Podcast

Creating a successful international strategy with AI and localization

Over a series of 3 episodes, Benjamin and Matija talk about the role of AI and localization in creating a successful international strategy. Matija talks about building translation memory, and what the future of AI and Machine Translation is.
Listen to the Voices of Search podcast: Creating a successful international strategy with AI and localization
“I think the machines will far overcome the capacity of humans and not only reach it. So not only reading and writing and expressing ideas but also generating and creating new ideas, is something that the AI is going to be able to take over from the human race.”
Matija Kovač, Co-founder and CTO @Taia Translations
Matija Kovac

Matija Kovač

Co-founder / CTO @Taia
Matija Kovač is the co-founder and head of development at Taia Translations, a company that bridges the gap between language barriers with the help of Artificial Intelligence and machine learning.
Benjamin Shapiro: Founder & CEO of I Hear Everything and Host of the MarTech & Voices of Search podcasts

Benjamin Shapiro

Founder / CEO @I Hear Everything
Benjamin Shapiro is the Founder and CEO of I Hear Everything, a media company that connects brands and content creators with their target audience. He is a President and Managing Director of Benjshap LLC.

What you'll learn

  • How AI and localization can help your business scale rapidly in the global market.
  • How to improve your localization process and SEO to boost your presence in foreign markets.
  • What translation memory is and what its role is in the translation workflow.
  • How translation memory can help you stay consistent and improve your global content output.
  • How machine translation is becoming the ultimate translation tool.
  • What is the future of AI and machine translation, and how marketers can use this to their advantage.

Episode 1 Transcript

Benjamin:
Welcome to the Voices of Search podcast. Today, we’re going to talk about the importance of localization in your business strategy. Joining us is Matija Kovac, who is the co-founder and head of development at Taia Translations, which is a modern translation platform where they help companies translate their documents, websites, and other content with an AI-assisted human perfected translation platform. And today, Matija and I are going to discuss why localization is critical for international business.

And this podcast is sponsored by Deep Crawl as Voices of Search podcast listener. You know that SEO and your website help have never been a more important part of your company’s marketing mix. Maintaining high rankings on Google has a direct impact on revenue and can help you lower your customer acquisition costs. But content and keyword optimization are only part of the picture. The technical health of your website is a critical ranking factor. Look, we all know that optimizing site performance can be an arduous and time-consuming task. But with deep crawls, technical SEO, and website health platform, your team will have the analytics and automation solutions you need to track your website’s technical performance, improve page rankings and stay on the top of search results in Google. So be smart like the SEO teams at Adobe, eBay, Twitch, PayPal, Microsoft, and Canva, who monitor their site performance using Deep Crawl to ensure your site reaches its full revenue potential. Visit DeepCrawl.com, the number one platform for technical SEO.

But before we get to today’s interview, I want to tell you about a new show that my company is launching. It’s called the Revenue Generator podcast, as it turns out, as marketers are under attack. That’s right. The walls between marketing, sales, and customer success teams are all falling down. And unless something changes quickly, your CMO is going to be calling him or herself a CRO in no time. And that’s why we’re creating the Revenue Generator podcast. The Revenue Generator podcast tells how innovators of revenue generation orchestrate teams to deliver world-class customer experiences that integrate data, SAS people, and processes to expedite demand and increase revenue. The show is hosted by my good buddy Doug Bell, who is a 20-year technology veteran that has been instrumental in driving revenue growth and scaling marketing organizations across some of the world’s best-known brands and nimble startups. And in each episode of the Rev Jen Pod, you’ll hear how industry leaders integrate sales, marketing, product, and customer success into a single business unit with a common goal of optimizing their revenue cycle. So if you’re ready to join me and Doug Bell as a member of the revenue generation, search for a revenue generator in your podcast app or head over to Revgenpod.com, that’s a revenue generator in your podcast, or head over to Revgenpod.com.

Alright, here’s the first part of my conversation with Matija Kovac, the co-founder and Head of Development at Taia Translations. Matija, welcome to the Voices of Search podcast.

Matija:
Hi, Ben, thanks for having me.

Benjamin:
Excited to have you on the show and thank you in advance for allowing me to butcher the pronunciation of your name. The Slavic names are a little difficult for me to pronounce. Give it to me one time just so I could hear what it sounds like.

Matija:
So it’s Matija Kovac from Taia Translations.

Benjamin:
It sounds so much nicer when you say it, maybe you should be the host today.

Matija:
I’ve been told I have a radio voice, so let’s try.

Benjamin:
You do. But hair for a video you got great hair too.

Matija:
And hair for video. For audio.

Benjamin:
All right. Well, everybody, you’re going to have to check out Matija’s LinkedIn profile to see the epic hair and, that said, he is not just a well-dressed, well-groomed man. He’s also an expert on localization, which is hopefully relevant to SEO. Let’s talk a little bit about that… Matija, why is localization so critical for international business?

Matija:
Well, we’re starting to work more with companies that are in this stage of very fast growth and don’t have a localization process in place. And what we’re seeing is that we can help these companies grow their revenue and grow their user base much faster by getting into multiple different markets, by exploring different levels of localization, and achieving a much faster growth than they would just by growing in a single language market. So imagine you’re a company from Germany, let’s say. And if you’re only targeting your audience in German, you have a very narrow audience at some stage, so even though Germany is a very big economy with a lot of people, there are a lot of limits. But once you start exploring localization, start investing in it. You can start growing much faster across the board.

Benjamin:
You mentioned that there are levels of localization, and I think that’s an interesting way to frame it. I’m here in the United States and we have English, and most Americans think that English is the language that should be spoken pretty much everywhere and get frustrated when they go somewhere and they speak a different language, right, with a very sheltered life. But that said, there are a couple of different flavors of localization that maybe people here in the states don’t think about where we think of localization, and it’s the difference between English, as in American English. And maybe there’s some translation into Spanish, but really, that’s probably not something that we’re thinking about very much locally until we go to different countries. And then it’s well, alright. Let’s just translate the words that are on the page and slap-up another domain with a dot.com, dot-something, and hope we have a localized language, but really, it’s much more complicated than that. You’re from Slovenia, am I correct?

Matija:
Yes, I’m from Slovenia.

Benjamin:
It’s the home of Luka Doncic.

Matija:
Luka Doncic; one and only yeah.

Benjamin:
Probably mispronouncing his name as well. My favorite…

Matija:
Actually, it was very good.

Benjamin:
Thank you. I watched a lot of basketball, and in countries like yours, where generally the actual territory is smaller, but also your neighbors speak different languages. Becomes more of a problem. So talk to me about some of the levels of localization, not just international localization. You have to think about different languages in the same countries and also sometimes you’re using the same language, but you still have to localize it. Talk to me about those levels…

Matija:
Right, so as you mentioned, I come from Slovenia. It’s a very small place. There are only two million of us and we normally start learning a foreign language when we’re about four or five years old, maybe six. And most Slovenians would speak at least one or two foreign languages. And some of us, a lot of us, go into studying languages as well. Myself, I learned English when I was very little already in primary school, and then I studied German and then Italian. And then I went to the university and I learned Chinese and all of us speak a little bit of Croatian because that’s where we go for our holiday. That’s where we keep our sailing boats and our condos and stuff like that.

Benjamin:
One of my favorite places in the world, we went to Croatia for my honeymoon. Oh my god, it’s so beautiful.

Matija:
You see, I get to go there every week and that’s awesome because it’s such a small country. I can just drive for an hour and I’m in Croatia. But what it also brings with it is aside from all the cultural mixes and getting the best Italian coffee, but also the Wiener schnitzel from Austria, it also brings in a lot of issues when you’re trying to expand your markets. So already when I started my first company in 2014, very quickly we came to our limitations. When it comes to market size for only two million people, there are just so many things you can sell right? So at some stage, you start looking abroad and you start figuring out how to get into other markets, and localization is the only way to do it right. And as I mentioned earlier, there are multiple different levels of localization. So with the high growth of machine translation quality since probably since 2016, when Google introduced their first neural machine translation solution and Google Translate suddenly wasn’t a joke anymore. The machine translation industry has gone through the roof. It has seen tremendous growth, and the improvement in the quality of machine translation output is really good. So this is already your first level of localization. You can have everything you have translated very, very quickly and very efficiently, but with a lower quality. So it’s not going to be as good as if a human translated, but it might do the case specifically in SEO. So let’s say you are, you were running an e-commerce platform or your store, and you have, let’s say, 10000 different products in your shop. You don’t have the funds to translate all of these products into all of the languages you want to aim for. But what you can do is you can translate all of them using machine translation and get them on the local market in another country and see how they react and see how your SEO positions can help your growth and grow the business there. And once you know which products and which categories are interesting to the local market, that’s when you can start investing in proper human-assisted or AI-assisted but human-perfected translations. Then again, there are multiple levels of that.

Benjamin:
Let me poke some holes here. I understand the philosophy of… And we go through this with transcriptions of the podcast. We will do a 10 cent a minute transcription of our podcast that we give to our team internally that they use to pull out show notes and quotes. But we would never publish that piece of content. We would spend the dollar per minute to get a human-vetted transcription, right? And so there are different levels of cost depending on the amount of attention you want to pay to the content. The problem that I have so talk me through this is you’re saying, well, use the 10 cents a minute version of a translation, see where there is demand for your products and services, and then double back and decide to spend the dollar version to go through the translations. But how do you know if what is causing specific products to have better conversion rates is not necessarily the quality of the translation? Maybe somebody is getting to a page. The translation isn’t any good, so they’re saying this obviously isn’t a professional company. I’m not going to buy the products. How do you know the problems, the translation, or the product?

Matija:
Well, yeah, there’s obviously this limitation that you just mentioned. So you have to be very careful about not having huge bounce rates just because your translation quality is poor. What you can do is you can learn from your existing markets. So let’s go back to the German example we had before. You were a German company, you want to expand into France. Let’s say there’s not such a giant difference between these two markets, right? So the product that might sell well in Germany might also sell well in France. So you might start investing in those sorts of products earlier, but before you have to invest heavily into the localization and don’t be mistaken, it’s not a very low cost if you want to do it properly. Even with all the AI- assistance that we can offer, you can go ahead and have everything translated with just a machine translation because you will start building traffic and you will start getting your SEO rates higher than if you were to wait before you have the funds to be able to translate everything perfectly. But as you said earlier, it really depends on the level of engagement you’re expecting. So we would always recommend our clients to translate their home page with the translation, with editing and proofreading. So ATP service, that’s the highest quality that you can probably find. We have AI and three professional humans on top of that to make sure that it all sounds perfectly natural and no one in that local market is ever going to figure out that this is actually a translation. Whereas for some other uses, a proofreader might not be necessary for your legal content, but the advisor would be. So if you have your terms and conditions, for example, you’d want to make sure that those are properly translated, but you don’t have to sound very marketing dramatic, localized entirely for the local region. Whereas with some content, let’s say your blog posts and stuff like that, that’s mostly there to generate content and generate traffic and maybe inform the audience, but it doesn’t have such a high value in your overall web presence. In those cases, you probably would be perfectly fine with just machine translation that’s edited by a human to make sure that it’s actually correct. But you don’t need all of this huge investment. But it really depends on the client, their budgets and obviously their entire needs.

Benjamin:
I think of this as building a product. You have an MVP. You’re coming up with a baseline. You know that it’s not perfect, but it’s good enough to work. So I guess the question is, are the machine learning transcriptions good enough to drive conversions, gain a signal from your blog post, figure out where there is demand? Or do you actually have to integrate humans into that process?

Matija:
Exactly. It’s very similar to building a company and having an MVP first and then building on top of that wants to see some traction.

Benjamin:
OK, so talk to me a little bit about the strategy of localization you mentioned. Start with the sort of MVP style, then go through the more resource-consuming process of building out true translations. But there are some other problems that you run into here, where you’re not just translating word for word in a language, you’re also dealing with cultural nuances. How do you figure out not only are the words resonating but are you writing what’s relevant for the specific culture?

Matija:
Yeah. So this is where collaborating with your translation team comes into place. So not even 10 years ago, localizing all of your company’s presence online was in the domain of companies who had huge assets who were able to build entire teams dedicated exclusively to localization. You would usually have a localization manager and then 10, maybe even 30 or more people in the company whose sole job is to make sure that their content gets translated into local markets and that it’s translated with the company voice, and that they follow certain guidelines and all of that. There was a lot of manual back and forth and emails sending up and down and all that. But with platforms like what we’re building, what we have here at Taia, things are getting much more simplified. So basically our goal is to be able to allow anyone inside your company, be it a marketing member, a marketing team member or a legal team member, or sales or whoever needs to get something translated to have it translated easily and effectively. So they don’t have to go through a specific localization department. They don’t have to have any special skills or any training. It’s a platform that’s so easy to use that if you show it to your mom, she should be able to use it instantly because it’s just a few clicks away.

Benjamin:
That sometimes is more challenging than what you think it is. My mom is a textile artist and I built her website, BarbaraShapiro.com. And when she wants to launch a newsletter, I still have to walk her through some stuff.

Matija:
Alright. Well, I’ll be curious if she ever wants to translate anything into us as is Spanish for Latin America.

Benjamin:
Actually, oddly enough, my mom is a language expert in addition to being a textile artist and translates textile books from English to French and French to English, but nevertheless enough of the Barbara Shapiro show. I think what you’re saying is that the cost for localization has gone down dramatically. So, you know, help me benchmark what…

Matija:
Not only the costs but the process itself as through the use of technology and through the use of not an only neural translation but also other text, it’s become much more accessible. Anyone in your company can start getting content translated now, and they don’t have to be a very educated person around this topic.

Benjamin:
So help me understand the differences where it projects on. You mentioned an e-commerce website with 10000 pages five years ago. What was the cost and timing to translate that content for a new market then, and then what the comparison to what it is now.

Matija:
On the one hand, depends on how you get your content across. So five to 10 years ago, APIs were not as common as they are now, so having stuff integrated and interconnected was a bit more complex and nowadays it’s much simpler, right? So this is the first major threshold where you can get your content across to your localization provider of choice in a much easier fashion than you would five or 10 years ago. So when we were just starting out with this company, most of the market was… About 99 percent of the market was still dominated by traditional language service providers or translation agencies, or however, you want to call them. And these guys usually didn’t have such high levels of technical skills or be able to integrate with their many products or be able to translate as fast using technology that’s out there. So we set out to change all of that, and we might not be the only company doing this sort of thing, but we do have some very special things that we’re pretty excited about. So nowadays what happens is that you would probably integrate using an API and get all the products across as they’re being input into your platform. So as soon as you launch a new product in your German shop, it’s already going to be halfway translated or even fully machine translated and available to your French audience, so you’re not wasting any time when launching a new product. That’s the first really important thing. And on top of that, you really reduce the amount of human workload necessary to get your content across. So that’s the first thing we solved. The second thing that you can do here is, as I mentioned earlier, you can have everything pre-translated with a machine translation. And once our system learns from your existing content and once you build up your translation memory, there’s a high chance that this machine translation is going to be very high quality and might not even need human interference at any point. But once you have all this content already in the platform and it’s already pre-translated with the machine translation, it’s much easier for human translators to just go ahead and added, postdated the machine translation, and have everything done much more efficiently than if they were translating by hand from scratch and typing everything down right.

Benjamin:
I understand that there’s a technical change where it’s easier to share your content. You’re taking advantage of some of the advanced technologies today. I’m still going to hold your feet to the fire. If it used to cost me $100,000 to translate 10,000 pages and I’m going from English to French and Spanish. What’s the cost now?

Matija:
It really depends on the quality of service you’re expecting, so it can be either one penny for word or it can be up to, let’s say, 20 pennies per word, depending on how much human interaction you require. And how much of a high quality you expect, but this is now. Five or 10 years without any AI assistance and without any fancy integrations, it could easily have come up even to, let’s say, 35 or 50 pennies per word and two that can sum up to quite high costs.

Benjamin:
So what I’m hearing is the translations have not only got easier from a technical perspective, which makes it more efficient for you to implement translations as you’re building your content. But also, the cost of translation has gone down because we’re able to leverage technologies that we’re talking about going from 35 to 50 cents to 10 to 20. So more than a 50 percent drop in your overall costs over the last five to 10 years for translations.

Matija:
Perhaps something like that, but the most important part is that it’s much, much faster. So if it took a week to get something translated five years ago, it’s usually up to 48 hours now. It’s much more efficient.

Benjamin:
And time is money, so again, more efficient. Alright, that wraps up this episode of the Voices of Search podcast. Thanks for listening to my conversation with Matija Kovac, the co-founder and head of development at Taia Translations. Join us again tomorrow when we publish our second part of this conversation. When Matija and I continue our conversations talking about building translation memory. If you can’t wait until our next episode and you’d like to learn more about Matija, you can find a link to his LinkedIn profile in our show notes. Or you could visit his company’s website, which is Taia.io., and also a special thanks to Deep Crawl for sponsoring this podcast. It’s time for you to be smart, like the SWAT teams at Adobe, eBay, Twitch, PayPal, Microsoft, and Canva, who use Deep Crawl to monitor their site performance. To ensure that your site reaches its full revenue potential. Visit Deepcrawl.com that’s deepcrawl.com for the number one platform for technical SEO. And don’t forget to check out our newest show, the Revenue Generator podcast, which tells how innovators of the revenue generation orchestrate teams that deliver world-class customer experiences through the integration of data, SaaS, people, and processes to expedite demand and increase revenue. So if you’re ready to join the revenue generation, search for revenue generator in your podcast app or head over to revgenpod.com, that search for revenue generator in your podcast app, or head over to Revgenpod.com. Just one more link in our show notes I’d like to tell you about if you didn’t have a chance to take notes while you were listening to this podcast, head over to voices of Search.com, where we have summaries of all of our episodes and contact information for our guests. You can also send us your topics, suggestions, or your SEO questions, and you can even apply to be a guest speaker on the Voices of Search podcast. Of course, you can always reach out on social media. Our handle is voices of search on Twitter, and my personal handle is Ben J. Sharp. And if you haven’t subscribed yet and you want a daily stream of SEO and content marketing insights in your podcast feed, we’re going to publish an episode every day during the workweek. So hit the Subscribe button and your podcast app and we’ll be back in your feed in the next business day. All right, that’s it for today, but until next time, remember, the answers are always in the data.

Episode 2 Transcript

Benjamin:
Yesterday, Matija and I talked about why localization is critical for international business, and today we’re going to continue the conversation and talk about building translation memory. Matija welcome back to the Voice of the Search Podcast.

Matija:
Hi, Ben. Thanks for having me.

Benjamin:
I’m excited to have you back on the show and continue our conversation. Yesterday, we talked about localization and how changes in technology have made it easier, faster, and more affordable to translate your content for different markets. And one of the things that you mentioned is that there’s this notion of not only being able to feed your data as you were writing it once it’s been published into a translation service, but also that the technology that we’re building can use artificial intelligence natural language processing to basically do the first round of translations before a human actually has to view them. One thing you said was you could start to build translation memory. So talk to me a little bit about what translation memory is and how does it work with making sure that you can internationalize your business?

Matija:
So translation memory is actually a very rudimentary technology that’s been around since the 90s or maybe even earlier. But it’s been it’s seen a lot of improvement in recent years, specifically with graph-based databases and similar technologies that have allowed for a much more extensive search through bigger piles of data. And what it is basically is just a database of existing translations that you’ve already built through your previous projects. So what you do is, for example, imagine you have a word document, you’re going to split it into multiple smaller segments like your header, your subheader, paragraph one, paragraph two, and so on. And for each of these segments, you’re going to restore the original text on one side and then the translated text and the other side. And this is basically your translation memory.

Benjamin:
So what is the purpose of translation memory? I think of this as the machine learning algorithms are starting to understand what you’re writing and basically understand your business a little better to make your translations more accurate. Am I my thinking about this correctly?

Matija:
I think you’re already a step ahead, so a translation memory in itself is a little bit simpler. That means that, for example, you’re translating a piece of content that has some content that you’ve already translated in the past. So let’s say it’s a product description that’s very similar to another product that you’ve already translated yesterday. And it’s a part of its description is similar or exactly the same. The system is going to fetch your existing translation from the database and feed it back and help you to translate things more consistently across all of your documents, or even to avoid having to translate an exact same segment all over again. So what does in effect is that it’s going to allow our translators to work much faster, and it’s also going to allow us to reduce the number of words you’re paying for when you’re ordering a professional translation service because a translator doesn’t have to go through the whole ordeal of having a segment that’s already been translated in the past, re-translated over and over again.

Benjamin:
It’s like this problem that every startup founder has where they decide they’re going to call their company one thing on day one when they need to write a description for their website or social media platforms. And then when they go to fill out another application or hire somebody, write a job description. They don’t remember what they called themselves. So they start rebranding the company because there’s no consistency across the copy that they’ve written. So using translation memory to basically make things consistent, describe yourself in a similar way. I sort of articulated a more advanced use case, which is natural language processing is good enough. Work understands the words you use and how they’re being translated. Talk to me about how you build rapport with the machines that are doing the translations, how they get to know you better to make sure that the translations sound like they are the words you would use your voice, your tone and et cetera.

Matija:
So that’s exactly where a translation of memory comes in. But it’s the first step in being able to build a machine translation solution that’s going to be able to learn from your existing content. It’s not something you can always afford. You would probably need about half a million segment pairs to be able to train an engine with any effect at all.

Benjamin:
You said segment pairs. What’s a segment pair.

Matija:
All right, sorry, so I might get too technical from time to time.

Benjamin:
It’s an SEO podcast. There’s no such thing as being too technical.

Matija:
So there are multiple ways you can name this. Even in the industry, there’s not usually a very straightforward lingo on this, but it’s as I mentioned earlier, you have segments or patterns or paragraphs or just sentences, or sometimes just a single word. This can be a segment right in the translation industry, and it really depends on the source text where it comes from and how the system is set up when it’s segmenting the source text. But a segment pair would be when you have a translation pair for your existing segments. So let’s say you’re saying Benjamin’s a good guy in English, and then you’re translating this into, let’s say, French and Italian in other languages, and you’re going to have this same sentence translated in other languages as well. So you’re going to need a lot of this sort of content that’s very specific to your brand and your company in order to be able to build a machine translation solution that’s going to provide substantially different results from what’s already available out there on the market. But that said, you might be building content much faster than you imagine. And companies that are not building their translation from memory, are losing a lot of time and a lot of money already on their existing translations, but also on the chance of having an opportunity to eventually train a machine translation model just on their data or retrain it along with their data.

Benjamin:
So talk to me about the size and scope of the project when you start thinking about doing these sorts of what I called translation memory, but it’s really an open segment pairing or whatever you want to call it, multiple different ways. When you’re thinking about working with a translation company and building rapport with your natural language processing, machine learning algorithm technologies stack. It does sound complicated, but this is a complicated topic when you’re figuring out how much you need to translate to start to get the value out of the technology you’re losing. What’s the sort of baseline of like: we need a million words, we need a thousand words… How do you think about figuring out how to give enough data to make sure that the machines you’re using are translating and understanding your brand effectively?

Matija:
So as I mentioned earlier, you would need probably around half a million or more segments that are pre-translated. And that’s just for one language combination, right. So if you have multiple different language pairs, you’re going to need to gain as much and more. But it also depends on the length of your segments. So if you’re just translating microcopy from your UI, let’s say just CTAs and, you know, other short strings, it might not even affect that much because those are usually very common across the board. You probably don’t have unique CTAs on your buttons, and it doesn’t make so much sense to train an engine just for that. But if you have a lot of content that’s unique to your, let’s say, blog and you make a lot of money with your blog, then yeah, that case might even make sense to train in machine translation and do that. But in order to do that, you would first need to start building the translation memory. And not every localization provider out there is going to be doing that for you just by default. So you’re either going to have to agree on that in advance, sometimes pay extra to have your translation memory. A lot of times you’re going to have to deal with them to be even able to take your translation memory and take it away with you wherever you want to go. Even though it’s basically your data and essentially you’re the owner of it. So make sure it’s when you’re choosing your localization service provider or a platform or whatever you want to call it, that you find someone who’s going to promise you to keep your data safe, to also keep your data, not just throw it away after every project, and to make sure that its data is kept correctly cleaned out and reviewed and stored in a way that it can be reused eventually for other projects.

Benjamin:
My takeaway here is that it makes sense to think broadly when you’re doing your translation. It is not just finding a vendor who speaks the language in the local market. You need to think about the data flow from where you’re creating your content, how you’re getting it to a translation company so they can start to aggregate all of the data that you’re feeding them because it’s not just converting words from one language to another. You’re also potentially building data that could be useful for future translations, which end up in future markets which generate revenue. And that wraps up this episode of the Voices of Search podcast. Thanks for listening to my conversation with Maija Kovac, the co-founder and head of development at Taia Translations. Join us again tomorrow when Matija and I continue the conversation and chat about the future of AI and machine learning translations.

Episode 3 Transcript

Benjamin:
Today, we’re going to talk about the importance of localization in your business strategy. Joining us is Matija Kovac, who is the co-founder and Head of Development at Taia Translations, which is a modern translation platform where they help companies translate their documents, websites, and other content with an AI-assisted human perfected translation. So far this week, Matija and I have discussed why localization is critical for international business, and yesterday we talked about how to build translation memory. Today, we’re going to wrap up our conversation by talking about the future of AI and machine learning translations.

All right. Here’s the last part of my conversation with Matija Kovac, the co-founder and head of development at Taia Translations. Matija welcome back to the Voices of Search podcast.

Matija:
Hey, Ben! Thanks for having me.

Benjamin:
Always excited to have you back on the show. Excited to wrap up our conversation today. And so far this week, we’ve talked about the strategy of localization and yesterday the strategy of localization, how the costs are coming down, how we’re being more efficient with localization. Because technology not only can ingest your content, but it can start to use machine learning and natural language processing to do the translations and also build up what we’re calling translation memory to understand how your company wants its copy to look, sound, and feel in different languages. Talk to me about some of the next-gen things that marketers and CEOs should be looking for. What’s the future of AI and machine learning when it comes to translations?

Matija:
It’s complicated, that’s for sure, and it’s very, very interesting. There’s so much going on in this field, actually. Machine translation was one of the first areas where A.I. has actually shown a lot of promise, also in application in production environment applications. So in both 2016, when the first neural machine translation engines came out, a lot has changed in the industry and there are a lot of new developments going on where we can, you know, use this technology to make translations more effective. More efficient and much faster, as you already mentioned. But what’s also coming and we’ve seen a lot of it in the last two years specifically with GPT three and similar large-scale models, is not only translation but also entire transcreation or even creation of content from scratch. So all of these huge models have great potential for understanding human speech patterns and writing patterns. And once trained properly, they can build content instead of you not only translate it into other content, into other languages. And this really excites me. I think this is going to be a very, very interesting arena in the future.

Benjamin:
OK, so what I’m hearing is the process for content creation used to be someone takes out a pen. They put it on a piece of paper. They write down words and copy them and hand them to people. Became newspapers. We move to typewriters. We move to laptops. People are still writing and typing. And then, you know, maybe we’re depending on computers and natural language processing to translate the content for us, but we’re still reviewing it. Now, it seems like we’re on the verge of the script flipping. Do you have a bad writing metaphor where we’re going to say, OK, natural language processing, machine learning algorithms go write me content, and then the humans are vetting to see whether that content is correct? At what point do we hit the tipping point here? One of the humans going to be writing the content and the machines are editing it as opposed to the machines editing? The humans are going to be writing it. How far away are we?

Matija:
Machines are already editing your content. I mean, you’re probably using tools like Grammarly and similar who are helping your writing skills and help you to build better content. Then there was no affiliation.

Benjamin:
I wish, I love Grammarly so much. It was always the problem at the beginning of my career. People like Ben, you’re a smart guy, but you cannot write grammatically correctly. You cannot write anything with any sense of grammatical correctness. It was always my Achilles heel.

Matija:
And you’re a native English speaker. Imagine how it is like for us Europeans to have to learn the language. And you know, I spend most of my day writing and speaking in English, and it’s not my first language. It’s not what I speak with my wife at home, right?

Benjamin:
You’re still probably better at it than I am.

Matija:
But that said, editing of your content is already here and it’s been here for a while that it’s improving dramatically as well. I mean, all of these tools like grammar and similar are have helping us on a daily basis to produce better content and produce more content as well. But when you’re flipping it around when there’s, I don’t know, GPT three and similar stuff generating your content instead of you, just by providing it with some inputs, and then you’re just editing that content to make sure that it brings across the right value and the right meaning that you want to convey, then you can actually produce much more content and to reach a broader audience for that.

Benjamin:
But there have to be formats of content that the machines can create that humans can. You know, I think of opinion. I don’t understand how… Sure a product description… Computers can understand where the words that are most likely to derive the best conversion results and start to use them to describe the products. If I’m sitting here saying, all right, I’m going to speculate on what the future of machine learning and translations is. You know, a computer is going to be able to sit down and be like, let me think about that. Actually, here’s the direction I think we’re going. So where are the machine learning and translation engine going to be effective? What types of content and where humans are still going to be sitting down with their proverbial pen and paper?

Matija:
I think you’re quite right here. So stuff that’s simple to produce, like stuff that’s very repetitive and already exists out there. You said it yourself very well. A product description is something, an AI can probably generate much better than your marketing intern with zero experience and copywriting can write. Possibly, or maybe he’s a genius machine and they can do a much better job at it. But when it comes to exactly that, you know, the human brain is just incredible, and we’re still not close to mimicking it with AI we’re probably going to see something similar in the span of our lifetimes, hopefully. I’m really looking forward to seeing where we get, but that’s way too futuristic for this time and age right now.

Benjamin:
Hang on a second. In our lifetime, you’re saying that machines and their ability to write should hopefully potentially reach the capacity of humans.

Matija:
I think they will far overcome the capacity of humans and not only reach it. So not only reading and writing and expressing ideas but also generating and creating new ideas, is something that the AI is going to be able to take over from the human race. Possibly in the next six to eight years with the current development that we’re seeing, it’s already overtaking a lot of the human ability. Can you imagine, like 20 years ago, it was entirely futuristic to think that AI would be able to recognize faces. Today is used on a daily basis, I mean, your iPhone unlocks just by looking at it, right? And it’s something we take for granted these days, even though it’s extremely complex technology. And it was unprecedented and completely unimaginable that a computer would be able to generate a text that would be indistinguishable to a human brain if it’s written by a machine and not a human. So the so-called Turing tests are being broken day after day after day. There are Open-Source libraries. You can run on your computer in half an hour and be able to produce texts that look as if a human has written them, even though there wasn’t a human around when the whole content was generated. So I think we’re quite close to the tipping point where a lot of the content we’ll be seeing is going to be so good. And not only reading content, I’m talking to audio, video. All sorts of content is going to be generated by the AI and is going to be generated so well that there won’t be a way for you as a human to figure out that it’s not generated by AI.

Benjamin:
Oh boy. So do we just pack it in? Are marketers done and artificial intelligence is going to take over in terms of content production… Where it is this type of technology gets applied. And I guess what’s the end game here?

Matija:
I think the first stage after this will definitely be that humans will be able to do better things and live a more meaningful life. So instead of a… The translator used to had to type away in search things in the dictionary and do everything manually and just open up a word file or something and start typing away. And sure, it sounds kind of romantic, but it also gets dull, very fast specifically for translating content that you’ve just translated yesterday. And now we have to do over and over and retype the whole text again with just some minor changes, right? And with the technology that we already have now, translators have to do way less work. They have to review what the machine proposes might be the correct translation. Now, this might be a dangerous area where they can sometimes forget to fix something, even though they should. But it also allows for a much faster turnover when it comes to human professional translation services. And when it comes to content generation, who wouldn’t have liked someone to assist you when you’re… If you have your writer’s block and you can’t pass something, you know you have this idea in your head, but you don’t know how to, how to put it down in words, and how to make it sound properly. And if a machine is able to generate five different options for you, all of them saying a similar thing but in a different way, isn’t that a helpful tool that you would want to use? And doesn’t it help you build a much more meaningful work where you don’t have to go and write it yourself in five different ways and then decide which one sounds better? But you can have a machine that helps you figure out how to do this faster and more efficiently.

Benjamin:
I think there’s always this sense of, I don’t know, the excitement of what technology can do and how it’s going to be able to expedite and make our lives easier. There’s always a little bit of anxiety as well, right? The feeling that what we’re doing and how we provide value is going to decrease. At the end of the day, I think of things like the industrial revolution and the internet age and all the new technologies that have come along. Everybody was saying, well, the light bulbs here. So everybody’s… The telephone is here, so no one’s going to have a job anymore. We’re just going to be talking to each other on the phone.

Matija:
People who are turning on candles with the candles take, you know, all over London. And this sort of place back in the Victorian era all lost their jobs because of the electric light bulbs. But would you want to sacrifice a life without an electric light bulb to provide this sort of job?

Benjamin:
I think the takeaway here is that, yes, there is some pain in the industries that are specifically impacted. Honestly, the translation industry, people that are translators, if that technology starts to become automated, those jobs probably go away. But it doesn’t mean that the economy doesn’t get better. It doesn’t mean that the overall workforce gets stronger. It’s just the transition for the people that are directly impacted.

Matija:
So, Ben, you said that a lot of people might be losing their jobs because those in the translation industry because of all the automation and AI that’s coming in. Other people’s work. But what’s actually happening and this surprised a lot of the people in the industry is that due to the effectiveness of localization, because of it, of the machine translation and all the other tech is making localization much more available, much more companies are actually entering the localization arena, and they wouldn’t have been localizing their content if it weren’t for the drop in prices. So the demand for human translators is actually growing instead of decreasing because the demand for localization is increasing because the technology makes it more available. So there’s always a silver lining to this sort of debate when it comes to how it affects our lives.

Benjamin:
I think that we’re going to continue to see the integration of artificial intelligence and advanced technologies into our life, and I do think that in some cases it will affect people and will change the workforce or change the way that we think about work. At the end of the day, I’ve never seen a scenario where technology has been what led to a decrease in the workforce. It always seems to be something that produces change, but in the end, ends up being a net benefit to the world. And that wraps up this episode of the Voices of Search podcast. Thanks for listening to my conversation with Matija Kovac, the co-founder and head of development at Taia Translations.

Liked this content?

Get notified when we publish something similar.

* We don’t spam your email or share it with anyone!