics with the technology itself. So I'm very excited to see where that goes. I think that, yeah, certainly things like the conversion rate, the bounce rates and things like that are important.
I think we've not yet covered, I think the full spectrum of metrics that we might want to think about in terms of like, I think that there is, as we start to forge this relationship between the localization teams and the content world, I think we're going to start to discover more and more that there are sort of these cultural nuances and stylistic nuances that feed into those metrics at the same time.
[00:15:36] Speaker C: Right.
[00:15:36] Speaker B: So yes, of course, the end goal, the outcome based evaluation in things like engagement metrics are important, but the degree to which we can measure the impact of the content also depends a little bit on the content itself. And there's some new ways I think, of thinking about that content that are coming to the surface now. So I think it's really interesting.
[00:15:56] Speaker A: Thank you. And of course, my question there, and I don't want to impose here, and I don't want to sound ignorant either, but how does phrase help its clients figuring these performance signals out?
[00:16:09] Speaker B: Yeah, this is an ongoing challenge for us. I've mentioned previously that of course, a lot of the users of phrase are buried within organizations and localization teams.
And a lot of the time we're sort of faced the same challenges in understanding to what degree the content is impacting. But the very first thing that we're interested in doing, I've mentioned this a little bit before as well, is going into those profiling exercises, right? You're sitting with customers and helping to understand what it is that their content teams are after. What are the outcomes? I think, you know, we talked about quality before and that's part of it. Part of it is understanding to what degree in particular language pairs quality. Linguistic quality is an issue and we're some way down the road of figuring that out. And then the other side of it is thinking about to what degree they want to reflect a particular style, what is the purpose or intent of a particular piece of content and how that's reflected in whatever it is we're doing with the technology itself. So there's a lot of work that's happening with customers directly and understanding and sort of trying to ring fence expectations. And we're working very closely with some localization teams who are trying to reach out into their content teams and trying to get hold of those metrics and bring them into the platform and understand what they can do with them.
So there's a lot of activity happening across a lot of different parts of the platform. I would highlight profiling as one of the most exciting ones for me. I think we're getting close to a point at which we understand that the output of these systems, the output of these generative systems, can be very, very carefully influenced. If you're very clever about what you give it, right? If you give it a good profile that tells you exactly what you want out of it, and you give it access to highly curated and highly managed assets that allow it to make its own decisions. You can get really, really cool results. And I think that that's where we're going to see most progress. So we're putting energy into that for the time being. And of course in analytics as well, is trying to build out an analytics platform that surfaces these things. We're not at the point yet where we have engagement metrics and things in the analytics platform, but that's a direction that of course, we'd be interested in if we can get there for sure.
[00:18:23] Speaker A: Craig and I almost asked you about Agentec, but I will do that in a little bit just for the use of our own magazine. Our own research on what we want to cover.
It's clear, right.
There was a huge conversation last year. We call it a multilingual the A of AI. There was a lot of fear, but I think this year we're way past fear. We've seen very high levels of adoption, very high levels of innovation, and that's consolidating certain names, including phrases name into the conversation.
And teams have been forced, many teams have been forced to really change the way in which they do things. And some of them now are coming to you. I guess in the past it was you going as phrase going to the clients and trying to sell something. Now, and I guess it's more prevalent that they actually come to you.
Could you speak a little bit about how expectations are shifting for localization content leaders in this world now where AI and automation are priority?
[00:19:25] Speaker B: Yeah, again, I think we touched on this a little bit before in terms of the challenges and the bottlenecks. Right. But I think a lot of the expectation, I mean, there's again, two sides to it. One of them is on volumes and cost reduction.
[00:19:38] Speaker C: Right.
[00:19:38] Speaker B: It's about get stuff out faster.
I think the leadership comes in with an expectation that they've seen ChatGPT and other LLMs and what they can do, and they're very impressed by performance in English. And they say, well, we need that everywhere.
[00:19:52] Speaker C: Right.
[00:19:53] Speaker B: Do that now in Japanese, Italian, everywhere else in the world. We want to use that power of scale to go everywhere and anywhere. Coming with a very naive assumption that that can be successful. I think that's where localization teams hit tension in the fact that they understand that there is cultural nuance. That means you can't just translate content.
[00:20:13] Speaker C: Right.
[00:20:13] Speaker B: You have to manipulate, adapt, and almost in some cases transcreate content to make sure that you're actually resonating with that particular market. And that's definitely become one of the pressures and the other is that you have a lot of just the pressure to just generally just become AI first and adopt AI. And there's a lot, of course, hype at the moment that we're still. I think things are maybe calming down a little bit, but I still see it everywhere. It's that everyone wants to do. You mentioned agentic. I look forward to the agentic question, but that's a big thing at the moment.
[00:20:49] Speaker C: Right.
[00:20:49] Speaker B: There's agentic. AI is the new buzzword. Everybody wants to do it. And so there's a lot of pressure to implement.
And we do see teams coming with we need to implement AI and so they are under pressure to do that.
And I think that there is some. I have a bit of a personal goal in this as well. I mean, so you'll know from my background that I was at one point a linguist and I've done some translation myself. And I do have a personal interest in seeing this transition through.
[00:21:20] Speaker C: Right.
[00:21:20] Speaker B: I really engage well with the linguists on these localization teams, in particular those who have a background and are worried about their careers and their future.
And I'm really invested in sort of seeing.
Figuring out a path and helping localization teams find their place within the broader content sphere and figuring out exactly how to tackle some of these new challenges.
Very, very exciting and motivating to me.
[00:21:45] Speaker A: I'm really glad to hear it. And it seems like at least this year that all paths take to the agents.
Very likely multilingual will call this year the A of agentech.
So many interpretations, so many definitions of what it is. And I understand I'm putting you a little bit on the spot here, and I'm pretty sure it's right down your alley. So is it the year of the agents? Where are we? Because one of the things that happens with the buzzwords with AI is that, well, many are so afraid of it that they just ignore the conversations. They don't even want to call it for what it is. They don't want to understand it. And it seems like with agent tech, something similar is happening where you have the vast majority of people who don't does not want to understand what it can do, what it is capable of. And then you have very small minority that actually it's getting into it and it's trying to figure out how. How. What it's happening around that.
I know what it's likely what the. What the position is for phrase. What is your position? What do you see in phrase? And if you can tell us what we can expect, then that'd be. That'd be fantastic.
[00:22:57] Speaker B: So let's like, yeah, I want to visit this from a particular angle. So yes, there's definitely a lot of. In terms of it being the year of agents, it's definitely the year that everyone's talking about agents. That's for sure. Right.
The value is yet to be seen. But for sure, everybody's. There's a lot of hype around agents at the moment. And I think two things happen at once. And the same thing happened with LLMs, right, is that we do see this hype bubbling up that means people take an interest in a particular technology or a particular solution or a particular way of doing things. And that generates interest and momentum and motivation to start looking at these things. What happens at the same time is like researchers like myself and the team that I work with start digging into these and understanding where it can add value. And that's the core, I think, principle that I stick by when I think about any of these technologies and particularly the agentic AI.
[00:23:46] Speaker C: Right.
[00:23:46] Speaker B: Is that where is the value? Show me the value. And so what we've been very, very heavily focused on in FRASE is that we're not just trying to throw out agents anywhere and everywhere, we're really trying to focus on the core areas of value. So we've very recently we've started on this journey as many of our competitors have, with the phrase agentic content system with our first agent.
And the goal of the agent that we're putting out is to really wrap itself around that localization translation process and fill in many of the gaps that you just cannot cover with raw mt.
[00:24:24] Speaker C: Right.
[00:24:25] Speaker B: We understand that there are lots of shortcomings of neural mt.
Agents offer a nice opportunity and tremendous value in closing many of those gaps and truly driving it full automation.
And that is where we're starting to see this sort of hype that happens and the excitement and motivation around AI actually convert into real value. And that's what I'm excited about for this year. And I think there's probably a nice bit of a curve to this year, if you want to think about it as well, is that people getting very, very excited and talking about it a lot. And I think this is where towards the end of the year we're going to start to see real value for localization. I'm excited. I'm also excited about the parallel breach into the content world and what that means for agentic AI as well. It's not just about solving localization problems anymore. It's about giving things like agents knowledge of the content world and engagement metrics and things downstream and feeding those in to create truly value based solutions, outcome based solutions.
And that's the direction, I think if you wanted to hint of where we're headed, that's exactly it.
[00:25:30] Speaker C: Right.
[00:25:31] Speaker A: That's very exciting because it makes it real, which is detached from the fiction of it will do everything and you won't even have to speak, it will read your mind and it will perform when the evolution and the first stages are looking like a place where there are very specific tasks that, that it can do very well and it can inform subsequent tasks or numbers of activities.
So it's really great to think about it in that way. It's also really great to see the transition in some CEOs, some teams. But there seems to be a need for an understanding on the technology at a granular level and also conceptually and also what the future could involve and also of the internal operations and all of that.
In your view, what are some of the most important mindset shifts, the leaders most embrace to succeed in this new global content race, I guess infinite game that we have going on right now?
[00:26:38] Speaker B: I think, I mean, two things. You asked for one, but I'm going to give you two.
I think one of them is we really need to start thinking about language as a growth driver. I think we need to start connecting language data and the biobehavior and the metrics downstream. We've talked about engagement, conversion rates, those kinds of things, like connecting the dots. There, I think, is a big for localization teams in particular to understand the place of the localization team in that bigger picture and start seeing the expertise that linguists and language providers bring to the table as the asset that enables that transition.
[00:27:14] Speaker C: Right.
[00:27:16] Speaker B: That's one of them. The other, I would say that is, I think, really, really important is I think from the leadership perspective, a bit of a kind of a leveling of expectation around the capabilities, the immediate capabilities and the value that pushing all of this technology brings.
[00:27:37] Speaker C: Right.
[00:27:37] Speaker B: It's very important to be aware in particular of the nuance across languages and the fact that just because it works in English doesn't mean it's going to work in other areas and to really invest in a process by which there's something of a unification of the content localization world. Right?
Not just pushing content teams to throw out content in any language and assume that it's fine. Develop a strategy that involves filters that are reasonably sized. Of course, you want to be able to scale. At the same time, you don't want the localization team to remain as a kind of a compliance barrier to outputting content, but create a world in which the content teams and localization teams are working together to generate content that is of quality. So as we talked about before, of linguistically accurate and, you know, doing the thing it's supposed to be doing, but also resonant, you know, highly resonant content that hits at the cultural nuance of the market that you're trying to target. Driving towards what you hinted at a little bit before was this idea and this dream of this future content world in which we are dynamically adjusting content to individual expectation Right. Is that you've got individuals interacting with your website, and that website language is just one feature or one dimension of that user's experience with your website and your brand, and that the user can come in and feel like the content that they're seeing was made purely for them.
[00:29:10] Speaker C: Right.
[00:29:10] Speaker B: That, that, that content that they're. They're being presented with was. Was made with them. That singular person in mind. I think that's the, that's the ultimate goal and where I'd love to see things move.
[00:29:21] Speaker A: CRAIG and what message do you have for your fellow linguist translators? I hate to do this because it feels very obvious to me when I think I was like, I wish I had the level of understanding in this specific evolution of our world where we have artificial intelligence as being in the middle, in the center of it.
What opportunities do you see, what perspectives do you have there for our fellow linguists?
[00:29:47] Speaker B: Yeah, and I'm thinking, as I say this of.
So I work closely. I'm still tied to Carnegie Mellon, work closely with the linguistic studies, the translation studies program there, and the students. I talk regularly with students there specifically. I think two things. One of them is don't panic, because I really think we're figuring this out. I'm feeling momentum behind these conversations at the moment, and I think we're getting to a place where we're starting to.
The fog is clearing a little bit, and we're seeing that the role of linguists as multilingual content strategists, for example, is pretty clear. There's a ton of things that linguists can be doing in this new world of hyper generation of content. I think that that's very, very clear.
The one thing I would suggest, I think, for anyone in this world is try as best you can to kind of let go of your current expectation around what it is that your role is and what your responsibilities are and think more broadly into that content frame of, you know, what we're actually trying to do is create these personalized experiences for people. And the role of linguists in that is. Is a significant central piece, but it's one piece of a broader puzzle that aims at resonance, engagement, and other downstream things in order to get there. I think the one thing I really continually push is, and this is really hard to do, but invest as much time as you can in the technology itself and understanding what's possible, because exactly as we were speaking before about this bottleneck and the pressures of executives bringing AI to the table and saying, okay, implement, just go for it.
The more that linguists can arm themselves with the knowledge to confront those conversations and to demonstrate their values through building understanding about what the capabilities are.
For example, that, you know, ChatGPT doesn't necessarily guarantee resonance and engagement in a particular market just because you prompt it to translate.
[00:31:44] Speaker C: Right?
[00:31:45] Speaker B: That kind of knowledge is really, really important for linguists to have and to be able to represent and defend their positions.
[00:31:52] Speaker A: Craig, thank you so much for, for, for these conversations. Before we go, any final comments, thoughts, quotes, book recommendations, anything on your mind?
[00:32:02] Speaker B: No, I mean, I think I just gave it that's a reasonable recommendation for anyone, for myself. But I'm really grateful for the time and thanks for having me. It's really, really, really great.
[00:32:12] Speaker A: Thank you, Craig. I think, I think we've had amazing, amazing feedback there.
Very, very great moments that will take us for our snippets for everyone to consume and enjoy. And this, of course, wraps up our conversation with Craig Stuart, Director of AI Research at Phrase. Once again, Craig, thank you so much for doing this.
[00:32:33] Speaker B: Thanks for having me. Great to see you.
[00:32:38] Speaker A: And of course, thank you to our wonderful audience for listening to Localization today. Be sure to subscribe and rate us on Spotify or Apple podcasts and stay tuned for updates. We'll be back soon with more insights from the people shaping our industry.
Until the next time, goodbye.
