#275 The Future of Language and Translation Education with JC Penet and Joss Moorkens

0:00Joss Moorkens:The main differences that we've seen as compared with earlier machine translation or AI translation breakthroughs are that LMs can generate text that is very similar to the text produced by humans.

0:18JC Penet:Hey everyone and welcome to this first podcast of 2026. Very excited about today's discussion we're going to have with JC Penet and Joss Moorkens. So JC is a Reader in Translation Industry Studies at Newcastle University, and Josh is an Associate Professor at Dublin City University. They're both co-editors of the recently released super interesting book called Teaching Translation in the Age of Generative AI, New Paradigm, New Learning. And we link it in the show notes. It's actually freely available, the entire book, 300 pages. So hi, JC and Josh, and thanks so much for joining today.

0:57Joss Moorkens:Hi. Well, thanks for having us here today. Great.

1:01JC Penet:So let's start with Joss. Tell us a bit more about your background first and then move on to JC as well. Quick self-intro to give our listeners and viewers a bit of background here. Sure. Thanks, Lorian. I'm an academic at Dublin City University in Ireland. I'm working and researching and teaching at the intersection of translation and technology, following on from my predecessors here at DCU, like Dorothy Kenny and Sharon O 'Brien, and have done this sort of research for a number of years. Initially, I was doing a lot of empirical studies, but recently I've been moving into looking at the bigger picture.

1:33JC Penet:I suppose inevitably, as our careers progress, we end up managing research rather than carrying it ourselves, so that leads to that sort of bigger picture idea. I'm also the science lead for part of a national research centre into AI called the ADAPT Centre, and I represent the university and the board of the European Masters in Translation Network, and that was where I met JC.

1:54Joss Moorkens:Okay, well, hi. I'm an academic as well, working for Newcastle University, as you said. I'm also a lifestyle and executive coach. And I've been educated across France, Germany, Ireland and the UK. And I've been teaching higher education for about 20 years now. and well got two passions translation and education and in the well for the last 10-15 years I've really tried to bring them together one of those uh the thing that I've done to achieve that is to co-found the association of programs in translation interpreting studies in the UK and Ireland with Olga Castro back in 2017 when the UK decided to leave the EU and we were kicked out of the we were not kicked out but we we couldn't stay in the EMT anymore and more recently I published working as a professional translator with Ratledge which is a non-country specific book that looks at what it means to work as a professional translator in the 21st century and the approach is very very much inspired by career coaching where the students are at the center of the process and they've got to make sense of the uh well the uh research-based industry information that we've got but to try and reflect on what it means for them in their own circumstances uh because everyone is different and uh and now i've co-edited uh this book uh with my friends joss and massara i'm saying friends because i've known them for a number of years now They have become good friends.

3:29Joss Moorkens:And as I say, we met at the EMT meetings in the mid-2010s. And it's been really lovely working with them on this book. Great. Thanks so much. That gives us a bit of background.

3:41JC Penet:So now tell us more about the book itself, what it's about, how the idea to compile, and then for you to edit the book come about. And maybe you should also acknowledge some of the other contributors that contributed to the book.

3:54Joss Moorkens:I think from the academic 2022-23, and that's how we work with academics in academic years. So following the release of ChatGPT in November 22, we could see, I mean, Joss felt that as well. There were quite a lot of anxious colleagues. I remember, if I remember rightly, Joss, you organized a workshop for the EMT around AI for translation teaching. And we could sense there was a need. talking to colleagues, we could sense that there was a lot of anxiety around what it would mean to have generative AI in the classroom, whether we want it or not. The European Language Industry Survey of 2024, for instance, showed that the university staff saw generative AI implementation as the greatest challenge for university teachers.

4:46Joss Moorkens:That's what university teachers said was their greatest challenge at the time, with 90 % of them seeing that as something quite negative, as an issue. But as we say in the introduction, adapt and implement we must. You know, we don't have a choice. Or as Slater said, you know, you've got to skate where the puck is going. That's something we quote in the introduction as well. So we need to kind of help our students have a sense of where things are going for the translation industry. But for many colleagues, the question remained, how do we do that? And we've got to acknowledge that colleagues teaching on translation programs have very varied backgrounds.

5:33Joss Moorkens:And so some come from a literary background and not necessarily engaged with technology that much. Some are from a language teaching background and maybe having been teaching for 30, 40 years, and that's a lot of change. So, you know, we've got to acknowledge that. And this book really comes as an answer to address that need to help colleagues with this issue or with AI in a collegiate and supportive way. And, yeah, you say you mentioned that we could mention some of the contributors. And as, for instance, in part one, we explore the new skills and competencies that were needed to help students develop in the age of generic TVI.

6:16Joss Moorkens:You've got Gary Massey and Maureen Ehrensberger-Dao talking about the digital skills they need to acquire. Ramon Inglada about the concept of suitability. We also have a contribution by Lynn Barker who talks about how we could teach data literacy to our students by using science communication as a starting point, which is quite an interesting angle. Masaru couldn't be here today sadly so one of the co-editors has a whole chapter around prompting Joss with Gokhandrogu has a chapter on ethics so quite a lot of different angles for the book trying to support our colleagues with this transition

7:01JC Penet:I might add a little bit there so as co-chair of the technology working group in the European Masters in Translation We sort of went quite quickly from panicked discussions of emergency remote teaching in the pandemic to discussions of generative AI following the launch of ChatGPT. So when JC came to me with this idea for the book in October 2023, it really seemed like a natural extension of the discussions we'd been having in the EMT network and beyond with colleagues in different universities. But because JC and I are perhaps a little bit sceptical about technology at times, it seemed like a good idea to ask Masaru Yamada to come and join us.

7:41JC Penet:Because as well as being a great fellow and a talented researcher, he's really positive about technology. We wanted to, you know, not be trying to give our own scepticism, try to, you know, push that on people, but instead bring his natural enthusiasm. So we came up with this list of authors, including the names that JC mentioned, Gary and Maureen, you'll know well, Florian. I do. And we asked them to contribute and they all said yes, amazingly enough. And we decided to go at Language Science Press in Berlin because my colleague Dorothy Kenny edited another book as part of a project called Multi-Train MT that came out in 2022 with them.

8:21JC Penet:And they really exemplify best practice for open access books. it means it's available to everyone and often we get requests for publications from people you know book publications that are behind paywalls from people whose institutions can't afford access so that was really high on our wish list to make it available for everyone for no charge that is amazing and again we're we'll mention uh we'll make the link available in the show notes so So go there and click. All right. Let's jump into the topic and talk about why do you think? What makes kind of today's LLMs AI? I mean, it used to be called Gen AI.

9:01JC Penet:Now it's maybe just AI. Let's see what it's going to be named in five years. But what's different? Why is it a different disruption than like earlier machine translation or AI translation breakthroughs? you know, maybe early 2000s statistical. And then, you know, we had the neural disruption in like 2016, 2017. What's different today?

9:24Joss Moorkens:Quite a bit, I think. Well, first of all, I think you said, Joss, and you're right, that we're kind of seen as, you know, skeptical when it comes to technology. But I feel interesting that actually depending on who I am with, I'm either seen as a champion of technology or skeptical. So that's very subjective. But I think, you know, in translation studies, we always felt or we felt for a while that we were quite well equipped to deal with the disruptor that Gen AI is because we've been dealing with, say, neural machine translation since 2015-16. So, you know, we'd kind of adapted already. But having said that, it's been more of a disruptor than we might have liked.

10:12Joss Moorkens:I call it a disruptor here rather than disruption, because disruptors can be good or bad, and often good and bad. The three main differences that we've seen as compared with earlier machine translation or AI translation breakthroughs are that LMs can generate text that is very similar to the text produced by humans. in a case for high language resources that the case in fact the tag they produce is so convincing at times that it's really difficult to spot where they got the meaning wrong or where they've hallucinated because you're just kind of compelled by the text when you read it one of my bugbears is when people say you know talk about human translation being intrinsically better than machine translation or or vice versa and for me or it's very vice versa but anyway but you know human translation is vast you know some some human translation is amazing and other human translation is very poor so but in in terms of writing here we can see that you know LLMs can produce things that are really really convincing the main difference as well is that they can be prompted uh you can ask them to take their context into consideration something that obviously neural machine translation tools couldn't do.

11:36Joss Moorkens:So in the words of the Functurist approach translation, for those who are familiar with it, they can do what previous versions couldn't do. They can take the purpose of the translated text into consideration. They can integrate the preferred terminology that you indicate, some kind of terms of phrase. you can prompt the tool to produce what you want them to achieve. And the third one, which is actually an important one, is public perception. There's been a lot of hype around what LMMs can do, with some headlines claiming that they've made translators redundant already. However, such headlines are based on a very narrow definition of what it means to be a translator, where that's how I feel.

12:28Joss Moorkens:If being a translator is just translating from one language to the next and it's just a language capacity, then yeah, you could argue that possibly that's being jeopardized by LLMs. But if being a translator is more than that, it's also ensuring client requirements and quality and all the rest of it, then that's different. And nevertheless, public perception has led to a drop in demand for some freelance translators. And it's also led to the perception that there is little to no point in studying languages or translation at university, although, as we know, this far from the case.

13:08JC Penet:In one way, it's a continuation of the NMT disruption, but of course, it's a lot broader. And as JC mentions, the hype is a lot different. Back when we were doing research on neural machine translation, we, you know, we could say and genuinely thought that, you know, our research on NMT was really about making translators better or faster. But now, you know, there are times and there are use cases when it's better or faster to use technology. and so increasingly it's about you know dividing what you know thinking about that use case and thinking what's the appropriate technology stack to use is it appropriate to use technology at all so for example I did a study with Harvard Medical School and Boston Children's Hospital they were looking at patient discharge instructions so the 20 percent of the the sick kids, families who come into Boston Children's Hospital speak a language other than English.

14:11JC Penet:And those languages are Latin American Spanish, Brazilian Portuguese, Haitian Creole and Arabic. So for those of us that know a bit about machine translation, we can imagine that the quality that will be produced by a machine translation system at best is going to differ massively between those languages, translating from English to any of those languages, similarly from LLM translation. So, I mean, we found that potentially using a GPT tool, we tested with GPT 4.5, I think it was, or no, GPT 4, and the quality of translations was better than what they were getting from a commercial translation provider.

14:54JC Penet:There were fewer critical errors. There were fewer errors to do with medical terms. But the problem is the hospital wants something that works out of the box for all of these languages, and current technology is not going to do that for them. Our other concern was if we publish this research, which we did. It came out in JAMA PEDS, which is the Journal of the American Medical Association for Pediatrics. It came out in July of last year. our concern was that in the US, it's legally required to have oversight for anything that's automatically produced. But then what if someone takes our research in another jurisdiction where it's not legally required and thinks, well, I can just use chat GBT for everything.

15:41JC Penet:And, you know, so then we're not going to get good outcomes. So it's become a lot more tricky. And so it means that maybe there are use cases where it's better to use technology. And this is something that's been happening since the NMT era. If we look at different companies, big companies, they've tended to split their content into something that can be automatically translated with little or no human involvement, maybe post-editing, and then heavy human involvement, depending on the level of risk, the value of the content, and the specific use case, maybe the shelf life as well. So let's go to the education part.

16:25JC Penet:And the book argues that the central question is no longer can students translate correctly, but can they judge suitability? Tell us more about that question, how you explored it and what you found.

16:38Joss Moorkens:Yeah, so it's partly the point that our colleague Ramon Inglada makes in his really good contribution called AI Literacy, the concept of suitability and core translation skills. As he says himself, he's inspired by Lynn Barker's MTA literacy project, and I believe that she came to talk to you about this back in, I think it was 2022. And Raymond argues that what he calls the traditional translator skills that most translation programs would agree students need to develop. So these are linguistic knowledge, translation ability, and cultural expertise. Well, these should not be allowed to disappear because they remain very important.

17:28Joss Moorkens:However, he shows how when translation students use large language models to produce translation, the focus moves from being on whether the content is accurate or not to whether it's suitable. or not. So it's a change in emphasis. Obviously, accuracy was a big, big thing 10-15 years ago, especially before NMT. But now, I mean, the way that we do on actualization programs is very different. Because, you know, we can do more. We can be far more demanding of our students. So that's fantastic. It's far more fun. But of course, it doesn't mean that it doesn't matter whether the content is accurate or not, because it still matters enormously whether it's accurate or not.

18:17Joss Moorkens:You could have like false friends or little nuances that are lost and they will make massive difference and that's very important. But as I said, the focus is changing. So to that effect, translation programs across the world need to help students develop two complementary skills. That's what Ramon shows. So these are skills that are already found in most translation programs, but maybe have been given less emphasis before. So these are selection and assessment. So these are the two skills that we should assess for selecting the solution. And you can see that if you work in a CAD tool, you might have a suggestion, well, your own suggestion in your head, a suggestion from your translation memory, a suggestion from machine translation, and you've got to select and then assess what's best.

19:12Joss Moorkens:And there is also one new skill. That's the core one here. That's the one that's game changer is effective prompting. So again, our colleague Masaru has an entire chapter on this, a really effective chapter on how to kind of help our students reflect around what's effective prompting. so again these new or the two complementary skills selection assessment and the new skill effective prompting do not substitute the three core skills that we mentioned before so linguistic knowledge translation ability cultural expertise but they build on those three core skills and the ultimate aim is for our students to understand their added human added value as human translators so they can properly engage with outside other stakeholders and say this is what I bring to the process and explain clearly how that helps produce augmented translation because we talk about augmented translation but what is that like in effect so suitability is not really a new concept but it's something It's been part of translator training since the 1980s, since the functionalist turn in translation studies that started in Germany with Vermeer.

20:34Joss Moorkens:But somehow this functionalist approach to me has regained all of its significance in the age of generative AI.

20:43JC Penet:Let's move to ethics, AI ethics. And I think, Jos, that's a central theme of your chapter that you've co-authored. So tell us more about ethical blind spots, Where do you see them with students or even professionals today? Sure. I mean, following on from JC's last question, you know, for us in Dublin, we've produced, we have an MSc and an MA in translation, and we've produced very few translators other than an Irish English over the last 10 years. Mostly our graduates move into a job that's tangential to language and translation or in the localization industry. But, you know, they're not necessarily translating.

21:24JC Penet:And that's a pattern that we see across lots of other programs. Perhaps it's more extreme because of our location in Dublin. I mean, for Howe and Pym, they showed the same in Melbourne. You know, that's 70 % of their graduates weren't becoming translators. So they're moving into this intersection between humans and technology or humans and data and languages and cultures, which, you know, is really interesting. And there appear to still be lots of jobs there. So that's sort of the context. And our feeling is that as people move their careers, they're going to start to make decisions that influence people.

22:00JC Penet:And those are going to be ethical decisions. So if we forearm them with, you know, some tools to help them with that decision making, that means they'll be able to sort of exercise their agency along their own sort of personal ethics. So coming back to the idea of blind spots, back in 2024, myself and some colleagues, Sheila Castillo, Federico Gasparri, Antonio Torral, Maya Popovich, we proposed an idea for a triple bottom line for evaluating technology for translation of people, planet and performance, each with equal weight. and this was inspired by an idea from Business Ethics of the Triple Bottom Line by John Elkington and various publications to do with sustainability and green AI.

22:45JC Penet:But, you know, it can only be a heuristic. It can only be an idea. You can't balance the effect on people versus whatever scoring system versus MQM, you know, or versus the effects on the planet. But the wider effects beyond performance are so easily ignored by us as researchers. I mean, if you look at the publications for any MT or LLM conference by PhD students, by professionals. So the point isn't to downplay the performance or in Alkington's original idea, Profit, but just to suggest that these should also be important. And the hope was that that would be transformative. And we've seen an example of this when we have a PhD student, myself and Sheila, Joao Camargo, who's just finishing his PhD.

23:29JC Penet:and he's created a class for teaching NLP students about responsible evaluation, taking this triple bottom line idea. And they really took to it because it has actually never occurred to them before. They were just using an automatic metric, write like blue or similar, and just trying to maximize the scores. So for translation students and NLP students, as they move on in their careers and get autonomy, our hope is that by inculcating these sort of ideas, they'll be empowered to make ethical decisions and having a positive effect, or at least an effect that's in accordance with their own personal ethic.

24:05JC Penet:There was also part of the book talked about the risk of AI generally kind of de-skilling students. I know it's probably a controversial term topic, but it's a big topic. But based on the findings of this particular book, do you think that this risk generally exists in certain areas, in the areas that we cover, or is it overstated? And is there maybe a risk of coming across as a little bit Luddite-ish if we're talking about, oh, no, the new thing is de-skilling the young, as it were?

24:40Joss Moorkens:I think it's a truly excellent question, this idea of will AI allow de-skilling of students? It's one that all higher education institutions, even secondary schools, are probably dealing with in all subjects. I'm not sure there is a definitive answer to it, but my sense, and I believe it's fundamentally what we argue in our book, is that to some extent, not engaging with AI is what truly risks de-skilling our students in the long term. So let me explain. There is a time and a place where it's important to keep AI at bay in our teaching. It's the case, for instance, in language teaching, where, say, you want to help your students develop as autonomous linguists.

25:24Joss Moorkens:And we've all done those kind of translations where there's no dictionary and when you translate it into your other language and the L2, and the objective is really to build that kind of linguistic ability. And in that case, AI would defeat the purpose, right? So there is a time where it's okay to say, okay, don't use AI, make your own mistakes because that's how you learn, and develop your own resources, linguistic resources. So basically learn to walk before you run. And that's sometimes the argument that's used on translation programs. So I've heard colleagues, some colleagues say that they would prefer to keep AI outside the classroom because they argue that using AI means that students don't learn how to translate for themselves.

26:10Joss Moorkens:They don't learn how to translate full stop. They're discaling. to this i tend to ask whether we should ask as a result of that we should ask our students to translate with ink and a quill because any kind of technology is cheating right uh should we say to our students no you're not allowed to use the internet as a resource because that's cheating um as an anecdote when i did my phd in uh in the late uh 2000 so 2006, 7, 8, 9, I can't remember. But anyway, my father-in-law had done his 30, 40 years earlier in classics. I always thought that it was cheating what I was doing because basically I could read articles online when he would have to travel to a library somewhere far.

26:58Joss Moorkens:I had a word processor. He was like, well, that's so much easier for you. But he not realized that the expectations were slightly different. Right? so of course when I say you know should we as I assume to use a quill and some ink that's provocation and I do see the point of that some colleagues make who rightfully argue that generative I is not just another tool and so Renée Desjardins made that point at an Aptis conference once I thought she made it beautifully well it's not just another tool because there are as just mentioned ethical implications to to generate tvi and and so we need to kind of discuss that uh this being the case my point remains that we can't really keep the world at bay so fundamentally i believe that what makes a translator a good translator their ability to leverage the right resources whatever these are uh so if we don't know if don't allow our students to to engage with ai on our program then we risk becoming complicit in the disconning of our students because they first of all most likely they most likely use ai anyway but they'll do it badly and with little awareness of the limitations or the ethical implications of ai and they the students who don't use ai or refuse to use ai because we have some students who refuse to use AI for ethical reasons, happens.