Stop Using the Term Full Post-Editing with Jakub Absolon

SlatorPod #187 - Post-Edited Machine Translation with Jakub Absolon

Jakub Absolon, CEO of ASAP-translation.com, joins SlatorPod to share insights into the evolving landscape of AI in the language industry and the role of post-editing in the translation process.

Jakub acknowledged the success of the past two years but noted that this year has brought new challenges. To stay competitive, the LSP plans to expand its B2B client base beyond the Slovak and Czech markets and explore languages in Central and Eastern Europe.

The podcast delves into the concept of full post-editing, which Jakub criticizes, believing it often results in a near-human translation, making it difficult for translators to justify their role and potentially leading to a decline in the quality of work and a loss of skilled translators.

Jakub expresses skepticism about the complete replacement of human project management with AI, as project managers serve as crucial connectors between clients and translators.

Subscribe on YoutubeApple PodcastsSpotifyGoogle Podcasts, and elsewhere

Regarding the use of large language models like ChatGPT, Jakub mentions that he uses them for content creation, but it requires a good understanding of the topic to avoid hallucinations.

Jakub anticipates that the industry will continue to grow, with a greater emphasis on post-edited machine translation and content creation rather than traditional translation.

Transcript

Florian: Jakub is the CEO of ASAP-translation.com, an LSP based in Slovakia, headquartered there. Hi Jakub, and thanks for joining. Before we get started, just tell us a bit more about your professional background, linguistic background. You work at Uni as well and run of course ASAP-Translation. Just share that story and your journey to becoming Founder, CEO of ASAP-Translation.

Jakub: Originally, I wanted to study geography, but I finished as a teacher of English and physical education. So my Master’s thesis was about translation process in translation of professional sports text. Then I have been working as a prison officer where I’ve been teaching English, but not prisoners, but officers and I was also a translator for Slovak prison service. Then I switched career and worked for Heineken Slovakia. But as my colleague from Netherlands said, you switched a lot, but you are still behind the bars. Then I worked as an HR specialist responsible for succession planning and recruitment. That was a great school and skills I have learned I am still using as an entrepreneur. For example, creation of job profile based on competencies, behavior, even interviewing again based on competencies, mapping of job processes and so on. Then short stay in HR department of Emerson Slovakia and then we started to build our own company together with my wife and colleague Katarina. She’s responsible for all the production and I’m responsible for strategic planning, technology, marketing. I’m also teaching at university as an external lecturer. I teach subjects like IT for translators, CAT tools and project management and so on.

Florian: All right, so on the client side services so ASAP-Translation, tell us a bit more. What’s your kind of client focus, text type focus, content focus?

Jakub: ASAP-translation.com is a medium-sized translation company with 13 employees which has operating on the market for more than 20 years. We are based in Slovakia, as we have mentioned and we have a branch office in Prague, but as I mentioned, everything is managed from Slovakia. We are a family business managed by me and my wife Katarina. At the time of establishment of our company, there were mainly large foreign translation companies on the market and a number of one to maximum two persons firm. Our goal was to create a midsize company that provides comprehensive localization services and uses processes of large companies, but also wanted to attain the efficiency of a small firm. We have succeeded in this I hope, and in recent years we have been ranked fifth in the Slovak market. In our over 20 years of existence, we have completed more than 150,000 translation projects in more than 120 language combinations. In addition to translation, we also provide other language services like proofreading, voiceover, post-editing, transcription, localization interpretation, subtitling, terminology database creation, SEO translation, CAT tool implementation consultancy and others. So you can see everything because the market is very small. Regarding the client segmentation, half of our clients are businesses from Slovakia and Czech Republic. But regarding revenue, the more important is our work for LSPs, where we provide them with translation with combination of Central and Eastern European languages like Slovak, Czech, Polish, Hungarian, Ukrainian. Mostly these languages in combination with anything. Domain of translated text has changed significantly where we started with general administrative text, then user manuals become the biggest segment. Currently we can say that we do a lot of medical text, marketing, still technical text, but also legal text, and of course e-commerce. Still more and more we provide our clients with specialized services like search engine optimization, translation, transcription and subtitling. We are also involved in educational and awareness raising activities where we’ve been providing CAT training for many many years. And in cooperation with Department of Translation Studies at the University in Nitra, we have managed to bring subject project management this year. We are quite proud of this. At a time when CAT tool were not a translator standard tool, we lectured at various conferences to show translators that despite the required discounts for repetition and matches, the use of CAT is ultimately a time saver for translator, so that the translator earns more per hour than without the use of CAT tools. We would be happy to do the same with post-editing of machine translation.

Florian: We’re going to talk about post-editing a lot, but first let me ask you about some of the big kind of LSPs, MLVs, Super Agencies, whatever you call them that you work with. So have you observed kind of any shifts maybe over the past 10 years in how they outsource a company like yours?

Jakub: Most of the time we are at the very bottom of the supply chain, so we are quite familiar with working both for so called Super Agencies as well as for regional LSPs or MLVs operating between us and the Super Agency. Last year I listened a lot about making supply chain lean. Even I have listened to the theory that LSP will work without project management team because project management could be replaced by AI. I am quite a skeptic in this, definitely parts of the project management process could be supported by AI, but I understand project management like a local HR department of localization companies. So I think that there will be still some project management. It would be possible if the translation and localization tasks were smooth and easy, but they are often very complicated and project management is here to be a connector between client and translator. So no significant change, I think, between last 10 years.

Florian: I mean, fully replacing project management with AI, I don’t see that. Like, I wouldn’t know how that would work completely, but of course all these systems get a lot smoother and so some parts are kind of removed, but maybe not by AI, just by general software, right? Now how would you describe kind of the business mood, I don’t know, for lack of a better term right now in kind of European Union SME LSPs? What are some of the concerns that you have or some of your peers have and some of the opportunities that you’re thinking that are coming up?

Jakub: Last two years we were surprisingly very successful, for us and as I know, the same for our competitors. This year looks completely different and I must say I do not know why. So it’s still more difficult to make a profit because there are more and more small projects which are very demanding but do not bring a big profit. So it start to be very challenging to manage standard company with full time employees. We are ready to become even more effective, but at the same time we need to work hard on our marketing, especially on united EU market. We strongly believe in EU project and regarding our business-to-business clients, I mean, no LSP clients, we mostly operate on very small Slovak and Czech market, so I see opportunity to open new markets for B2B clients. And regarding LSPs, we currently provide our clients with Slovak and Czech language mostly, but we are able to manage other central and Eastern European languages, especially Hungarian, Polish, Ukrainian, Russian, so maybe this will be our direction.

Florian: Do you work mostly with, I mean, probably like 80, 90% with freelancers, I guess, and do you retain them in some shape or form?

Jakub: Yes, we do with freelancers, but in case of end clients from Slovak and Czech market, we call it foreign to foreign language. Yes. If we do not have project managers speaking with native language, they asked us, so we use some partner agency. For example, if there would be a demand from Japanese to, I don’t know, Portuguese, so we definitely use company, LSP where we know the process is the same as in our company. At the beginning we tried to use freelancers also for these combinations, but it was not easy and it brought many problems. So this is the standard way how we work.

SlatorPod – News, Analysis, Guests

The weekly language industry podcast. On Youtube, Apple Podcasts, Spotify, Google Podcasts, and all other major platforms.

SlatorPod – News, Analysis, Guests

Florian: Now you’ve been in post-editing, kind of thinking about post-editing, looking at post-editing, reviewing post-editing and also of course implementing in your own company for a long time. I mean, when we did our research, I think the first mention was like 2015. I remember in my LSP days it was kind of borderline gibberish. It was like rules-based statistics, so post-editing just didn’t make any sense, but obviously with 2015, 16 it kind of changed. But just in a nutshell, how has it evolved since you started looking at it?

Jakub: Yes, you are right. We were testing usage of machine translation from the very beginning. We focus on modern technologies and processes, whether it is in the translation process itself, where we were also among the first users of CAT tools in Slovakia and Czech Republic, but also in communication and marketing, so we are trying to use all the modern techniques and technology. We also tackle machine translation where we have tested about, let’s say, eight machine translation providers. We currently use only two of them and I must say that they are generic, not custom MT engines. Also we have tested also create our own custom MT engine. So now we use the two of them that best suit our needs in terms of language combinations and domains. We try to build custom models, as I have mentioned. However, it is not suitable for small and middle size LSP company, I believe, because we have many small clients in different domains, so it’s not easy to build something special for them. If the customer does not forbid usage of machine translation because it happens, it is forbidden to use it, as long as if it is not a problem for translator because it also happens. We always connect MT within a CAT tool. We have considered for many years machine translation as one of the translation resources complementing translation memory and terminology database. We have also devoted a lot of time, effort and money to the post-editing of machine translation, which we provide as a separate service for our clients. Our competitive advantage is that we used so called dynamic translation quality management. That is also why we are not certified on 17100 because we believed that there is not only one solution for our clients, which in practice means that we can provide different levels of translation quality from raw MT through post-editing or MTPE, as now people like to call it, through regular translation by professional translator to TEP System or ISO 17100 compliant translation, so we can do all this range for our clients. In terms of terminology, for post-editing we still use PEMT as a product, yes, for post-edited machine translation and MTPE for process as machine translation post-editing. It could be also as a kind of tribute to MT pioneers because in 1950s, 60s when the ALPAC report also uses the term post-edited machine translation. So dynamic quality management is also our huge advantage for e-commerce clients where there are huge volumes of translations and it is practically impossible for them to use regular translation or TEP process. Also, my PhD thesis was about post-editing of machine translation and we also published ebook Human Translator which deals specifically with the role of human translator in the age of machine translation. So we did a lot with post-editing and currently, yes, there are still higher demand for these services. But unfortunately for full post-editing, there is almost no demand for classic post-editing on the lower level of quality.

Florian: Let’s dive into this term full post-editing and you’ve been a vocal critic of it. You wrote some pieces. You actually have something on YouTube as well, so outline your criticism of the concept full post-editing in a bit more detail. What do you mean by that?

Jakub: Yes, so as you can see, we are not definitely against the technology.

Florian: No, it’s certainly not because that’s what on our side when we’re researching like initially, oh he’s a critic of post-editing and we dug deeper. No, he’s not a critic of post-editing, of full post-editing because the expert on post-editing, right?

Jakub: Yes, because we are surprised by the effort to promote the so called full post-editing where the desired result is qualitatively at the almost same level of like a human translation. The theory of existence of full post-editing and light post-editing was also reflected in 2017 in ISO standard where full post-editing is defined as a process of correcting machine translation output to obtain a product comparable to human translation. I’m reading this. The very fact that the standard uses the phrase or term comparable is almost grotesque to me because I believe that the standard should precisely define its subject and not just to set values that are open to interpretation. It’s something similar when somebody uses good enough quality. Yes, it’s the same. It’s funny for me. Unfortunately, the IT company’s marketing… lobby is very strong and raising this topic on conferences is not very popular. Instead, we hear more and more about the end of translators and amazing performance of NMT and LLMs. Yes, I am a great fan of NMT and LLM. It’s fantastic. I still can’t imagine how they are able to do it. The practice, however, is quite different. People still demand quality, but thanks to the full post-editing theories, they would like to buy a quality translation at half the price. The client then believes that after all, the MT engine has actually already done everything and all translators need to do is find few errors made by robot. It’s like putting a needle in the haystack and pay the needle seeker or needle hunter by the number of needles found. MT has clearly brought more efficiency to translation process and it is more than 100% certain that there will be soon no translation without MT or without pre-translating, just as it happened with translation memory. However, a human will still be needed to check and correct the translation if necessary. It is the human translator or otherwise known as a post-editor, who is ultimately responsible for the translation, responsibility is the key. If you use one of the available machine translation engine on the internet, for example Google or DeepL, you must first agree that MT engine, it means the company that runs the service, is not responsible for the translation, yes? So regarding post-editing, if you think that the translator who reads text at the speed of light cannot make a mistake, I think that you are very wrong. So that’s why PE service, post-editing service should also have limited liability, up to me. If the translator is to be fully responsible for translation, he must be given time to do so, so the translator can do not only revision, but also research and decision making. Thus, if we demand 100% quality from the translator at the level of his maximum ability, up to me, it’s a human translation, not post-editing, even if the translator used MT or LLM in the translation process.

Florian: That’s an interesting theory. Let me just try to unpack this a bit, so what you’re saying is machine translation in this sense is basically a productivity enhancement? But if you’re telling me to do full post-editing, what you’re essentially demanding is a human translation and I don’t really care if I got to this human translation using no technology, using translation memory, or editing something that a sophisticated AI outputted, it doesn’t really matter. If you’re demanding from me full post-editing, you’re demanding complete human translation, and it doesn’t really matter what led to this end result.

Jakub: Yes, and clients don’t care what I’m using, maybe I’m using some books or something, but they don’t care. I’m not talking about the prices, I’m talking about the process and I believe that the human translation is the same as the full post-editing. And by the way, we are using as I have mentioned, our translators are using MT for many years, so it’s nothing new. Only what has changed that we are given pre-translated text by client. What is even worse for many of the translators, because they have to edit it and they ask us for half of the price, yes. Okay, let’s talk about the prices, but do not talk about the full post-editing. Yes, so that’s my theory, because I believe that professional translator use or should use MT only as a tool. But translator is still a process owner, so it’s up to his decision or her decision, what is he or she is going to use. In contrast with the ordinary user of MT, where the MT serves as a tool for basic understanding. Yes, for example, I’m on my holiday in Spain, I don’t understand Spanish, I see the sign, I turn my smartphone on, scan the sign and I found out, for example, the bus stop is temporarily out of service, so that’s absolutely different usage of MT. A professional translator, I believe must never get to the level of early mentioned consumer and must not use MT to understand the translated text. Under no circumstances should translator use MT to translate a text that would not otherwise understand or unfamiliar topic. Yes. Doing so, translator risk fatal translation error. Instead, translator uses MT just to increase the speed of translation and together with online resources and translation memories, MT allows him to obtain wider range of possible solution of translation problem. Yes, and I believe that quality comes hand in hand with guarantee or liability. Therefore, regarding quality assurance, I believe there are four types of translation services. First, raw MT where there is no guarantee for quality, no liability. The second, post-editing or MTPE with limited liability because we expect there are still some minor errors in the post-edited text. But there is also a significant risk that post editor can make a mistake because of high speed of the process. And the third, it’s a regular translation by professional translator with full guarantees. But this service is still not suitable for high risk or high visibility texts like, for example, expensive prints or manuals where the error can cause some serious injuries or even deaths of user. Yes, and the last type of the translation is TEP process or 17100 translation with full guarantees. So the risk management which we call dynamic translation quality management is necessary before we choose proper quality assurance for our translation. That’s why we have problem to accept definition of full post-editing and light post-editing. In full post-editing, they want you to be fully liable for your translation and that is why we believe full post-editing belong to human translation as I have mentioned. Also it is assisted by machine translation. So now let’s have a look on post-editing, how it is defined in ISO standards. In ISO 18587 and 17100, both define post-editing as edit and correct machine translation output. However, this definition does not apply to situation where the translator sees and uses the suggestion of machine translation tool that is a part of computer-aided translation, so it is omitted from the standard. This is not a post-editing. I don’t know why. Then there is a special standard for post-editing, 18587, where full post-editing is a process of post-editing to obtain product comparable to product obtained by human translation. And light post-editing is the process of post-editing to obtain a merely comprehensible text without any attempt to produce a product comparable to a product obtained by human translation. It was read from the standard. But in my opinion, there are other variables that distinguish human translation from post-editing and they need to be included in the definition, so I’m using this definition. But first we should ask question why did post-editing come into existence? Because we wanted to reduce what? Translation costs and delivery time. And for this reason we define post-editing as follows: post-editing is a method of rapid translation. Yes, so rapid translation in one important thing using automated machine or machine translation with subsequent checking and correction by a translator. So still there is a translator, we can call a post editor. So that the quality of a final product reaches the desired level and this desired level is also very important because as I have mentioned, I hate the so-called good enough quality because I believe that you must define the quality. Yes, it must be defined somewhere. And to meet the speed requirement, the following assumption are necessary. Post editor must be familiar with topic and terminology of translated text and must understand process that the text is describing. And also post editor must be proficient in typing and advanced editing techniques to be able to work fast and also familiar with software or application. Usually it’s CAT tool where the post-editing is done and to meet the quality requirements we need the following prerequisite of… Yes. Raw MT output can only be automatically copied into a translation if there is a presumption that the vast majority of MT translated segment achieve the following quality. Either they are applicable without modifications, so absolutely perfect pre-translated text or applicable with minimum modification, so almost perfect translation. Otherwise you are losing time as a translator and it happened many times that still the translation is not so good.

SlatorCon Remote June 2024 | $ 180

SlatorCon Remote June 2024 | $ 180

A rich online conference which brings together our research and network of industry leaders.

Buy Tickets

Register Now

Florian: What do you do then? I totally agree, right? I mean, if the translation is close to perfect or, well, perfect, whatever that means, right? I don’t see an error, fine, I’m faster. But if it’s not, then I’m reading, losing time reading, then I’m starting to edit and I’m like, well, I can’t really edit this, so probably in the end I’ll just delete the whole thing and start from scratch.

Jakub: Various problems that many studies are made on students or amateur translators or I don’t know who. So for them it’s absolutely perfect and they are 10 times faster with MT. But when you imagine translator who is working 15 years every day, eight to 10 hours, he is really very fast, such a translator. And for example, I am using MT very often, but if I am using it for translation just for me or for just information purposes, it’s absolutely perfect. I do almost nothing with it but when it is some material that will be visible, yes, so I am editing almost every sentence, so that’s the difference.

Florian: How good is it for Slovak? Because my benchmark is always English-German and obviously that’s the top combo almost globally, maybe English-Chinese, English-German, so it’s quite good. What about English-Slovak?

Jakub: English Slovak is very good, but really it differentiates from the domain. Yes, so the domain is very, very important and I remember time before DeepL that we had problem with German to Slovak. Yes, because it was almost impossible to use, but DeepL, great job from them. So I would like to add something, yes. So if we compare human translation where the translator sees and uses suggestions, yes, and has to achieve maximum quality with the so-called full post-editing where the translator modified the output, I believe the results of both these procedures is technically the same, identical. But if full post-editing is half as fast and cheaper than human translation, human translation lost its reason for existence. The other possibility that it’s just misleading of the customer and full post-editing should be included among the human translation or just simply forgotten. That’s my idea, so for this reason, I think that it is necessary to revise this standard on post-editing and stop using maybe this term full post-editing. As for discounts for translator, this will always be the results of agreement, so I’m not talking about the prices. However, it is necessary, of course, to offer a price which the translator can operate in the region because with such a discount, now they are asking for really sometimes 40% of discount, the hourly rate is coming down and it’s not good for them.

Florian: No, it’s not good, but it has an end to this because if the hourly rates going down, eventually the true experts, linguistic and subject matter experts, are just going to leave the field and then you’re going to have actual issues. Because, I mean, if productivity goes up and you still earn the same amount of money per hour, sure, you can make a living. You’re an expert in finance in Slovak and English, well, let me be a translator, maybe, right? But if the hourly rate then drops, then you’re just going to say, look, I’m taking another job, right?

Jakub: That’s what I am talking about. That we are risking that we will lose our translators and yes, you will have another translators, but this will be some au pairs or something like this with a knowledge of language, but if somebody… Means that some doctors or engineers or somebody that they will be translating or post-editing during their free time, I do not believe in this, yes. And even I want to tell you about one more process because now some translation agencies are also using process and they call it MTPE plus revision. Yes, so another… Very strange for me, very strange process, because for me it’s similar to when you cook an instant soup and ask the chef to season it to the level of five star restaurant. And absolutely funniest thing when you are asked to do it by ISO-certified company, so very strange. And also what is another not good experience it’s sometimes they ask for post-editing with cloud-based CAT tool and this CAT tool is overloaded. Yes, the servers are overloaded, so the translator waits literally few seconds until they receive the suggestions from MT and similar is the case when the text contain a large number of tags. Yes, the speed of post-editing itself may be satisfactory, but we must add an effort and time needed to apply the tags correctly in the target text. So such a text is up to me, also not suitable for post-editing. Or you should add some price for tag editing, so that’s it.

Florian: Tell me more about the tags. I haven’t been in this, for about on the tag side, so they still show up in the segment? Like tags that you have to kind of manipulate around, wouldn’t that be mostly taken care of by the tool?

Jakub: I don’t know what MT they are using, but there is a lot of work with the tags, so you have to change it, you have to insert it and everything.

Florian: Let’s go back to this kind of the talent question so what about expertise versus linguistic skills? As these AI tools get better and better and better, where do you think translators or linguists or language experts, whatever we call them, should put their focus? Do you think they need to focus more on the expertise side because kind of like, okay, you got these AI systems and you have to be a lawyer to really add some value to it? Or do you think linguistic skills matter a lot because maybe we can figure out the expert terminology through the AI or maybe it really depends on obviously the type of text? What are your thoughts there about expertise versus linguistic skills?

Jakub: I think that MT really has fundamentally changed the system or work of translator and opens the door to use of the bilingual subject matter experts to perform post-editing, where, thanks to their perfect mastery of the terminology and process discussed in the text to be translated, they can actually achieve a higher quality output than a professional translator who is not an expert in the field, unfortunately. It is therefore vital to adapt the university curricula in translation studies as well as focus more on terminology management and revision, of course, or proofreading, which will play an increasingly important role in the translation process. So it would be also nice to offer some specialized education to students of translation studies like, for example, basic nursing or something like this in foreign language and vice versa, but also in your language, native language, and also vice versa offer basic translation skill to medical or technical students. But we must not expect that graduated physicians will translate text after they work, as I have mentioned. Yes, so I believe that flexibility of graduates is the key. Also, specialization is very important, but it’s very difficult to specialize on small market. So it’s maybe easier for, I don’t know, German market. But here in Slovakia everybody is always asking what are you specialized in? And I just say we do everything because we have to do it and all of us we are using the same translator. There are maybe 500 translator active and all companies use the same translators. Yes, of course, we have in our pool and in our database that this translator is perfect for medical texts because maybe his wife is physician or something like this but we are using the same so it’s difficult to…

Florian: It’s difficult. Yeah, you’re right. I mean, the size of the market obviously matters in terms of how much you can even, yeah, if you can make a living being an absolute niche specialist. You can probably in Germany or China, whatever, but maybe in a smaller market, tougher. Let’s move to LLMs. I mean, obviously now for a year it’s been ChatGPT here and there and everything and AI. What do you think about this? Has it changed your day-to-day? Is it mostly hot air so far for you in day-to-day? Or do you really use it already? Have you implemented anything related to large language models, GPT, etc?

Jakub: I’m using ChatGPT or Geppetto, we call him Geppetto, almost day-to-day basis for content creation. That’s why it works perfectly. However, you must know first what are you looking for and you must know the topic. Otherwise you can be caught in a trap of hallucinating robot, of course. So in ASAP we have kind of discussion group to help solve creative problems in translation. Last week there were a nice question how to translate proverb “burden shared is a burden halved”. Our internal translator found immediately solution for Czech language, but we had quite a problem to translate it to Slovak language. So great idea struck me, I’m going to use my friend Geppetto to solve it. I wanted to help him, so I created prompt. Something like find all Slovak proverbs starting with word X. The answer was there is no such a proverb in my database. Then I gave him just easiest prompt, translate this English proverb to Slovak, and Geppetto reacted immediately and started hallucinating and created oven proverb without any sense. If I’m using big translation, it was something like, “in my company and evil the hour is short”. So it resembled my favorite science fiction movie from Soviet era, where the young girl needed to overcome two robots on guard. She asked them logical question which burned them their brain processors.

LocJobs.com I Recruit Talent. Find Jobs

LocJobs is the new language industry talent hub, where candidates connect to new opportunities and employers find the most qualified professionals in the translation and localization industry.

LocJobs.com I Recruit Talent. Find Jobs

Florian: Tough one for our friend ChatGPT to get that. All right, to close off, you were a founding member of the association of Translation Company of Slovakia, ATCSK. Is this still an ongoing association? If yes, what are some of the activities that you’re doing and what are some of the member benefits people are getting there?

Jakub: Yes, we have to break ourselves. I’m good in English. How to say it? So Mid-Central Europe just several weeks ago, the regional conference which is organized by ATCSK together with Hungarian, Czech, Austrian Association, won the first EU ATC association initiative award, so we are happy. And regarding future events, we are preparing now one very interesting webinar for our members and the topic is LLM and NMT and we are preparing it with cooperation of Institute of Formal and Applied Linguistic of Charles University in Prague. So I’m really looking forward to be participant of this webinar. And regarding some regular and ongoing activities, we are working on connection and cooperation with universities which provides translation studies program because cooperation with educational institutions are, from the very beginning of ATCSK, our priority. It’s mainly focused to help them to connect education with the practice, so still alive ATCSK.

Florian: Let’s make sure that we are still alive in two to three years. So what’s your outlook for the broader industry, for your company in the next, I don’t know, let’s say two to three years because five years, who knows, right?

Jakub: I believe the industry will grow, but it will be still more and more about post-editing and content creation rather than regular translation. And that’s why we are also offering multilingual content creation and SEO translation in our company. And regarding the supply chain, I really do believe and agree that some clients will create automatic system to skip LSPs and will go directly to freelance suppliers. But in case of need for high-quality translation and in case of complicated projects, there will be still need for preprocessing, post-processing, vendor management. So I believe we will be still here, but it will be better to grow than to risk the position at the very end of supply chain.