AI and the future of translation. In search of evidence — Transcript
Full transcript
- 0:01This is a talk for Translators and Interpreters Australia, who are very concerned with protecting
- 0:09the rights and well-being and the privileges and the pay of translators and interpreters.
- 0:15It's about AI, artificial intelligence, and its possible impact on our professions.
- 0:24And my starting point is that there's lots of talk around, there are lots of instant experts, and
- 0:33we're not very sure of how much of that talk and the predictions, especially the more catastrophic
- 0:39ones about the challenge to humanity, to the environment, to our professions, to our work, to
- 0:46communication, to the quality of language, all those things, we're not sure how much of that is true.
- 0:53It seems to me that what we have to do is get down and look for some evidence.
- 1:01Some of you may remember the millennium bug, the fear at the end of the
- 1:09second millennium that all computers would go wrong because nobody had fixed up the date to
- 1:16to change the big numbers. There were a thousand instant experts around and the
- 1:23role of an expert, ladies and gentlemen, is usually to spread fear and be negative. Why are experts negative?
- 1:31Well, because if they're wrong nothing bad happens - everybody's happy so you forget about
- 1:37the expert. And if they're right, then you can say I told you so. So for experts it's
- 1:45just basic risk management strategy to be negative if in doubt. And that apparently is what could be
- 1:53happening here. Let me start my investigation as best I can and I'll start by just
- 2:03pointing out what I'm dealing with or what I'm talking about here.
- 2:10The one you've all heard about I presume is Chat GPT because you can get on that for free at
- 2:18the moment and play around with it. It's been with us since November 2022; GPT 4 has been released
- 2:26publicly so you can play around with that too and I've also been playing around with Google Bard
- 2:32which has been released in experimental mode and is available in Australia.
- 2:41Chat GPT happens to have the fastest growing user-base in the world ever for an app. So people
- 2:49are playing with it; people are having fun; people are exploring it; and I hope you are too. I have
- 2:55been now. I must admit that when I first went into that little world, I thought, right that's a big
- 3:04change. That is impressive. It can translate and it can give us information about a lot of other
- 3:11things and it's going to be incredibly useful but it might be this moment of singularity when the
- 3:21processing capacity of artificial intelligence is greater than that of the human brain.
- 3:27That's a moment of singularity. It has been reached in the game Go and in chess.
- 3:36Perhaps in translation? Perhaps interpreting, too? We don't know. But certainly my first encounter
- 3:45was one of shock and surprise. However, as I continued, things became a little less frightening.
- 3:54Now, a commonsense approach out there among experts of all kinds is this one is from Sasha Luccioni
- 4:06who tweets profusely on this topic: it's the iceberg approach
- 4:16where we recognize that there are actual benefits to the new technology but underneath
- 4:23the water there are lots of threats. The complete version finds a lot more threats, some of
- 4:31them truly worrying, such as the environmental consequences - we've got tons of carbon emissions -
- 4:38and the ones about representation might concern the work of translators
- 4:48as indeed do privacy and copyright infringement So there are lots of things there to be
- 4:56scared of certainly and a few things up there to rejoice about. Lynne Bowker, who knows as much
- 5:05about this as anybody, who's worked a lot on machine translation literacy
- 5:13has adapted that to translation. So neural MT which is
- 5:23what we've had with us - neural machine translation - since 2016 can be analyzed in much the same terms,
- 5:32with a few advantages up the top and many possible disadvantages
- 5:35down the bottom. Hallucinations are just crazy results. What's interesting is that since 2016
- 5:49if you look for actual evidence of growth in these negative things
- 5:56it's quite hard to find the evidence and that's going to be my main point of entry.
- 6:03I want to talk about these things that are said about
- 6:08firstly neural machine translation but then AI such as chat GPT and Bard.
- 6:15What's the actual evidence for the bad things? I'll be dealing with transparency first,
- 6:23which is a big point, job destruction, changing skill sets, anchoring, which
- 6:29we'll come to in a minute, which languages benefit, which will be marginalized, wage dispersion
- 6:36in the industry and intellectual property theft. That's a lot of things to deal with. I'm going to go fast.
- 6:43First, let's be very clear what we're talking about..A lack of clarity
- 6:52was evidenced around me where I live and work in Spanish and Catalan. We had people
- 7:01who work with subtitles complaining that Squid Game, which was a huge success on Netflix,
- 7:10was post-edited: Netflix would give a translator a machine translations and ask
- 7:17them to correct the machine translations. There was a lot of public outcry in various forms
- 7:24of the press and online and then a particular talk there that I picked up and recorded,
- 7:31by the head of the Audiovisual Translators Association in Catalonia,
- 7:38which is where I am at the moment. She talked about the massive threat of post-editing and I went
- 7:48through her talk you can see I actually took an automatic speech-to-text and then automatic Google
- 7:56Translate version and produced the subtitles, which actually worked. I produced subtitles
- 8:02automatically for her. Going through the talk, I didn't know what she was talking about
- 8:09because it was all mixed up. She talked about post-editing as if it were raw machine translation
- 8:18as if all the mistakes that come from raw machine translation are the mistakes in post-editing.
- 8:24But they're two different things. The question is: it advantageous to translate from
- 8:31scratch without any electronic aids, which few of us actually do, or put a text through
- 8:40machine translation and correct it? That's the only question. Is it better to use
- 8:46that passage through machine translation and correct it or do it from scratch?
- 8:52When I teach, we just do experiments. I don't tell students whether one is better than the other.
- 8:57I let them find out for themselves. We measure results in terms of time spent
- 9:03and the quality of the translation at the end, which is assessed through peer reviewing.
- 9:09So it's important when you get into these conversations
- 9:15to stress what you're talking about. We're not talking about raw machine translation.
- 9:21In the case of GPT and Bard, what they can do for translators is translate
- 9:31but the quality is actually not much better or on a par with DeepL and Google Translate
- 9:40for example, and there are many other systems out there. It's not a huge qualitative advance
- 9:49when we give the prompt "translate from this language to that language".
- 9:55Try it for yourself. That will help calm nerves, I think. However, the new technologies,
- 10:03the GPT technologies, can do a lot for us and help us with what translation memory software does or
- 10:12language automation in general, if I put it together, can do all these things in addition to
- 10:19translate, well or badly as a machine translation. It can obviously bring up your
- 10:27previous translations. This is translation memory technology and you can revise it.
- 10:32You can use different feeds. You can have different machine translation feeds and the GPT
- 10:38feed which you now get in Trados Studio to get alternative translations to choose from.
- 10:46Lilt gets you to type and then offers proposals and they reduce as you type and
- 10:52that's one interesting thing that I'll come to in a minute. I might also add that even a
- 10:59free online translation memory suite like MateCat now has a terminology tool linked to GPT.
- 11:08If you're in doubt, you can click on that and you you get an explanation of what the term means in
- 11:14the particular context you're dealing with. That can be tremendously useful for getting
- 11:20translation suggestions about what you're doing. Propagation automatically making a change in the
- 11:28whole text when you decide to translate a term in a different way. GPT can be used
- 11:34to pre-edit the text to translate. If it has to be simplified, we can simplify
- 11:40it. or f the grammar has to be corrected. The text will then go through machine translation better.
- 11:46It could do that for you. If the sentences have to be shorter, it can do that for you.
- 11:52It can prepare the text so that it will work better in machine translation. There are online tools where
- 11:59we can exchange notes with other translators. We can use speech-to-text; that is, we can speak our
- 12:05translations, which often produces a more natural- sounding translation. We can also go from text to
- 12:12speech and I always do that to to revise my translation: I sit back and listen to it being
- 12:18read to me. There's no better way to spot the typos and the grammatical gaps that happen to be there
- 12:26You can revise a translation - the spelling quite obviously and for grammar and for
- 12:31readability and so on. In fact, we're discovering more things that it can do. For example, aligning
- 12:39text your old translations so that you create a translation memory for reuse,
- 12:45GPT technology could do that for you. So I think it's important to insist that
- 12:54we're not talking about just machine translation; we're not talking about just its quality; we're
- 13:00talking about all the many different ways you can interact with it and where it can help you.
- 13:05And it the main help is not the core task "translate A to B". It's all these other things that
- 13:14can assist us in our task. Let's just continue the list: revise the punctuation; translate for a specific
- 13:21reader and purpose - this is where GPT really comes in - you can say translate for a particular reader.
- 13:27Terminology management and extraction as well propagation.
- 13:33You can set up your preferences and check consistency. Now all of
- 13:43those things concern revision all of those particular things concern opening up the range of
- 13:54things that you can choose between and improving quality. So my point is:
- 14:02what's new here is not the key task of translating; what's new is everything that's around it.
- 14:11We can see this in Audiovisual Translators Europe that has a I think a very good paper
- 14:17on automation in the industry. We can see you the linguistic translator here using tools
- 14:26for all these things. They're aware that all these things can help you so they
- 14:34make use of the term "augmented translation" - this is not you being replaced,
- 14:41it's you being extended in all these different ways. I agree very much with their conclusion
- 14:47here that the real problem with the technology is that a lot of it has not been developed in
- 14:54cooperation with the people actually using it, with translators and interpreters, which means
- 15:00that there are some serious limitations in it and that the technologies can be much better.
- 15:06but refusing the advances of automation on all those fronts, no that is not the way forward,
- 15:13they argue and I generally agree with them - I think that's a fine paper.
- 15:20I have however also been looking at another big paper, a position
- 15:25statement produced by the French literary translators. You can see it there.
- 15:34We have translations online into English and into Spanish
- 15:45and their headline is transparency. They say they're not against AI, not against any of the
- 15:52automation, but what they want is transparency: just make it clear who is using what.
- 16:00I agree. Who could possibly be against transparency? It's just basic honesty.
- 16:08However, there are catches. For example, translators have forever been using dictionaries so in my
- 16:15published translation should I list and declare and thank all the dictionaries I've used?
- 16:22I ask people around me often about particular words and usages. Should I list all of them?
- 16:28Perhaps I should in the name of transparency. Would it hurt to list the machine
- 16:35translation systems that you've been using and how you've been using them? No it won't hurt at all.
- 16:43In fact, I'm doing this now. I've been engaged in a collaborative translation and
- 16:49I do in the preface explain exactly whom I've worked with and which machine translation
- 16:55systems I've been using. Why not? One of the things that intrigue me with
- 17:02GPT is that if you ask it to translate, it will translate and give you a text
- 17:10and if you ask it to adapt to a particular readership, it will do that too and Bard will
- 17:16do it very well and give you an explanation at the bottom of exactly what it's done.
- 17:24But if I ask that system anything about a legal matter, not a translation matter, as
- 17:30you can see here I've done that: Should I file for bankruptcy in Australia? I wonder why I'm asking that...
- 17:36it will tell you it's crucial to consult a qualified professional. It will
- 17:46declare its limitation and point you to the use of a professional.
- 17:53It's protecting itself in legal terms. That is transparency. I think that it should do exactly the
- 18:01same for translations. It can translate for you but it should say if you want a completely trustworthy
- 18:07or you want an adaptive adaptation or you want to be sure that the translation is going to fulfill
- 18:13the purpose, go to a professional translator just as you would go to a professional lawyer
- 18:21or accountant or tax expert. So I agree very much with transparency
- 18:28and a lot more could be done.
- 18:36Let's see...
- 18:42If you are a translator and you believe your work has been stolen...
- 18:46I asked them have there been cases of translators who've had their work stolen because of
- 18:53machine translation databases or translation memory databases because that's one of the
- 18:59big risks that's always put out there. There have been
- 19:04cases where translators have sued companies but none of them have concerned AI, that is automation,
- 19:11or databases as far as I can tell. What's interesting is that you could if you're a
- 19:20translator and you believe your work has been stolen, consult with an attorney.
- 19:25Ask a professional. So that's transparency. I would like them to do that for translation.
- 19:32I have written to the organizations involved and made that request and they have
- 19:40not got back to me. Does automation destroy jobs? This was in Politico:
- 19:50Who killed the EU's translators? Automation is creeping into European Union institutions
- 19:56and translators are among its first victims. This is in May 2023 and we can see
- 20:06that the employment of translators in the European Commission
- 20:14which actually has the biggest translation bureau in the world ever
- 20:21is decreasing. The problem is what's the cause of it. If you can see here,
- 20:31from 2012 to 2022 where does the money go? It goes to outsourcing
- 20:42so the number of in-house translators has decreased because it's cheaper to outsource
- 20:48so you can adapt to variable demands. It makes sense as well and they've managed
- 20:54to reduce their staff and their budget that way. The cause is not automation it's just outsourcing.
- 21:01So the report is just a load of rubbish. Automation might be in it but we've got no
- 21:07numbers to actually show that. We can see here the budget for translators
- 21:17in the last year - these are the most recent numbers I can get - actually increased by 5.4 percent for
- 21:25in-house staff and 11 for contract staff so they have increased their budget for a translation
- 21:31despite the numbers that you just saw and if I go back here, the cost per page, which you can
- 21:39get from the published numbers, has actually increased although if you go back 10 years
- 21:45before then it was about 130 euros so it's got better but it's still pretty bloody high as a cost.
- 21:54So if they're using automation to get rid of people and cut costs, they're doing a pretty bad job of it.
- 22:01We should get some real data before we start publishing headlines about automation.
- 22:09This is a slightly more serious study about the same thing.
- 22:14They seek evidence from Google Translate and Chat GPT. So or what they did was look at the
- 22:23advertisements for translation jobs on something called a large online labor market - I think that
- 22:31must be ProZ but they're not saying - over the period in which Google went neural:
- 22:38from January 2016 to May 2017. They assumed there was a qualitative change and they
- 22:44wanted to see if that affected the employment of translators. They did find that the demand for
- 22:51translations declined by 13 to 20 % , that's a pretty big band but anyway, and interestingly
- 22:58found that the demand for transcreations these are non-routine cognitive and interactive tasks,
- 23:05the kinds that neural machine translation was not good for, did not decline. And over that
- 23:13period they found no significant change in the rate of pay. Then they assume that something
- 23:20like that can be seen for Chat GPT but they don't have a get convincing numbers.
- 23:31I find the study naive. If you know about the translation industry, I don't think you get
- 23:37immediate impact from technological advances. So Google Translate goes neural and all the others
- 23:44go and you're all in the same years 2016-2017, but it takes a long time for that to filter through
- 23:52to any consequence that is substantial for the industry. It is quite possible that lots
- 24:00of clients went and tried Google Translate in that period, which might account for a decline,
- 24:08and many of them may have come back the year after when they found that they could not rely on it.
- 24:16This is a similar study but a bit more serious. It looks at advertisement for jobs in
- 24:25translation over the period 2014 (not 2024!) to 2019.
- 24:43They find that the advent of a neural machine translation in 2016 did not lead to decreases
- 24:51in the demand for language professionals, although it did increase the demand for transversal skills
- 24:57and skills in the use of automation: skills which are most important for "high quality
- 25:07translations of text that machine translation tools cannot deliver with accuracy."
- 25:13So there are two increases: one of them in the technical skills to work with technologies,
- 25:18and the other to do the jobs that machine translation was not good at at the time.
- 25:27This in fact contradicts the previous data.
- 25:32So is there job destruction? There's no firm evidence of that happening.
- 25:41Globalization increases, therefore the number of translations in
- 25:45the world increases, so even if our slice of the pie as professionals gets smaller as more and
- 25:51more people do it themselves, the work available is still considerable.
- 26:00We're not losing work because the amount of texts that are translated is increasing enormously.
- 26:08Now let me go back to the French literary translators who said they are speaking
- 26:13from experience. I went through that long document looking for the actual experience of AI
- 26:21and this is the only piece of evidence that they present.
- 26:26It's by a researcher in Vienna and they give this example where
- 26:40people using machine translation all gave the same translation and the people not using machine
- 26:46translation gave different translations. They call this "anchoring" and this would be the problem
- 26:52of using machine translation: everybody gets anchored to what the MT suggests and
- 27:00therefore you get tremendous homogeneity in the group and your culture becomes very homogeneous.
- 27:07I do think this actually happens so I'm not arguing against the fact.
- 27:16I went to the actual text in question: this is the beginning of "A Very Short Story"
- 27:21by Hemingway and we're interested in this sentence here "Luz sat on the bed".
- 27:29That sentence could mean "she sat down on the bed"
- 27:35or it could mean "she was sitting on the bed". It's just ambiguous
- 27:40and nothing else in the story depends on how you resolve that ambiguity.
- 27:48If it were important, Hemingway would have said "sat down" or "was sitting" but he didn't: he said "sat on the bed".
- 27:55For me, it's logical if you have to make a decision and you're
- 28:05going to invest effort in making the decision and somebody has given you a suggestion
- 28:11and the suggestion is as good as all the other alternatives and the MT says "Luz sat on the bed",
- 28:17that's the one it's logical for you to go for because it's less effort
- 28:22and you lose nothing. So I'm really not upset by the fact that people can
- 28:31solve a problem with the help of automation when there is no good or bad solution to the problem.
- 28:38I'm fine with that.. Lots of terminology problems
- 28:44are solved this way and solved very well. Another point that's raised very often
- 28:51with respect to machine translation and GPT appears in recent work
- 28:59on using GPT3. The sentence is this one: "I'm honored to meet you the future leaders
- 29:07of Great Britain and this world". This is Michelle Obama speaking to young people.
- 29:14The translation into Portuguese for "I'm honored" comes out male, which is wrong, and then we have
- 29:22female future leaders which is wrong too, because there are men and women. We know that
- 29:29machine translation screws up when there are long sentences and when there are pronouns and
- 29:36especially when they're gendered pronouns. MT can be incredibly sexist but here
- 29:41is just a question of being wrong. I translated this into Catalan
- 29:49and it gave me correct genders when I said "Michelle Obama".
- 29:56I specified the context in the prompt. And I said "a group of young women and men".
- 30:06I specified in the prompt and it says feminine and masculine future leaders.
- 30:13So if you do the prompt correctly, the machine will solve the problem really well.
- 30:19The sexism is in the society that generated the databases that go into the machines.
- 30:26I suggest the lack of a specific enough prompt was the problem.
- 30:34But let me show you a different way of solving this.
- 30:39So this is the offensive translation: sexist here and sexist here. The first one is easy
- 30:46to fix up. Just click on there and "es un honor" (it's an honor) and we solve that problem.
- 30:55we can also work on the ST. For example, "future leaders" becomes "who will lead",
- 31:05using the verb, and we make it plural over here, and we change this and...
- 31:21I've got a non-sexist translation just working here in this lovely little DeepL sandpit.
- 31:36What I was doing - I hope you could follow - was just going into DeepL and playing with
- 31:43the alternatives that it suggests. So the the anchoring phenomenon you recall suggested that
- 31:51whatever was proposed by the machine translation would attract everybody to it, but if you can use
- 31:57DeepL this way, modifying the translation here and modifying the start text over here,
- 32:03you can work with those drop-down menus and play with the translation until it solves the problem.
- 32:10So I solved the problem. Instead of saying "I am honored", I put "It is an honor" - I've removed the sexist
- 32:17problem. In "To meet you who will lead Great Britain and the world", I've removed the sexism.
- 32:25Now the problem for me is that this sort of interactive translating that I
- 32:33can do in DeepL online should be put into the mainstream translation memory software.
- 32:41You can do it in CafeTran, you can have the DeepL interface there at the bottom,
- 32:48but I really want machine translation to put that capacity to have the drop-down menus and
- 32:56select and play with it as you're translating into the software so that we can all use it to
- 33:03enhance our creativity rather than be stuck to the anchoring phenomenon.
- 33:12So yes, anchoring is a problem. I think sometimes it does matter but the case
- 33:19the French translators gave is a trivial one. The technology can help us
- 33:24overcome the problem if we get better technology. I don't think we can accuse interactive MT of
- 33:33devaluing rich human practices, as in the list by Lynne Bowker,
- 33:38or of a lack of creativity. I think that what I was doing there was a highly creative process.
- 33:44Let me move on now. We found that GPT works well for the main languages I've been working with:
- 33:57Spanish, French, English, Chinese, German, all in combination with English - no problem,
- 34:03and between themselves no real problem. But we tried it for Armenian: GPT just gives complete
- 34:12rubbish and the neural machine translation is questionable at best. The people in Dublin
- 34:21tried it with Irish and they found that GPT was much worse with Irish.
- 34:29Catalan, although a small language, has a pretty good set of assets and works
- 34:36really well and in fact GPT seems to work directly with English, not going through Spanish
- 34:44as is certainly the case in Google Translate. So it's getting even better with GPT.
- 34:51The problem is that these solutions are working well for the languages with big resources not for the
- 34:58languages with small resources. These are the most frequent languages on the
- 35:05Internet that's English that's Russian that's Spanish French and there's a very long tail of
- 35:11very small languages with limited resources. The answer I think is not that we need
- 35:20less automation: we need more of it; we need more electronic resources. I don't think I've really
- 35:27got time to go into this but the work was done by this group of people working with Icelandic
- 35:34and Hausa - small languages comparing it with French and German and we
- 35:41can see the problem that the GPT is really not doing anything for the smaller languages.
- 35:48What's the answer? More automation, build up the databases, and the machine translation people
- 35:54are very interested in that and getting machine translation to generate databases by analogy.
- 36:03I'll move down very quickly to what GPT is very good at, and Bard, so the large
- 36:10language models in general, and that is adapting a text to a particular audience.
- 36:18This is the text in the prompt "translate". We get that in Catalan and that's a good straight
- 36:27translation. There's no problem with it. If we say "translate for a 12-year-old",
- 36:33and we actually gave this example because we previously used it in class and one of the
- 36:41students realized that when Whitlam went to Canberra he was 12 years old, so that if
- 36:50you say "translate for a 12-year-old" you're inviting personalization of the
- 36:58text for the particular reader as was done by GPT: "born in Kew, Melbourne in the
- 37:07summer of 1916 - so they've adapted July - they say it was a really hot day...
- 37:15Whitlam went to live in Canberra with his father when he was 12 years old. Brilliant! His father
- 37:23"was a lawyer", so that's much better than a "deputy Crown solicitor", too complicated for the reader.
- 37:30Whitlam studied at the school in in Canberra and then at the University
- 37:34of Sydney. It's done brilliantly by adapting this "quan tenia 12 anys", when he was 12 years old.
- 37:45However, as some of you will have noticed, it has adapted the text as if it were in Europe.
- 37:51In Australia, particularly in Melbourne, July is a miserably cold month, ladies and gentlemen,
- 37:58it's not the summer. So beware: the more you ask the technology to adapt, it will make changes that
- 38:08it thinks are cute - it really does seem to think - that it calculates are going to
- 38:14be attractive to the new reader and the more changes are made, the higher the risk there
- 38:20will be an error. As indeed there is there. It shouldn't be summer, it should be winter.
- 38:26You can re-submit the same prompt and you get different answers.
- 38:34Another mistake was made: "Whitlam was born in Kew, a city of Melbourne".
- 38:43I don't know. I've been I've been living there for seven years. I think
- 38:48it's a suburb. It might have been a city I don't know back in 1916, but I certainly wouldn't
- 38:54use that explicitation. It was quite happy just putting Kew, Melbourne. That's all you
- 39:02have to say. The more you try to explicitate, the greater the risk of error and confusion
- 39:11Just some more numbers that can be found. Most of this work is 2023.
- 39:19Comparing GPT with neural machine translation in our own work at the University of Melbourne with
- 39:25colleagues working with Chinese and with colleagues in Spain working with Romance
- 39:31languages and Armenian, we have found that GPT is not much better than what we get from NMT.
- 39:39Here we're going Chinese into English with Google, DeepL and Tencent and you
- 39:48can find here that the highest quality is with Google in this case, with Tencent in this case,
- 39:56and all of these are much higher than GPT. German into Chinese the winner is
- 40:05Tencent and again and Romanian into Chinese...
- 40:12Beware! When you look at these numbers,
- 40:18they prove pretty convincingly that the existing
- 40:24machine translation systems are better than GPT. The new one was there for GPT-4 by the way.
- 40:32The research was done by people working at Tencent, so they happen to pick examples or
- 40:39languages where Tencent came out a winner in at least two of those cases.
- 40:44So there's a lot of playing around in the examples that are being used.
- 40:50This is a similar thing, just showing that as you can see at the top for these things the
- 40:59the machine translation is doing better than GPT. So GPT isn't a threat except there are some
- 41:05cases where it does do better. It is not a big threat to translation. In fact, as translation scholars
- 41:12and translators in the community, we've had time since 2016 to come to terms with this. We shouldn't
- 41:20be completely outraged. I'm moving to a conclusion but I want to step back and have a
- 41:29call to calm. Automation has been going on you know since the Industrial Revolution and before that
- 41:35You can study it over time. Computer -based automation would be affecting professions,
- 41:45occupations, so a sociologist of work just gets a list of occupations in the United States from 1950
- 41:52another one from 1910. We assume that the impact of computers has had an effect. Which is the only
- 42:02occupation that disappeared? Can you guess? Is it translators? No, we still exist.
- 42:08Lift operators or elevator operators in the United States. There are
- 42:16still some around - I managed to find them that's that's the only job because there is a logic
- 42:22by which automation increases productivity which increases demand which increases work so there's
- 42:29not necessarily job loss through automation. That's what we see over the big picture.
- 42:36But what we also find or what is found in these studies is that the more automation comes into a sector
- 42:44the more there is wage dispersal, inequality, that the people who can use the technology to their
- 42:52advantage get richer and the people who can't will stay where they are or get slightly worse off.
- 42:58That's a real question for our industry. I have been fascinated by certain numbers.
- 43:05This is from a European Competition Index that is now a bit dated: over a seven-year period we found
- 43:14that in the translation sector the growth was huge for the big companies and was negative
- 43:21for the small companies. That is worrying. I'm looking for more recent information.
- 43:29I find that as companies get bigger up to here, their profitability increases and out here it's hard to
- 43:36say because there are so few companies that it depends on mergers and acquisitions.
- 43:41Similarly up to this point out here, the bigger the company, the more the profitability,
- 43:49although there's some doubt going on there. So there is some evidence that in general
- 43:59the bigger the company, the better the profitability, and we know from other information
- 44:04that the bigger the companies, the more they invest in the technologies so the technologies
- 44:11I've been talking about here are... Well Trados is now owned by the biggest communication and
- 44:19translation company in the world. It has an interest in exploiting the technology.
- 44:27Final point: intellectual property theft. This has been a problem since the early 1990s with
- 44:34translation memories: Who owns the translation memory? The company or the translator?
- 44:41As I pointed out, I have not found any court cases dealing with intellectual property theft
- 44:47via translation technologies. There are plenty of cases where translators have sued companies
- 44:52and other translators for plagiarism but not via translation technologies so perhaps I'm
- 44:59wrong to look for it. In summary, is there a lack of transparency? Yes and I think
- 45:07we need more transparency and I think it would be a good thing to apply that across the board.
- 45:13Tell us which dictionaries you use; tell us who you are; tell us who helped you. Why not?
- 45:18Let's see translation as a collaborative social activity. Evidence of job destruction? Not yet.
- 45:25I haven't seen firm numbers and the numbers that are published are deceptive. Changing skill sets? Yes,
- 45:33towards augmented translation, as we saw in the audiovisual industry report and I
- 45:40think I can see that just tracing where our graduate students go. AI anchoring?
- 45:48Yes, that happens but it may not have negative consequences and the technology should improve
- 45:55to move to get rid of the anchoring, to make MT more of an aid to a creative process.
- 46:03Evidence of language marginalization? Yes, automation is helping the big languages and not
- 46:09helping the smaller languages enough. The answer is more automation, more language resources
- 46:15for the smaller languages. Wage dispersal? Yes, but indirectly.. It happens because
- 46:25people know how to work with the technology and the wage dispersion seems to correlate
- 46:32more with company size than with anything I can see dealing with numbers about the technology itself.
- 46:38So yes, I think that is happening but the mediating factor is company size.
- 46:43Intellectual property theft? Perhaps it is going on and we haven't heard about it.
- 46:54so there might be an iceberg. There are certainly some very good things at the top. I do enjoy
- 47:00working with these technologies. I enjoy working in that DeepL sandpit and I want to see
- 47:05that integrated into mainstream translation memory technologies. And there are some serious threats
- 47:12that we don't see under the iceberg but there's a lot of misinformation out there and a lot of panic
- 47:20about that hidden part of the iceberg. I think as professionals we have to sit back
- 47:28and see how it really is affecting us rather than enter into panic mode. Find out about
- 47:34the technology, play with it, experiment with it, find out what works for you.
- 47:40As an academic, I have to do research and and find reliable research before making
- 47:47these huge statements about AI being a threat, not just to translators and
- 47:53interpreters but to humanity. I don't know. Not yet. I haven't seen the evidence of that.
About this transcript
This page contains the full transcript of AI and the future of translation. In search of evidence by Anthony Pym, generated from the public captions YouTube serves with the video. The transcript has 5,887 words across 388 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.