A recent (civilised) squabble among renowned academic and industry leaders in translation and interpreting on social media caught my eyes. I’m not going to lie - I enjoyed reading it immensely. While this is an extreme oversimplification of the matter, in a nutshell, one group of people believe that translation and interpreting institutions are living in denial and refusing to face the reality that AI is going to make their mode of teaching and training irrelevant soon. On the other hand, some think that the industry is overexaggerating how good AI is at translation and interpreting and T&I training is doing just fine. Well, this is a wild generalisation, to be filled with deeply embedded nuances to both arguments.
Why am I interested in this debate? I consider myself someone with a foot on each side. I am a practising translator and interpreter and I have just started my research in machine translation. I’ve been reading literature on machine translation and absorbing information like a sponge. I’m also a lecturer in translation and technologies. I’ve witnessed the profound changes in the provision of translation and interpreting work. I am not famous or accomplished in either world (yet) but I feel passionate about this topic.
This is by no means a research paper. Far from it. Substack is not a platform to publish research. This is an anecdotal account of my personal experience in the industry and academia. It will be different from that of other translators, interpreters and lecturers, because we work with different languages, in different markets and are based in different places. It’s all relative.
Translation
My first piece of paid translation work goes back to 2010, when I was still an undergraduate student in China. I still remember that assignment. It was a medical paper and it was notoriously difficult to translate. It was around 6,000 words and it took me two weeks to translate from English to Chinese. I did not use machine translation, or CAT tools (I didn’t know they existed), nor did I have any professional training in translation to speak of. I was simply helping out my university professor and she insisted on paying me when I refused. She hired me because I ‘speak decent English’, according to her ;-)
I carried on translating like this for a few years, until I came to the UK for professional translator and interpreter training. The UK market was completely another kettle of fish. I did my first piece of paid translation work in the UK in 2016. The rate was much higher than in China. I used MemoQ for this assignment, but no machine translation. It took me a couple of years to figure myself out in the market and my freelance business started to take off. I was lucky and work kept coming. Then I started hearing about Machine Translation Post-editing (MTPE). My friends hated it and some of them changed profession because of it. They said translation has become meaningless and tedious. I decided to give it a go and it was not long before I received my first MTPE job. I loved it because MTPE enabled me to work much faster than translating from scratch. At least it was the case at first. Then I realised the pay was not great. Companies saw it as an opportunity to cut costs and many translators around me suffered. I thought, okay, this was it. Perhaps I should start looking for a job in other industries and gave up on translation and interpreting… hold that thought.
Interpreting
In my interpreting work, it was predominantly human work. I have not seen much impact from AI yet. Maybe it’s because I’m not important enough as an interpreter. I must confess that interpreting has been a much smaller proportion of what I do today than before I started as a full-time lecturer. However, I have little experience interacting with AI in my interpreting work. I’ve worked with international organisations as well as local SMEs, and everything in between. I now use ChatGPT to do some terminology work, but nothing more than that. I was once at a conference where a combination of speech recognition, live subtitling and machine translation was employed to provide multilingual subtitles. Interpreters only worked into English and machine translation took care of the rest. The quality was not great but no one seemed to care. It was just another conference where the speaker said what they had to say and the hundreds of audience were staring at their phones the entire time, only to offer a warm round of applause in the end. Other than that single experience, my clients seemed content with hiring interpreters with flesh and bones. Of course, I’m not denying that machine interpreting exists, but I have not heard of my colleagues in my small bubble (yes, I admit I live in a bubble) talking about using machine interpreting or interacting with it.
A few years ago, one of my clients decided to stop using my services because she found a magical device that would automatically ‘translate’ what she says into Chinese, and of course she was able to understand her Chinese business partners in English through that said device. She was brutally honest, and told me, “sorry, Faruk, you’re simply too expensive now that I have this!”, proudly demonstrating her new toy to me. I was sorry to lose a client, but such is life. She came back to me after some time, not because she’s stopped using the device, but the business negotiation for which she needed me was far too important to rely on an overpromising stick. For the record, I’m not equating that device to the entirety of machine interpreting. I’m simply stressing that I have little interaction with machine interpreting, so I’m not ready to comment on its performance.
Back to translation
Between 2018 and the summer of 2021, MTPE assignments flooded my inbox. I didn’t mind them, nor did I particularly like them. I negotiated my rates with my clients and arrived at a comfortable financial spot again. Roughly from the autumn of 2021, with no particular watershed moments, my MTPE work took a huge dip. All of a sudden, my clients wanted me to renew their NDA with me and started sending me conventional translation work again. I asked a few of them why, and they told me, MTPE made a huge mess and their clients were simply not happy with the quality. The translation product is not performing its intended function. Plus, translators don’t give a sh*t any more, because MTPE is not fun or rewarding. They’d much rather do the translation themselves. Fast forward 3 years, now MTPE accounts for only 5% of my translation work. While it has become much more difficult to negotiate a decent rate, but I do mostly translation, not MTPE. There is a clear distrust of machine translation and AI from clients from where I stand. For what it’s worth, I’m still actively involved in MT quality annotation work as I believe MT can do much more good than harm. It’s easy to lose sight of the big picture because translators often tend to see MT as a threat to their work much more than how the general public benefit from it in everyday life.
Quality of Machine Translation
Is the quality of MT really that good? The short answer is yes and no. According to my students’ many case studies, MT is very good at translating technical, legal and function-oriented texts, not without flaws though. However, when it comes to creative and promotional texts, it is far from perfect. Large Language Models (LLMs) are working wonders on many fronts. Sometimes the translation output from LLMs leaves me flabbergasted, as I would have never thought of something that clever. On the flip side, sometimes they hallucinate and mislead people. They can make mistakes that humans would never make. Want an example? Make sure to read my post on LLMs speaking Uyghur. Trust me, you cannot afford those consequences. If you think LLMs only hallucinates in under-resourced languages, think again. The term ‘Artificial Intelligence’ is misleading, as it does not process information in the same way the human mind does. You can argue that AI is insanely clever, and I’d agree, but you can also argue that there is no intelligence to speak of. It is simply (yeah, right!) an imaginary machine with a vast amount of data and a far more powerful computational and learning capacity than the human brain. It can detect patterns from data in ways humans can never match, but it does not truly ‘understand’, whatever that word now means, what it’s dealing with. It will, for the foreseeable future, need a human pair of eyes for translation, interpreting or anything else for that matter. AI can fly aeroplanes now, but I would never board a flight if I know there is no pilot.
Translation and interpreting training
I agree that institutions providing translation and interpreting training need to adapt. However, the adaptation process should not be haste, or filled with anxiety or fear of obsolescence. I’ve heard of admission numbers dropping in many institutions. This is definitely true. This means that teaching translation in its purest sense is no longer ‘sexy’ or sufficient. However, I struggle to point finger at a single institution that still stubbornly holds on to linguistic training only. Changes are happening, and institutions are still in the haemorrhaging stage from the impact of AI. I understand the argument from the industry along the lines of ‘translators will be out of work soon because AI is coming for your jobs!’ It is true to some extent because the industry is unfortunately also enabling this trend, driven by a relentless chase for lower cost and faster turnaround. It may sound like I’m judging people for doing this but I am really not. It’s a business after all. However, I am also hearing stories from clients who are desperate for high quality translation but are not getting any. Many of them do not trust language service providers (LSPs), especially large ones, because these clients are left with a bad taste in their mouth far too often. For one reason or another, AI cannot fully meet the needs of end users of translation and interpreting services, at least not yet, but many LSPs are pushing AI too hard and branding it as a miracle worker. Perhaps they should sit down with their friends in the academia to have a calm conversation about what they can do to improve quality and make their clients happier. Perhaps they can also borrow ‘Introducing Translation Studies’ by Jeremy Munday, Sara Ramos Pinto and Jacob Blakesley and have a light read. I do not have the answer to this question, but how many LSP owners can genuinely say that they have kept their clients happy over long terms by employing AI and MTPE only? Are they truly ready to embrace a future without their translators? I have my doubts.
Translation and interpreting training needs to link up with the industry, but the industry also needs to listen to what academics have to say. I have worked in other industries as well, and it seems to me that (again this is anecdotal) the gap between academia and industry is vast in translation and interpreting. We need to work much harder at bridging the gap. So perhaps the online feud is not a bad thing after all. It seems to reflect frustration from both sides, not of each other, but out of working in isolation and lack of conversations.
Much love.
Comentarios