Welcome to Working With Languages. My name is Sonia Kampshoff, I’m your host and a multilingual digital marketing consultant.
In today’s episode I’m joined by Marina Koleva and Manuel Augustin from Yoast. If like me, you’ve ever worked on a website built on WordPress, you‘ve probably come across their work. Yoast is the company behind the hugely popular SEO plugin, helping over 13 million websites improve their visibility in search engines.
The plugin is best known for its content analysis for readability, keyword optimization, and now even AI powered suggestions that make writing easier. Their mission is to make SEO accessible for everyone, from bloggers to large organizations. As linguists and developers, Marina and Manuel bring a unique perspective blending language and technology to shape how the plugin works.
I’m excited to dive into this conversation.
Sonia: in today’s episode of Working With Languages we have not only one guest, but two. We have Marina Koleva, a linguist and developer at Yoast. Hello Marina.
Marina: Hello
Sonia: and Manuel Augustin also working at Yoast. He’s the product owner and plugin release lead. Hello, Manuel.
Manuel: Hello.
Sonia: It’s very nice to have you both here. I’m very excited about this episode. So can I ask with the first question to both of you: what is your favorite word in a language that you speak?
Marina: Yes, sure. So I’m Bulgarian and Bulgarian is my native language. And since I was little, my most favorite word is the expression for, I won’t, I’m not gonna do it, which is няма.
So if, if someone says, could you please close the window, you can say няма.
Sonia: Short and concise, to the point. Manuel, what is your favorite word?
Manuel: So I thought of two words which are very similar in two different languages also. It’s Spanish sobremesa and Dutch uitbuiken. So both of these refer to kind of the feeling you have after a meal.
Sobremesa is more about the about the act of just having a nice conversation after a nice long meal, whereas uitbuiken, and that refers more to kind of the feeling when your belly’s really full and you’re sitting there just, also enjoying the feeling after a meal. So I thought both of these words kind of evoke the same nice feeling and I think it’s nice when workds really evoke a feeling. So that’s why pretty, pretty quickly I came up with those two. Yeah.
Sonia: Very nice. It sounds like you’re not only a linguist and developer, but you’re also a foodie.
Manuel: Yes, exactly. You have to feel, feed the body as well as the spirit.
Sonia: Right. That’s true. That’s true. Marina, so you mentioned you are from Bulgaria.
Can you tell us a little bit more about your background, where you learned the languages you speak and so on?
Marina: Yeah, so I don’t actually speak that many languages at all. But I have studied quite a few. And this is actually what made me want to study linguistics. So I’m from Bulgaria, in school I studied a few European languages, and then I did my bachelor’s in Japanese.
I also did an unsuccessful bachelor that I didn’t finish that included Greek and Turkish. And then right now I live in the Netherlands, so I’ve picked up a bit. But yeah, it’s always fun for me to. Compare languages that are similar. And then it’s also just always exciting to learn a language that follows a completely different structure and logic.
Sonia: Yeah, it’s interesting that you also learn Japanese and Greekwhich have a different writing as well. So that also gives you even more insight of how a language works.
Marina: Right. And Bulgarian actually also uses cyrillic, which is…
Sonia: That’s true, that’s true
Marina: …close to, to the Latin alphabet, but not, not completely. And to the Greek one, of course.
Sonia: Excellent. Manuel, what about you?
Manuel: Right. So about languages, well, my native language is German and I have lived in the Netherlands for quite a long time for I think almost 15 years now. And German, Dutch are very similar, so I also speak Dutch. And so those are languages that kind of, well, and of course English as we all need to speak it.
So those are languages I usually work in, that I use on a daily basis. otherwise, I’m actually also, I study Japanese, not as a major, but let’s say more of a, as a hobby. And recently I also got into studying Korean. So I think I also have a thing for east Asian languages.
Sonia: Interesting. How did you both get to work at Yoast? What’s your career path? What brought you there?
Manuel: I came to Yoast first and that kind of also touches on the history of working with languages at Yoast. So I started as Yoast as the second, let’s say linguist working there. So Yoast is doing a lot with language analysis. So at some point they figured out that they need a linguist and they hired the first one who was actually a friend of mine we had studied together, Irene Strikkers, I’ll send it to her. So shout out to her because I wouldn’t be there without her.
And yeah, so she was the first one and when they were, they realized, oh, they might actually need that second person who is linguistically versed and also is willing to get into the software parts.
Marina: So actually. I thought of applying to Yoast a few years before I actually did, but I didn’t have the confidence to do it. And then at one point I just had to find a job, so I thought I’m gonna apply whatever happens, happens. And then I remember very well the application process because a part of an interview there was also an assignment, and the assignment included a way to flag or to tag suffixes in Turkish. And because I had studied Turkish before. I couldn’t speak it at all, but I was familiar with the grammar. So this helped me a lot with not just doing the assignment but having the confidence that, oh yeah, this is something that I can do.
I get the logic of it. So this is one of the main examples in my life where I’ve learned that with languages, even knowing a bit is actually very useful. You don’t have to be a master of the language to be able to use it. And, yeah, the other thing is that I joined when we started the project of introducing Japanese support.
And I’m not sure if that was the reason why I was chosen, but maybe it was a, a consideration there as well.
Manuel: Yeah, I didn’t, I don’t remember. I didn’t remember that we gave you such a mean practical assignment for the job into your Turkish suffix as well. So, yeah, mean, I think the fact that you made it through that proves that you’re a, a very capable linguist and also have a knack for software.
Marina: Thank you, Manuel. Oh, that’s very kind.
Sonia: So it sounds like it’s of advantage at yours to speak many languages, even though maybe not at perfect level each of them, but just to have an understanding of as not as many, but many languages. That’s great.
Marina: Absolutely. Also because when we start working on the support for new language obviously all effort is concentrated there, but then you want to continue the support, right?
So you find that a certain path can be improved and all that. So I’ve had to pick up tasks connected to Farsi, for instance, or Arabic, which are languages I’m completely unfamiliar with. I’m not used to just the direction right to left is already weird for me. The fact that the cursor changes direction is weird.
But yeah, I think the more used you are to being uncomfortable kind of in facing a new language, a new system, the better.
Marina: That’s a very good way of putting it. What’d you do in your day job, both of you?
Manuel: I am the product owner for the Yoast plugins. Yoast has a plugin for WordPress, which is our main product.
So this, this product does a lot of stuff for SEO. So it helps you write better contents, but it also does a lot of technical stuff. So I used to be working mainly on the kind of content analysis side where Marina is also working on. But I’ve also branched out to do everything that has to do with the product management of the Yoast SEO plugin.
Sonia: SEO, or search engine optimisation, is the practice of making your website more visible to search engines like Google, so that more people find it when they search online for things related to your website. The goal is to get your website to appear higher in the organic (non-paid) search results, so that people click on it and visit your website.
Marina: So for me it’s mainly about thinking what kind of features we would like to add next, for example, things like that.
Sonia: And for people who are not familiar with what a, what an SEO plugin should look like. What are the features, for example, that you added recently?
Manuel: One of the big topics, of course, for everyone everywhere has been AI. So maybe Marina, do you want to take that one since you’ve been more hands-on with all the AI things we’ve been doing?
Marina: Yes, sure. A big feature that we’ve worked on for the last two years already, I think, or a year and a half, has been, we call it AI optimize. It’s because it helps you to optimize your content with AI.
The idea is that we have certain checks that are automatic, that are, that run on scripts, basically that check, for instance, does this word appear in this text? And then if not, if we expect it to appear and it doesn’t, for instance, a keyword should appear in the title. And if it doesn’t, then we just flag it and we say: fix it. That’s our recommendation.
But then with AI, we can actually recommend, we can suggest in what ways you can fix it. So for instance, for a title, we can recommend five different titles, five suggestions that you can choose from. So you can of course also do this with your AI of choice, right.
But it’s different when you can do it within the, the editor, within the interface of the plugin itself. So this is what we’ve been working on, offering this kind of AI support for different checks and also expanding this feature to work in different editors because in WordPress you have different editors and people get very attached to their editors. So offering it in just one is not sufficient.
Sonia: How are these five suggestions generated? How do you make sure that they don’t get repeated in various different places?
Marina: Right. So this is basically something that our team works on a lot and it’s mostly we depend on prompts engineering. Yeah, so writing prompts that work, testing them extensively because with LLMs you get a result that is not consistent always, right?
So you need to figure out whether in like 90 or 95% of the cases for instance, the title fits within the length you want it to be, but then in terms of, even in terms of, for instance, variety of the sentences, even though the, the LLMs are pretty good at it, you should, at least in our experience, you still have to ask for it so that you don’t get the same or very similar suggestions. Yeah, there are a lot of underwater rocks when dealing with LLMs for sure.
Sonia: Very interesting. I read your short bio on the, your website and the first paragraph, I loved it. The first paragraph was a little bit about you, where you come from, a couple of jobs that you had before Yoast. And then the second paragraph you just went straight to into, this was a very long paragraph. Did you like the length or would you have found it easier to read it in smaller sentences and so on?
And I thought that was perfect for the work that you do at Yoast because it is easier to read shorter sentences. So I know that one of the things the plugin does is also suggests: the sentence very long. Can you make it shorter?
And I think that fits also well into what you’re saying of how do you write, how do you make sure that the copy that you’re writing. It says what you want to say, but it expresses, you know, the sentiment and, um, the technicality of what, what you want to say, but also making it easier for the reader to read it.
Marina: Yeah, and also a point that I have to mention here is the fact that we always recommend our users to make sure that they check and they edit the text, the suggestions by AI, because we still think that the originality is very important in online content. You can, you can often notice whether content is generated by AI.
So even, for instance, with sentence like what you’re saying, we can help, like AI can help people shorten their sentences, but still you need the, you need a human eye to go over them as well to make sure that it’s not just a short sentence, but it also fits with the style, with your tone and all that.
Sonia: How many languages do you work with? With a plugin?
Manuel: I think for the part where it’s analyzed languages, it’s basically 22. So that’s the languages through we have optimized the language analysis and that has always been a very intensive process because, well, it’s for every language that we supported, we basically very lovingly handcrafted the analysis and did research.
So we also employed native speaker consultants, for example, to, as you mentioned, for example, sentence length. It’s a bit different also what, for example, writing manuals recommend for the sentence length and different languages. So then we do not a lot of languages, but for example, to look into the specific recommendations for Arabic and things like that, you would need a, a person who speaks a language.
So we’ve worked with consultants to help us for every language to develop these. It’s been a very interesting process because of course, even though you might have a native speaker as a consultant, you need to know what kind of questions to ask them to get the information that you need.
So I think that’s where, that’s where the point Marina made also comes in that, you don’t need to speak all the languages, but you need to have this kind of mindset to know what you’re looking for.
Sonia: Perfect. And I’m guessing English is the most popular language, as in terms of downloads, which are the languages are the most popular ones?
Maneul: Ah, that’s a good question. I think basically the bigger European languages are among the most used ones. So, for example, and well then especially the ones that are also spoken in South America, so like Spanish and Portuguese, but also German, and I think Dutch because we’re of course as a Dutch company by origin, we have a bit of an advantage in that in the home markets by, just by the fact that, that we’re most well known there.
Sonia: And if you were to add a new language, how would you go about it?
Manuel: So one thing of course you can look is how many people speak a language, but that’s not always rips. But since, since we’re really looking for text optimization on the internet, when we chose new languages, we looke at how much content is written in a specific language on the internet. So there are statistics about that.
For example, how many, what percentage of websites is written in a certain language. And also of course in which countries, the WordPress, the CMS or plugin is working for is most popular. So you get some discrepancies. For example, you might think because some languages, like for example, I dunno, Hindi spoken very widely, but in that specific market, let’s say, English is much more used on the web.
So not every big language also has the same representation in the written web content, for example.
Sonia: And what was the last language that you added? Do remember?
Manuel: Marina, what was the last do remember? Do you remember?
Marina: I do. The last one was Japanese.
Manuel: Oh, yes.
Sonia: Amazing.
Marina: That was a difficult one. It, yeah, it, it presented challenges.We didn’t consider maybe at, at first, for example, the fact that for all of the other languages that we supported before that, and not all of them were Indoeuropean sentences, were were separated by a full stop. Then either a space or a space and capital letter. So then you can very easily distinguish where one sentence ends and another begins.
With Japanese that was more of an issue, but another one, which was actually much bigger was the distinction between, words usually have spaces between them, right? In Japanese and in Chinese, so how do you distinguish between where one word stops and another begins? That was, yeah. I don’t think we’ve had this issue with any other language, and…
Sonia: so can you have words in Japanese with more than one character?
Marina: Yeah. Yes.
Sonia: and how do they show that a sentence has come to an end if they don’t use a full stop?
Marina: so basically with that, we decided that you do use a full stop. Let me correct myself a bit. You do use a full stop in Japanese, but in online content, very often they skip the final preposition.
And so they just start a new line. So you would just see every new line has to be a new sentence.
Sonia: Okay. So even. Say in an e-commerce website, in a product description where it could be a couple of paragraphs, do you have the text going to the next line a lot more frequently? Is that what you say?
Marina: I think it’s much more common in blogging. But generally you would have lines which are not as long in Japanese.
Sonia: Side note, it’s on full stops. One reason Yoast looks at where sentences end is because clear sentence boundaries make text easier to read when readers and search engines can quickly understand where one thought stops and another one begins, your content feels more structured and user friendly, which is great for SEO.
Sonia: is there anything else you want to cover about what you do in your daily job at. Well, we’ve given you a bit of a taste of what we do language wise. I think it’s also, well, maybe it’s a fun anecdote.
I think we have some colleagues who also work on the software part, but not the language part, who, when confronted with parts of the code base that, for example, use a lot of linguistic jargon, for example, I don’t know, to detect passive voice we need to detect the past participles. So oftentimes colleagues who do programming but not language work are very puzzled and will say, uh, but I’m not a linguist.
Sonia: Just a quick side note before we continue. You will hear Marina and Manuel talk about writing in active voice. For example, Sonia eats the apple as opposed to the apple is eaten by Sonia, which is in passive voice. The reason is simple. Passive voice makes sentences longer and harder to follow while active voice is clearer and more engaging.
This means that if you use sentences in passive voice, search engines may view your content as less clear or useful, and that can hurt your SEO.
Manuel: So I think that’s quite unique to yours that you have as a software company that has a code base that is so heavily focused on language analysis, that it’s really well, you have to be a specialist of both language and, and programming to really understand what’s going on in some parts.
Sonia: So I understand you, you’re not part of the same team. Marina, you’re on the lingo team and Manuel you’re on the product team. Are there any other teams other than that somehow related to the language side of it?
Mnuel: strictly speaking to the language side, maybe not, but we have other teams with, who also do programming, but then they will be working in the same code base.
So they will come across the language parts and sometimes that can be quite tricky.
Sonia: and Manuel, you’ve been at Yoast for a few years, five or six years, if I’m correct. Have you seen a change in how you work, how you do things, and how you apply the linguistic side to the work you do?
Manuel : Yeah, I think our original analysis is, we have this analysis that gives you recommendations that is very much handcrafted because also technically speaking, it’s something that runs within your browser. It doesn’t send any data anywhere, so that was really, it’s very much linguistic work of coming up with rules.
I think recently as we just briefly touched upon, with AI you have access to these language models that you, that actually support a lot of languages out of the box. So you don’t have to, you do have to still test whether everything works with different languages the way you want it to. But I think you have to do a bit of less of the nitty gritty work of making sure that your software has been, when it comes to language, and analysis works for every language.
But yeah, as Marina said, then the downside is that AI, as you all know, doesn’t always give you exactly what you want, so you have to argue a lot with it.
So whereas in the past as a linguist, maybe I had to really rely on yourself for setting up, for coming up with a very precise rules. Now you have to, it’s a bit more fluid work of arguing with the language models to give you information that you want, which I think is the same.
Well, it’s true for a lot of with AI that you have these days.
Sonia: I find that AI as well in the last. One, two years that it’s more been about, it has also changed, it has become better and more accurate. I don’t rely on it much, on data kind of things, but in, in terms of idea generation, it’s become so much better.
And I can see how also for your work, it just evolves in a way with, so you learn how to use it and it also evolves on its own. So that’s very interesting. And I guess considering how some people have a fear of AI taking over jobs, you have really embraced it and you work together with it. Um, you develop together and you use it to your advantage by developing your products in tandem with how it develops itself.
Manuel: Yeah. I think it’s, uh, one of the challenging but also fun parts of the work that we’re doing that. The environment changes so quickly and, uh, we, well, we need to be, um, at the forefront, forefront of developments. So, uh, never a dull day.
Sonia: Great. So, um. You use a lot, AI at work, do you use it also in your private lives?
Marina: well, so we actually discussed this in the morning with Manuel and I, I was saying how it’s funny that it’s. Us that have joined you today because we are like representatives of the two camps. Manuel has, um, enthusiastically embraced the AI and its, um, all of its capabilities, and I’m more on the skeptical side.
More, more selective I would say. I am a bit concerned with, um, for instance, the environmental influence and consequences of using AI. And the fact that you can’t rely on it a hundred percent, so it gives you some results, but then you still have to check. Those results. So for, for me, if I can look something up in a search engine, I would prefer to do that rather than have some reliable but not completely reliable results from ChatGPT.
But at the same time, it’s part of our world now, and it definitely proves, as you say, like every six months I, I see and hear something that really impresses me about AI. Just this week actually, I had a, a colleague who presented this incredible prototype of a site that he’d built with AI for a day, which honestly blew my mind.
Yeah. So that, that’s where I am. I do use AI, but really, really selectively. And then Manuel…
Manuel: Well, yeah, I think I also mentioned, uh, before when I was talking about this to Marina, that probably she’s more skeptical because she has so much more hands-on experience with testing the AI generated results when we were developing new AI based, uh, features and products that she, she knows how, what the limitations are, much better than I do as a casual user.
I think, I think you have to be, um, aware of the limitations. So especially these days, I think if you want to look up facts, then it’s a bit of a, yeah, may, maybe, maybe not always 100% reliable, but, I think as you mentioned, Sonia, for a dear generation, it can be really nice in my, my, um, experience, for example.
Well. One, one example where I use it a lot recently I think is kind of looking at recipes saying, oh yeah, I want to have these in these things in my fridge and I want to make this kind of recipe. What would you recommend? And then it gives me something, and then I talk about, oh, could I substitute this or that?
So it’s a nice conversation part. In that sense, it kind of come up with good things. It, I think it helps that it’s a domain where I can kind of make a judgment call myself if what it says makes sense, because I think we all remember from the early days of ChatGPT when it would recommend people to cook with glue or whatever.
Yes. I think, I think, you can use it in to advantage with a good dose of common sense.
Sonia: Yeah. I agree. I agree. And I was probably a late adopter in, in the timings of AI in that I didn’t jump on the ship immediately, but only I would say a few months later, which is very late in terms of AI and how the, the speed and how it develops.
But I find that what made me change my mind is hearing examples of how people use it, and that intrigued me about different ways of doing something that I was already doing in my own way. So that is that is how I started. I’m sort of copying examples and sometimes the way, it was nice to test it, but it was not relevant to me and my life.
But then I, I just got other ideas of how I could use it otherwise. So that was very interesting and that’s really how I use it at the moment. Just testing, oh, what would it say in this case? Uh, or how could I do it this way? And I find that also between different AI tools, there’s quite a difference.
So I have my favorite one and the ones that I use less frequently, but every now and then I, I want to double check what they’re saying and I use the same prompt in two different tools and I get different results, which is also interesting.
Manuel: I think it’s also good to think about when to rely on AI and when maybe not, because I think, well, at the time of recording, I thinkChatGPT 5 just came out and you see a lot of memes at the moment making rounds about people asking ChatGPT 5 how many bees there are in blueberry.
And it consistently comes up with the wrong answer of three Bs because AI is notoriously bad at counting. And letters, which is something we’ve also encountered. So that’s why, I think our traditional, and of course at yes, we do a lot of, I think like counting words and counting letters.
So this, this, there’s still a big domain word, traditional rule-based string processing and programming has, uh, has a place. So we’re not about to replace everything with AI just yet.
Sonia: I didn’t know that AI just so bad at counting things.
Manuel: Yes, it is. So it’s one of the thing that does so such magical things in so many domains, but then it’s quite bad at counting.
Marina: But it is after all, the most famousAI models are language models. Right? Large language models. Yes. It makes sense. Yeah. It’s a humanities major, not a science major, so to speak. There are other models and neuro networks and all kinds of machine learning advances that are better at calculation, and maths, in general.
But we are just so used to using LLMs for everything while they have, you know, they’re very good at certain stuff and then just not that good at others.
Sonia: Yeah. Interesting, interesting. Thank you very much for your time and your conversations. I’m gonna add links to the show notes, but is there anything else that you would like to add before we finish?
Manuel: I would just like to say thank you very much for the opportunity. Well, if 10 years ago someone would’ve told me that I would once have the opportunity to talk to work with languages to start with because I was a linguistics major and at that time you don’t really think that there are, well, you have a very limited idea of what jobs therea are with, when it comes to working with languages to know that 10 years later I would not only work with languages, but even get to talk about it on a podcast that’s about working with languages. That’s just a really, really great opportunity. Wouldn’t have dreamed of it.
Marina: Seconded.
Sonia: Well, thank you again and bye-bye.
Marina: Thank you very much.
Manuel: Thank you very much.
Sonia: Thank you for listening to this week’s episode. If you liked it, please subscribe and share it with someone who may enjoy too. As a new podcast, we value your feedback, so please leave us a comment or email us at the email in the show notes.
