AI, Translation and Quality: A Client‑Side View on Multilingual Content
27 minAudio
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! How is AI changing the way companies think about translation, quality, and multilingual content? In this episode, I talked to Hristina Karagueorguieva, localization manager, to see what goes on before choosing tools, vendors, and workflows. We explore questions like: What really changes when a company starts using AI for translation instead of relying only on humans? How do you decide who should translate your content: freelancers, small agencies, or big providers? What do you need to have in place before you choose tools and partners, and how do you prepare if your company suddenly needs to grow into many new markets? I also ask Hristina what “good quality” means now that we expect basic things like spelling and grammar to be correct by default. We talk about why simply counting errors is no longer enough, how people actually read and feel content, and why we need to rethink quality so it reflects real user experience and business goals - not just word‑for‑word correctness.
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Hi, everyone. I have a special guest. I don't know anything about Hristina. Let's see how she pronounces her name. Hristina should be the one. Let's see. This is always that moment. So we have been connecting on LinkedIn for a little bit now. We have the obvious interest in common, but Hristina actually comes from the side of the client, let's say. So let's see. Thank you so much. Cristina, for being here. This is really hard. How do your friends call you? You'll tell
me. Everyone calls me Cristina. OK. So it's great to have you here, because I love having all perspectives possible in the language industry. And you kind of come from the side of the client. So I cannot wait to hear everything about it. But first, let's hear a little bit about you, where you're from, and how you got into all this language shebang. Thank you so much for being here. Thank you for having me. So I can start in two possible ways. So the first possible way would be the
LinkedIn way that asked me to describe myself in 300 characters. And that would mean solutions minded expert who finds structure and complexity and thinks outside the box and design sustainable programs or delivers measurable quality gains. But I'm sure you're all that. But I could also start with the way I like and would start with my love story for translation and localization that started when I was 15 years old. I translated
the poetry of A. A. Milne, better known for Winnie the Pooh, for a birthday present for a younger cousin. And then this continued with me reading the two volumes of the English -Bulgarian Dictionary taking notes. then followed by English studies at the Soviet University and me dedicating myself to literary translation and editing. So I'm not only a localization specialist, I'm a translator by heart. And I'd stayed so all these years,
which is around more than 25 years now, almost 30 years of me being involved in translation. And I have both experience on the client side. and on the vendor side, and also as a translator myself. So this is a testimony that this has been the professional love of my life. Yes, I totally hear you because I'm exactly like you, right? So that's why we met somehow in the universe of LinkedIn. We came together. So thank you so
much for this, and it's you in a nutshell, I know. Let's now go a little bit slower. So you just... finished university and you had already been practicing and experimenting with translation because you thought that your cousin had the right to be able to read Winnie the Pooh in their language, right? So that's amazing. I'm not that romantic. I admire that. So then in the real world, in the world of work, how did that unfold?
Well, basically I started in localization. as a freelance translator and then I graduated to the English service of the Bulgarian National Radio when for about a year translation was all about faxes and typewriters. Unbelievably so, my translation career started on a typewriter and then I graduated to using a computer and ever since... Basically, I've seen the transformation of the localization industry and how we do translation.
At that point, it's not even called the localization industry. Different technologies from the dongles that you had on Trados, which was the first translation tool that I was using, to doing everything on cloud and now integrating AI solutions into translation workflows. So this is how it all started. I've worked on the client side and on the vendor side as I already managed. And I've seen the transformation
of the industry from the need to scale up technical content really quickly, which actually coincided with globalization. Basically us having to reimagine ourselves and also seeing this element of translation returning to its roots nowadays with the localization industry trying to Focus more on the value that
localization brings, not only in terms of cultural adaptation, but also in terms of translating the world to otherness, also for commercial purposes. This would, for example, include conversion, speed to market. the emotional part or the emotive part of marketing content that should speak not only the cultural language of the user, but also
their private language. That's true. That's true. Private language. Now there's a concept. I like that. We should explore that a little bit. So what do you mean? Yes. So basically the way that people speak in their real world, like what we call in linguistics, their ideal act, right? Like the particular way in which one expresses themselves, right? Which has consequences in everything that we do. Absolutely, right? Whether
if it's in our own language or our second language, third language, whomever has them, right? So how do you live today this whole intersection of things, given the fact that you work with a multinational company, right? So they're not your client, they're your company, right? So you're a part of the team, which is completely different from most of my guests who usually work for clients and my situation too, but you actually work there. That's your job. So what's
the story these days? The story these days is that localization for most of my clients, and I probably have to mention this, that I'm a localization consultant. So even though I work on the client side, not someone who is internal or hired. So this is why for many of the companies that I consult on the client side, I actually see that localization moves upstream because clients are
working on their own AI based solution for multilingual content development at times. and there is a lot of experimentation going on which I think is a very positive outcome because we are developing use cases or people within the corporate environment and also outside of it are experimenting with AI which in the end will bring so much knowledge and so much trial and error that this would support us building better processes for localization
as well. So I thought you weren't there. I guess I was wrong. But I guess it's irrelevant because you have a specific role that nobody else has, right? So that's irrelevant. Can we say the name of the main client? In principle, I would try to avoid mention. Okay, that's great. We shall not pronounce the name, right? So what is your work really like? Do you organize content? Do
you contact vendors? What do you specifically do in your daily activities with that, let's say, main client? Basically, I target my services to the current demand of each client. Um, this can be very varied work for some clients that would be deciding what kind of vendor model to use, because I see recently that with AI, some clients want to have more ownership on their own processes and want to structure content completely differently. So then localization will be part
of content creation and therefore they would rather use for some sort of an internal team to reach that type of ownership that they think is necessary when they implement AI workflows. Other teams, for example, that don't have an internal localization team, or for example, when content development is more dispersed, might want to centralize this through hiring a localization vendor, which might be an LSP, and that also require having a TMS provider. It is really very
different depending on the setup of the specific client. Some clients contact me because they have specific issues with localization quality, which is what I'm trying to specify right now. I do think that we are now beyond the vanity metrics in localization. We are beyond the MQM metrics. And what we need to work on is basically reconciling the strategic value of localization. with the quality metrics that we implement and
potentially upgrade the quality metrics to actually match the increased demand for very targeted quality on the parts of our clients. Because no one is talking translation anymore. No one is talking about replacing words. with other words, which might have been the rationale and which might have been the paradigm when we started scaling up, um, simpler technical content. And I'm not saying that technical content is simple and not at all very specialized is very heavy
on terminology, but it was simpler to actually scale when we started designing our first TMS tools. Now this is the basis. So basically, all clients demand good localization. All clients demand that there are no spelling grammar or terminology errors. And they would like to see the added value of localization is that they are starting to look, okay, does this local this
specific localized content? Does it help me realize my business objectives? And these business objectives are very complex within the hierarchy of an organization because also different stakeholders have different needs. Something may be relevant to the product team that is not relevant to the marketing team. So this is really these type of functions nowadays and consultative functions have a lot to do with
basically talking to all stakeholders, checking what their pain points are. and seeing how this can be integrated into a holistic approach that is also feasible and deciding solutions around it. So this is what I've been doing. So you mentioned TMS is on the client side, we'll call it that. So how's that process when you come in that's already there or do you advise the client in the options in the market? Because I can only
imagine that for a company that is not an LSP, like a classic type of language service provider, it must be a very specific type of shock to actually be faced with choosing something like that that will have such technical consequences with all the possible matches or non -matches, compatibilities or not. It might have with everybody else's tools and TMSs. So have you ever been faced with with
this question, like arriving and not having one and having to choose or implement one? Yes, a lot. And this basically depends on the maturity of the company to integrate technical solutions into their own internal processes. Whether there's, for example, an engineering team to support this, because usually this also requires specific APIs to be. developed and then connected between the content management system and the TMS of the client and different clients are on a different
level of maturity in terms of them being able to onboard a localization exercise or rather decide to outsource the localization exercise to a strategic partner. And then the issue would be which LSP and which structure would be able to best match their architecture of needs, if you wish, to the complexity that many companies actually have when it comes to expectations for
content and localization. So you're helping this process of not only implementing technology on the client side, well, our specific technology for our industry, right, or our way of doing things, and then matching that vendor. So what kind of process do you go through? I love these things, right? It could also be combined. What I do, I just audit the client for whom I'm working. And basically, I also try to evaluate their knowledge
of localization per se, because then a different solution will have to be put in place. And then what might also be relevant, for example, is to establish to what degree stakeholders in this company really need to know a lot about localization. Because sometimes this is not really necessary. what they need is a solution for a pain point but they don't need to have all the knowledge of how to do it because it is us experts who
should be able to advise them and build the structure that accommodates their needs better and then advise on virtually everything or arrange what is possible for that specific company. So it was that kind of interplay between demanded supply and also orchestration. architecture that would allow the client needs to be met best also in view of the future because Sometimes companies
start small but then all of a sudden they need to onboard 20 new markets Because they grow they're very successful. So the question mark always is how to build anything, any solution or any type of infrastructure in a way that is sustainable for the company long term. And this is always a puzzle. This game is always a puzzle. So sometimes we get it right. Sometimes we get it wrong because sometimes it's also very difficult to predict where this company is going the long term in
order to provide better decisions and then basically to provide advice as to what the company can do. in case of changes of scope, so that their transition from one maturity level to another is a success story. Because we all know what a pain it is to move between two TMSs, for example, or two translation vendors, or maybe two people translation vendors. So talking about that, I suppose that for the big clients that you work
for, you actually have... multiple translation vendors. Is that the policy? I know it's a procurement policy. There's reasons why you should have only one or reasons why you should have multiple. So how do you decide and then onboard those strategic linguistic partners within the workflow that you already know and helped create and orchestrate with the original client? Each type of supplier
Has specific benefits and then downsides for example if we look at the Freelancer model. Mm -hmm. You could hire very proficient freelance translators, but they're completely not skate scalable. So if you are translated sick Then you have to wait. So there's always that balance between full ownership full one -to -one contact with the person in charge of localization for your local market and turn around time because
or scale for example if you need a lot of content to be delivered within the short period of time of obviously the internal linguist model will not work on its own I have to build up something else there are specific benefits for example with working with boutique localization agencies that are very much specialized and they are wonderful examples of LSPs, small LSPs that are focusing on a specific area and can deliver really fantastic
quality when it comes to their area. For example, extreme sports or for example, medical translation agencies. For sure. Or for example, creative agencies that combine copywriting with localization and maybe also SEO or other services relevant to e -commerce. There's also the big agencies such as the major LSPs that definitely have added
value when, for example, the service level is going to change over time because they can absorb any service level expectation and easily onboard new services to their offer. Also, they have very efficient project management. And they also do, in most of the cases, they do have the technology and the technological know -how to be able to meet any type of goals that the organization
needs at the moment. What the downside of it would be is that sometimes because of the outsourcing model, developed for other times, for times of scaling up. When it comes to very specialized translators, the quality or translations, the quality may not be uniform, just because usually big LSPs work with a huge pool of translators. Thousands and thousands. Which is a challenge that some of them face now because when it comes
to very targeted quality And to being able to basically monitor quality on a granular level in order to provide very targeted solutions They might struggle with this just because for example, the Vanderpool was created for different times So what about when you talk about quality, right? This is that keyword that I always say that it's not even a factor these days. I mean, who tells quality, right? Every single person, it's supposed
to be the case, right? So I guess it's no longer a selling factor. However, from a linguistic point of view, It's the goal, right? It's the golden rule. So how do you within the context of a final client, right? How do you make sure that everything that is happening and coming to you from your vendor, from your chosen group or single LSP, how do you monitor that? Do you have like a third party? review every now and
then? Do you have internal technology that will somehow flag if there's something wrong with it? I know that all of these possibilities exist, so I'm just wondering what your experience is. Maybe if you allow me, I would just walk one step back. For sure, absolutely. Maybe start with a bigger picture. What you would usually hear in our industry is that quality is very subjective, and this is something that I definitely do not agree with because we know what good quality
is when we see it. Objectivity of this is that we have designed our quality system based on metrics that don't necessarily measure quality. They measure technical parameters or what we can very easily call now vanity metrics such as spelling errors or grammar errors or punctuation errors. that come from the pre -industrial so
to say. I mean localization if you would allow me to actually use this term. Whereas now this type of quality is already expected. So we built our tools and this has to do with how we work in localization. We work with silos. So you would have the linguists and the language experts on one side and you would have the engineering experts on the other side. So the way we integrated quality
into the technical solution very often showed this disconnect between linguistic expertise, which is not so much based on the technicalities of word -by -word translation or whether there's a typo here or not, rather on Discourse analysis because we as humans don't think in sentences We don't think in segments whereas we built our translation tools to segment by sentence, right?
So basically breaking down language Into something that is unfeasible from the point of view of the linguistic mind Our problems with localization quality came So this is why, for example, when we have the technical architecture that still limits linguists as to what they can do. And the quality metrics that was designed for this technical architecture, this technical architecture does not know how to actually absorb feedback.
Because let's say you have two linguists. One of them follows the MQM metrics verbatim and only fixes very obvious things. Right. And then another linguist who actually thinks in terms of discourse and in terms of the emotive capacity of language and in terms of fluency and in terms of user perception, they would make make completely different. corrections to the same text. And
then in the end, whoever would be arbitrating this text would say, yeah, but this is very subjective. Whereas we basically know that this has to do with how we orchestrated or how we designed our legacy systems. And this is why I think what we need now is to completely redesign how we conceptualize quality, because now we're using LLMs. LLMs don't make spelling mistakes. And the expectations from our clients is that now
that we are equipped with these far more sophisticated tools, that we should be able to deliver far more than just translation. Okay. And therefore I think that the next big step in our industry would be to sit together, potentially also bringing universities and researchers. and technology partners and experts within the localization
industry with handouts and experience to basically design ways to go or how to accommodate this paradigm shift because AI is here to stay. There's a lot of shadow localization happening on in which people will translate using AI. on themselves, but then they would lack the governance that is required to really deliver that type of very
targeted quality that is expected of a professional. Wow. Well, I mean, we're going to end here just because I think you went so high right now. You said everything that I think needs to be said and heard and internalized by everyone in the industry, is that this is not our single person or single vendor, single company game. This is
now a much more universal conversation that we need to have all. stakeholders and parties need to be involved. We need to talk more about it. And this is why you're here, right? And this is also why I do this because I think it's impossible to keep going without opening the conversation to the world because this is no longer our industry. It used to be with our tools, our secrets, and now the cat is out of the bag, obviously, and clients are a lot more informed, which has a
huge impact in everything that you do. Thank you so much, Cristina. I have to prepare my throat for this. So it was great to hear from you. Amazing. This is a unique perspective from someone who has a lot of experience. But way beyond that, you really take this thing, put it in your mind, and I can see that you live it with your whole body. And that is amazing. Thank you so much for this. And we'll be here to continue the conversation very soon. Thank you so much. Thank you for having
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Common terminology status values include preferred, admitted, deprecated, and forbidden. In the context of...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Terminology decisions are business decisions and leaving them unmanaged can come at a cost. In this episode, I...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Terminology management doesn't happen in isolation. In this episode of The Language Worker, we look at the...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! In this episode of The Language Worker, I sit down with Sonia, a German and Italian native speaker and performance...