CHAT-GPT

[Geoffrey Trousselot]

I know it is not the most cheery of topics, but I'm starting to feel like the latest releases of CHAT-GPT have let the cat out of the bag regarding the near future capabilities of AI. 

I have this image of many translation companies going under like Nokia did when the smartphone hit the market.   

One positive thing being in the Japanese market is that aggressive change is not that common, but I am seeing AI that can write perfect English and be proficient in Japanese at the same time. The CHAT format and the lack of commercial direction of OpenAI are possible hindrances, but the capabilities are undeniable. 

I can't quite see why every white collar worker is not freaking out right now. Am I missing something?

Geoffrey Trousselot

[Warren Smith]

Pretty scary indeed.

 

What scares me the most is that it is the epitome of "group think." (I have run some interesting tests.) Heterodox opinions will be squelched even more effectively through the "authority" of this new AI that codifies and then reinforces through positive feedback opinions that are held by the ignorant masses. I cannot help but think that it will usher in an age of global mobocracy...

 

W

 

---

--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CADJAnY2eUtLcYQBsMrvAteFr%3D6VQcwfWABe_CsxMw92KqgmXMQ%40mail.gmail.com.

[Joe Jones]

You could have said the same thing years ago about typewriters, electronic calculators, word processors, spreadsheet software, etc. These innovations didn't really displace white collar workers but did make it possible for individuals to produce higher-quality and more nuanced work product more efficiently.

For our field specifically, ChatGPT is no better at translation than the existing MT engines like DeepL and Google Translate. It has the same flaws and limitations as those platforms, which boil down to the fact that the engine has no semantic context to understand what it is seeing and is simply matching it up to how equivalent language has been translated in the past. It mimics rather than actually understanding and creating.

Same applies in other areas. In the art field, Dall-E and similar engines draw human hands with weird numbers and positions of fingers because they don't actually know what a hand is supposed to look like and are simply churning large numbers of hand pictures and deciding what they seem to have in common. In the law field, ChatGPT can write a convincing-looking legal argument that is supported by citations, but the conclusions are often totally wrong, and the citations are often nonsense made up by the AI to mimic actual case law citations.

AI will be a very powerful tool for people who want to get more done with less effort, and who are patient enough to review the engines' output, but until we figure out how to actually replicate a human brain using technology, AI will not be a suitable replacement a human who can think critically and creatively.

-Joe Jones

[Dan Lucas]

I agree with the tenor of Joe's argument; a sense of perspective is required.

In this specific case, projects to mimic human speech have been around since ELIZA, and even chatbots have been around for several years now. This latest iteration packages itself nicely with some additional functionality, for which it has received a good deal of attention. Nevertheless, every time a new toy appears people coo (or groan) over the bells and whistles, and the flashing lights, but that doesn't mean that the toy in itself represents a great leap forward.

In more general terms, I'm not arguing that technology doesn't advance, just that it may not advance in the way that the majority expects, achieve the successes that the majority fears, or generate the consequences that the majority predicts.

I thought StackOverflow's (temporary?) banning of ChatGPT was interesting for this comment:

"The primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce."

That's an observation that would seem to apply to MT in our own line of work.

Regards,

Dan Lucas

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/257feb82-4897-4281-aac6-b74823d38fe4n%40googlegroups.com.

[Tom Gally]

A comment on a comment by Joe Jones:

ChatGPT is no better at translation than the existing MT engines like DeepL and Google Translate

I've just started testing ChatGPT's translation ability compared to other MT engines. In at least one respect—consistency of pronoun reference—it does seem better. I put my test results here:

https://www.gally.net/temp/202212chatgpt/translatingwithcontext.html

I need to do a lot more testing to find out whether it is actually better than the other engines overall. The interface is not as easy to use, but it does have a function that the other engines don't: If its first translation into English seems awkward or unnatural, you can tell it to "Rewrite that translation, without changing the meaning, so that it will sound more natural in English” and it will produce something better.

Whether ChatGPT will put translators or translation companies out of business, I have no idea. But I am willing to speculate that it will have a big impact on second-language learning and teaching (my current field, though I still do translation occasionally). Three videos by me on the topic are here, if anyone wants to sit through them:

https://www.gally.net/temp/202212chatgpt/index.html

All the best,

Tom Gally

Yokohama, Japan

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/c73c715a-3bab-4421-9dcf-59f8a3155a03%40app.fastmail.com.

[Dan Lucas]

Many people on this list have been doing translation for many decades. To offer a different perspective as a relatively "new" entrant to the industry, the first time I remember machine translation being discussed on this list was in the huge 'Race to the bottom?' thread of late 2014. That's eight years ago, which is a fifth of a typical 40-year working life, or two-fifths of a second career. Not an insignificant span of time.

In 2014 I had only just begun looking at translation as a possible full-time occupation after a couple of decades in the financial markets, so I followed that thread closely and thought carefully about what was being said. I also did a lot of reading on ProZ.com and other forums. It seemed to the me of 2014 that the threat of AI in translation was overblown, at least in this pair, for the foreseeable future (5-10 years). That seems to have been more or less the right conclusion.

Have we all suffered that much over that time? Anecdote is all that we have to go by, but personally I went full-time in 2015 and my own business has been solid since then. Fluctuations have been few, mild and short, though I have misgivings about 2023 due to the threat of recession.

While several people who were on the list in 2014 are sadly now no longer with us (and it's worth considering the implications for supply, in this exotic and difficult pair, of translators retiring or dying), most people on this list seem to have survived in professional terms. I'm aware that there has been some turbulence in patents, but even those freelancers seem to be mostly still here.

Anybody can point tremulously at the spectre of MT and quaver "Look, my lord, it comes! No, honestly, this time it really does come!". But will it come? And even if it does come, will it displace us or simply augment us? I don't see why translation should be destroyed by MT any more than translation was destroyed by the shift from typewriters to word processing in the 1970s or from unassisted translation to CAT that began in the late 1990s. Of course, if you were unable to accommodate the changes driven by those technologies, you probably struggled. If you were flexible, you got by.

It's hard to take a Panglossian view of the commercial world when one has watched so many companies and industries rise, fall, or change beyond recognition. Nevertheless, even if there is some short-term impact from a cyclical recession in 2023, the outlook for translation in our pair in the medium term doesn't look that bad to me. I am determined to take a cheerful view as we approach the end of another year of work.

Regards,

Dan Lucas

[Dan Lucas]

Anybody can point tremulously at the spectre of MT and quaver "Look, my lord, it comes! No, honestly, this time it really does come!".

Sorry, just to clarify that when I wrote this I wasn't referring to anybody in particular (and certainly not Geoff, whose comments have seemed pretty judicious to me in the past) but more generally to the gloom-and-doom tendency that one sees on, say, ProZ.com forums.

Dan Lucas

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/4e0300da-e290-4e95-a3d9-e04fec2d4107%40app.fastmail.com.

[Geoffrey Trousselot]

The reason I am raising it is because I see immense uncertainty, and I fear that disruptive entrepreneurs are going to have a field day with the existing translation industry in Japan.

The thing about ChatGPT is that it doesn't come with a user manual, and its output is governed by the input given to it. This is just an interaction I had with it this morning. I don't know how best to collect all the possible abilities, but I have a suspicion that they are vast.

Me:

In Japanese, there is the expression 現場主義. I believe in this. I believe it is important because if managers only make top down decisions, they don't appreciate efficiencies or stresses at the employee level. Also employees such as engineers can have amazing plans for the overall system, but their place at the bottom of the management structure stifles their opinions. Put yourself in the place of an employee, who has many innovative ideas. You believe in 現場主義 and you want to express this belief in English to your superior managers.

ChatGPT:
In English, the term that most closely corresponds to 現場主義 is "on-the-ground approach" or "on-the-ground management." This refers to a style of management that emphasizes the importance of being directly involved in the day-to-day operations of a company or organization, rather than making decisions from a distance or relying solely on top-down decision making.

By expressing your belief in an on-the-ground approach, you are advocating for a management style that values the perspective and insights of employees who are directly involved in the work being done, and recognizes the importance of being present and actively engaged in the day-to-day operations of the company. This approach can be particularly beneficial for encouraging innovation and empowering employees to contribute their ideas and suggestions for improving processes and systems.

[Tom Gally]

As Geoffrey notes, ChatGPT doesn’t come with a user manual. It also can't do web searches, which are essential for a lot of commercial translation, and it doesn’t have any APIs that would make it easier to use for repeated translation-specific tasks. But all of those things will be coming soon, either from OpenAI or from others that develop similar software.

Because it already has translation-related abilities superior to those of DeepL et al.—despite not being trained or designed for translation—and because it seems to be able to grasp the meanings of words in context, I think that it will indeed pose challenges to human translators. If I were still a freelance translator, I would start thinking about how I might shift my business into language consulting—working with linguistically naive clients to help them solve their cross-language problems.

Tom Gally

--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CADJAnY1ns1spybsbgpD_MeoxpyN355xrzgnWFcUbgAppsP1hsA%40mail.gmail.com.

[JamesALockhart]

On Sunday, December 25, 2022 at 1:13:50 PM UTC+9 Tom Gally wrote:

Because it already has translation-related abilities superior to those of DeepL et al.—despite not being trained or designed for translation—and because it seems to be able to grasp the meanings of words in context, I think that it will indeed pose challenges to human translators. If I were still a freelance translator, I would start thinking about how I might shift my business into language consulting—working with linguistically naive clients to help them solve their cross-language problems.

When Tom says something like the above, I tend to sit up and take notice.

OTOH, seeing in his blog post that none of the systems understood the nuance of 「 曲を世に送り出し、歌い続けてきた 」, rendering it backwards and not realizing the meaning of the 連動計+続ける, I have a bit more confidence in my future now.

—Jim Lockhart

Yamaguchi

[Dan Lucas]

When Tom says something like the above, I tend to sit up and take notice.

I don't, because I see no evidence that anybody on this list, including Tom, has a track record in making verifiable predictions about the future. This is something distinct from ability in linguistics and translation, which Tom may well have in spades. Creating semi-public forecasts was a large part of my job for nearly 20 years. Frankly, I often got my forecasts wrong (for which I was duly held to account) but at least I acquired a realistic understanding of what the process entails.

Nevertheless, you can make a stab at forecasting if you have a clear starting point. If you're projecting sales for a particular company or the population of a specific country, you have the most recent year of data, for example. That's not the situation that we have here in translation, at all. Even the basics elude us.

• We don't have useful and transparent stats even on the size of this wildly heterogeneous industry (quick now - how large is the annual market for JA-EN translation?)
  
• We don't know the extent to which MT is used in each niche of each sector of each language pair (see above), and how well it performs relative to traditional translation
  
• We don't know the relative change in the usage of MT from one year, five years, or 10 years ago
  
• We don't know the extent to which MT deployments have been successful, and the costs incurred or saved, and we don't know about the failures that have undoubtedly occurred (this is information that most LSPs are going to keep close to their chests)
  

Without such hard data all that we have is anecdote. We share stories about MT like medieval travellers clustered around the fire on a winter's night, whispering about ghosts and boggarts while glancing fearfully out of the window.

Vague statements about the future of AI in the industry are easy to make, free of consequence, and essentially unactionable. If we did have the kind of data described above, I would be challenging individuals to take a view and put their money where their mouth is. 

If we could be fairly confident that MT accounted for (say) 31% of the industry on a value basis in 2022 in the JA-EN pair, I'd be asking for specific forecasts. Oh, so you reckon MT will hit 40%? By when? What effects will it have, for example on rates? How will we measure them? Are you confident enough to pledge $500 to a designated charity if your forecast isn't accurate?

A few years of people attempting such projections and getting it badly wrong might remind the intelligent and educated members of this list what they already know but may be forgetting: creating useful and accurate predictions is bloody hard.

I agree with the general thrust of Tom's argument that translators will need to increasingly add value beyond "simple" translation. I would go as far to say that if you are one of the people on this list still in the game of full-time translation - some are not - and earning a comfortable living, you must already be offering such value.

I should add that naive optimism about the future isn't my bag. My business has been doing well, but I'm perpetually struggling to offer a high degree of availability and capacity, and to balance that with rates and quality. Maybe one day one of those factors will tilt emphatically away from me and I'll be in trouble. Maybe MT will drive that tilt; I don't know.

So yes, all of us who need to work should have a Plan B, and that's the case in any industry. There are very few safe havens or sinecures left out there.

Regards,

Dan Lucas

[Tom Gally]

Dan makes a lot of good points. Thank you.

From the point of view of this former freelance translator, though, quantitative information about general industry trends wouldn't seem very useful when trying to decide whether I would be able to continue making a good living from freelance translation. Instead, I would have to focus on the clients I had and whether I would be able to keep them, or find similar clients, as the competitive environment changed.

Two examples of changing competitive environments from the distant past: When I was freelancing, from 1986 to 2005, there were two big changes to the J-E market in Japan that threatened to drive down rates. One was the appearance of overseas competition from low-wage countries starting in the late 1990s. The other was the arrival of translation memory software like Trados, beginning, if I remember correctly, around 2000 or so. I was active on Honyaku then and I read a lot of discussions about both. After following those discussions, and after looking into both, I decided I didn't need to worry about either.

I knew that living in the Tokyo area, especially being able to meet clients face-to-face occasionally, and the quality of the translation services I provided protected me from low-price overseas competition. (That situation has probably changed since then.)

A lot of translators on Honyaku did adopt Trados and were forced to accept lower per-word rates. However, I was doing mostly small bespoke jobs for direct clients, and consistency of terminology was not as important as it was for large jobs handled by multiple translators. So I was able to ignore Trados as well.

Since becoming a salaryman, I have done only a little translation, some for publication that paid a pittance in royalties and some, especially in the last year and a half, PR-related work for my employer. (I get a reduction in my overall workload, so I am not paid separately for that.)

For the latter, I have taken to running the manuscript through DeepL and working from the output rather than translating the Japanese from scratch. That saves me both time and effort, as, at least for me, the most mentally exhausting part of the translation process is the first draft. The multiple revisions to bring the English into something suitable for public consumption still take time and skill, but my overall output per hour is definitely boosted by DeepL. The quality of the finished translation is the same.

I assume many freelance translators in the commercial market must be using MT as well. Am I right? Has MT therefore led to downward pressure on rates?

If so, then it’s hard to see how even better MT, made possible by more meaning-aware language models like ChatGPT, will not have the same effect.

Tom Gally

--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/e56f44bf-aaef-45a0-a28c-6bf7fe21a020%40app.fastmail.com.

[Kevin Johnson]

One important point to make is that the incentive to say "everything's great" is non-existent, as nearly as I can tell.

I tend to bite my tongue on this topic wherever I see it come up. I've typed and deleted probably a dozen posts talking about how well my business is doing. The reason seems obvious to me: I don't want to encourage competitors. If you think the industry is experiencing downward pressure on rates and everybody's making less money, great. From where I'm sitting, it is actually in my interests for you to think that way.

So I think there is a natural trend for negative voices to dominate the conversation. The people motivated to talk about the issue are those who are suffering, nervous, or not even in the industry. What motivation do I have to poke my head in and say, "Actually, everything's great"? Some kind of perverse pride...? A desire to help others find their best life? Anyway, I tend to choose silence. My inbox is overflowing and I want to keep it that way.

I would suggest that if you're on the outside looking in, you lack crucial information needed to be making predictions. You will also have a damn hard time getting that information, because I am disincentivized from giving it to you!

Dan's right. Objective data is needed. Everybody else is just skimming the most outspoken voices.

Kevin Johnson
Osaka

[Perry E. Gary]

Concur.

--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/7e5cc06b-8e1b-4620-8cf6-f75e4addb66bn%40googlegroups.com.

[Dan Lucas]

Tom, I do appreciate the thoughtful response.

From the point of view of this former freelance translator, though, quantitative information about general industry trends wouldn't seem very useful when trying to decide whether I would be able to continue making a good living from freelance translation. Instead, I would have to focus on the clients I had and whether I would be able to keep them, or find similar clients, as the competitive environment changed.

Ideally, yes. In reality, I argue that as part of that process one needs to make explicit assumptions about the competitive environment, which in functional terms is indistinguishable from the industry.

As previously mentioned, I don't have any objective data about the use of MT in our industry, so I find that approach impractical. I conclude that the evidence base for the threat of MT/AI is not robust enough to justify a switch to another business with all the negatives that would likely entail. The hard evidence I do have is my monthly order inflows, which have been stable and continue to trend upward gently over time (see Kevin Johnson's post from a few hours ago).

The multiple revisions to bring the English into something suitable for public consumption still take time and skill, but my overall output per hour is definitely boosted by DeepL. The quality of the finished translation is the same.

I am glad to hear that. I am all for tools that help the translator become more efficient. By way of corollary, I have been happy to use Trados and other CAT tools, and it hasn't prevented me from earning multiples of the median income for my country. One of the reasons for this is that I get plenty of business, and I get plenty of business in part because I have a positive approach to the use of CAT tools.

All private-sector industries are under pressure to cut costs. By squeezing out boilerplate the agency and the freelancer can work together to reduce the expense of translation for the client, and that in turn stimulates demand (ceteris paribus). I see that as a good thing. Freelancers who relied on translations of text that have been reduced to boilerplate will no doubt have experienced it as a bad thing. If the text I translate goes the same way, obviously I will be unhappy.

I assume many freelance translators in the commercial market must be using MT as well. Am I right? Has MT therefore led to downward pressure on rates?

Others are better placed to comment on this, because I really don't know. My clients have taken a look at MT (in some cases, a very close look) and seem to have concluded that only limited adoption makes sense, if they have adopted it at all. Maybe it is their clients that are behind this abundance of caution, or maybe it is just that MT performs badly in more complex contexts. I don't think any of my existing clients permit the use by freelancers of MT such as DeepL or Google Translate, and most specifically prohibit it in their service-level agreements.

Nevertheless, I would be happy to experiment with MT if my clients were more open to it. Translation is a business for me, and I'll consider almost anything if it seems likely to enable me to maintain or enhance my earnings.

Regards,

Dan Lucas

[Dan Lucas]

On Mon, 26 Dec 2022, at 08:58, Kevin Johnson wrote:

What motivation do I have to poke my head in and say, "Actually, everything's great"? 

Kevin, I feel the same way, and I too find myself biting my tongue a good deal. Being a somewhat combative personality by nature this is not easy for me, and occasionally I am unable to restrain the urge to take a poke at the unwarranted consensus that freelancers are struggling and, in effect, doomed. The declinists and catastrophists don't speak for me, and I don't find their unsupported assertions persuasive.

Please note that I am not including Geoff (who started this thread) or Tom in that group! I may disagree with certain of their opinions, but they have made their points well and clearly, and without the corrosive certitude that characterises the statements of some people. Readers are free to conclude what they will based on the arguments that have been put forward.

Regards

Dan Lucas

[Warren Smith]

I don't know about everybody else, but my income in nominal dollars is down from about $300k in 2006 to about exactly half that last year (with the second half of the year substantially worse than the first half of the year).

 

This makes me extremely nervous, especially as there has been 48% inflation in the US since 2006, meaning that my effective income is down by a whopping 2/3 since then. Regardless of where an income starts, this is a difficult adjustment to make, and the trend to reduced income seems to be accelerating.

 

I don't know if that is because I am in the wrong specialty (patent translation), or because patent translation work has shifted more to Japan from the US, because my particular clients are getting less work, because there are fewer cases in front of PTAB, or what, but I do know that I am seeking other options for employment right now, and am considering taking a job that is yet another 35% cut in income (from last year's level), just to have a salaried position that will be more stabilized.

 

W

 

---

--

You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/5c48fad8-40f6-4771-b2a9-646ddc5e2d48%40app.fastmail.com.

[Dan Lucas]

On Mon, 26 Dec 2022, at 13:35, Warren Smith wrote:

I don't know about everybody else, but...

Warren, that's interesting, thank you. Clearly you've been an outlier on the upside. I would conjecture that those excess returns stemmed partly from a very high degree of specialisation (and competence!) during a time of abundant JA patent activity and partly from being one of the early pioneers of digital tech, CAT tools of a home-made variety, and so on. Nothing lasts for ever, but it sounds like it was fun while it lasted. Good for you.

However, if you're still making $150k then you're doing pretty well. I wouldn't let the decline you describe panic me into another line of work if I were not already fed up of translation. Particularly in the US, would a salaried position really be more stable? As far as I understand it you can be canned at any time, unless you can find work in a unionised part of the public sector. And with tenure in decline, areas like tertiary education are no longer as attractive as they once were. 

If you were to be on roughly $100k then in most parts of the country you would be regarded as a high earner, and that would make you a potential target if your new employer were to need to cut costs.

The period of time you mention is quite long. Sounds to me like you had a peak, then a lower peak, and that you're rumbling along at a lower level still. But what does it look like if you chop it up into shorter periods, and take the median income for 2012-2016 and for 2017-2022? (I'm not suggesting that you disclose the figures, just that you do the sums if you haven't already.)

My thinking is that if you've been fairly stable at this lower level for several years, you shouldn't be voluntarily giving away another 35% of your income to take up a post that may have no legal guarantee of stability, and which would involve the rapid and irreversible decline of your translation franchise. You won't have the bandwidth to pin down a demanding job while also maintaining your status as a high-end translator.

Regards,

Dan Lucas

[JamesALockhart]

Thanks, especially to Dan Lucas and Tom, for the informative exchange. I really appreciate the extended elucidations.

Jim Lockhart

Yamaguchi

[cpta...@ozemail.com.au]

Is it too late to weigh in on this?

Dan and others make many good points, but I have been reading most of these posts and it sounds like another world, or at least another industry to me. AI/MT has probably driven changes to my business over the years, but like the invention of chairs, pens, dictionaries, PCs and the internet it has contributed positively much more to how I invoice people for translation services.

I think if your income is seriously threatened by AI/MT then it might be that your entire range of business activities is mechanical and highly vulnerable to take over by machines: only written translation, only one language pair and direction, mostly through agencies, no diversification or employees, focusing on turning over as many sentences in as little time as possible and then rolling your eyes and complaining about the stupid in-house checkers.

Yes, when they send the robots out hunting they are coming for you first.

I struggle to think of a part of my business that could be taken over by AI/MT. When my client gets an invoice from me they are paying for the  transfer of all the risks of miscommunication in a multilingual setting to a commercially accountable entity, who gives a shit about their interests, is a god and not a slave of language, who will whip the ass of ANY armchair, risk-free, in-house checker who gets it wrong, and when they get it right we will show them how we are even BETTER than them at discovering and revealing and correcting our own mistakes.

AI/MT can do none of those thing. You need a business model, or at least a definition, like that.

 

Here’s a short piece of advice I wrote seven years ago for client.

https://pooletranslation.com.au/file/3296/371

 

Chris

--

You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/db3669ba-8bce-4db4-b60e-4ace941b9511n%40googlegroups.com.

[BTB]

I think ChatGPT is just another step in the evolution of translation tools and technologies that will have an unpredictable impact on the livelihoods and roles of human translators (because it is new) , but people will always be in the translation loop for high-value (risk) content. Even ChatGPT doubts itself as a game-changer for translation. 

"It is difficult to accurately assess the potential impact of ChatGPT or any other language model on the translation industry without more context. Language models such as ChatGPT can be useful tools for generating translations, but they are not intended to replace human translators. Translation is a complex task that involves not only converting words from one language to another, but also conveying meaning, tone, and cultural context accurately and appropriately. Language models can assist with certain aspects of translation, but they do not have the ability to fully understand and interpret the nuances of language and culture in the same way that a human translator can. Ultimately, the role of language models in the translation industry will depend on how they are used and integrated with other tools and resources." (written by ChatGPT)

Brian Boland 

California, USA

[Dan Lucas]

On Mon, 26 Dec 2022, at 20:18, cpta...@ozemail.com.au wrote:

Dan and others make many good points, but I have been reading most of these posts and it sounds like another world, or at least another industry to me.

It's not another industry Chris, but you're making a category error within the same industry. You're erroneously conflating the process of managing translation with the process of performing translation.

Based on past comments, including those in a previous "Price Erosion!" thread in June 2022, your panacea for a freelancer struggling in this market seems to be for them to switch to outsourcing work, managing other translators, branching out into other pairs, and marketing this basket of services to end clients. To become, in effect, a small agency, as summarised by you here: ↓

I think if your income is seriously threatened by AI/MT then it might be that your entire range of business activities is mechanical and highly vulnerable to take over by machines: only written translation, only one language pair and direction, mostly through agencies, no diversification or employees, focusing on turning over as many sentences in as little time as possible and then rolling your eyes and complaining about the stupid in-house checkers.

If the agency approach suits you then that's fine, but as I pointed out in that June thread, success in this area requires a completely different skill set (and usually a different kind of personality) to that of the typical freelance translator. Many people do not enjoy business processes and want nothing to do with managing other people. There's no point in saying "Oh, don't worry, it's easy enough once you get used to it". It's a material difference, and effectively a change of career.

Additionally, I don't see that your proposal mitigates overall risk. In the scenario implied by your thesis, you end up not avoiding competition but engaging in competition with a different set of entities. You're no longer threatened by MT, agreed, but the trade-off is that instead you're threatened by every other LSP on the planet.

When clients want to know your unique selling point, are you going to say "I will whip the ass of ANY armchair, risk-free, in-house checker who gets it wrong"? Because every other agency can and will say that. How do you argue that you are better and more efficient at managing this service than, say, Transperfect, or (slightly closer to you) Straker, or any other top 100 agency?

As far as I can see, your approach requires swapping the bundle of risks faced by freelancers for the bundle of risks faced by a small agency. Again, if the latter is preferable to you then that's great. 

However. If you are the sort of freelancer who doesn't want to engage in agency-type operations then your advice isn't particularly useful. To put it another way, your response may be valid, but only in answer to a different question, namely something like "What can I do in the field of commercial linguistics, other than translation, to make money?"

Regards,

Dan Lucas

PS I have often wondered where the successful freelance translators who enjoy and are good at the business side of the industry go, and have come to the conclusion that traditionally they are the ones who go on to start boutique agencies

[Andreas]

As to Warren’s predicament, this article may shed some light on the matter.

https://xtech.nikkei.com/atcl/nxt/column/18/00434/022500028/

 And yes, MT does seem to play a part (rants in denial always make me snicker). Apparently their peculiarly linear style makes patents a prime target for MT application. According to some records, rates for patent translations in Japan go as low as 8 yen per word.

[Warren Smith]

Thanks. This was very interesting.

 

W

 

---

[Dan Lucas]

Agreed, thanks to Andreas.

The writer only has data up to 2019, so it doesn't address the issue of the sudden drop in patent litigation from 2019-2022 raised by you in an earlier email from December 3. I know it sounds simplistic, but do you think that perhaps that could have been driven mostly by COVID and its aftermath? I can't see why COVID would have affected 2019 so obviously? Economies were already decelerating in 2019 though...

Regards,

Dan Lucas

--

You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.

To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/F9CEE6C10A254A27B55D12B981D81AA6%40WarrenSmithDell.

[Joe Jones]

Just to contrast with some of the comments on patent translation, I work in the corporate legal field (contracts, due diligence reports, litigation materials) and that field is still very active and commands decent rates. Some clients expressly forbid the use of MT or even translation memory because they don't want excerpts of their sensitive documents (much less the entire documents) floating around on third-party servers, and some clients have specific style and usage preferences that off-the-shelf MT is simply unable to consistently replicate, so I don't see the human touch going away for this kind of work anytime soon. I believe there are vendors trying to come up with secure MT/TM solutions for this kind of work, but it's tricky to make it work well in practice.

Joe Jones

[Jon Johanning]

@Joe Jones

"Tricky" is putting it mildly. I've been reading up on the subject of "artificial intelligence" (or maybe it should be called "arty intelligence") lately, and mostly what I see is experts in the field pointing out how lousy it continues to be with anything having to do with natural language tasks, despite the humorous chatbot samples we keep being regaled with.

Computers wrestling with natural (i.e., human-produced) language have certainly wiped lots of work we used to do in the good old days (can't compete with free, instant "translations," no matter how good or bad they are), because they can simulate the real stuff moderately well in some respects, but we just have to search for consumers of translation who recognize that if you want good furniture, don't go to Ikea.

Happy New Year, all!

Jon Johanning

zenn...@icloud.com

www.johanning-translations.com

[Warren Smith]

This matter of security with MT is (or at least should be!) a huge issue in patent translation as well (especially when it comes to translation for filing).

 

I address my article that was recently published in a legal journal. (If you have not read it, I would hope that you would, as I am trying to use it to educate clients about issues like this.)

https://law.unh.edu/sites/default/files/media/2022/06/smith_bernard_pp43-75-sm.pdf

 

Unfortunately, this article has not generated as much traction (and business!) as I had hoped it would.

 

Warren Smith

 

 

---

Joe Jones: "Some clients expressly forbid the use of MT or even translation memory because they don't want excerpts of their sensitive documents (much less the entire documents) floating around on third-party servers..."

[John Stroman]

Andreas commented, "Apparently their peculiarly linear style makes patents a prime target for MT application." 

 

I want to insert my two-cents on J>E patent translation. We need to understand that there is a difference between:

 

(A) patent translation for information, and

(B) patent translation for filing in an English-speaking country.

 

(A) is a case in which an English-speaking customer, usually a company, is trying to gain information about its competitors' inventions from published Japanese patent applications. This type of translation doesn't require human translators with subject matter expertise because the purpose is to get the general drift. (If the customer suspects the content affects its own R&D or infringes on one of its own patents, it will then turn to an expert human translator for a detailed translation.) Patent translations for information used to provide lots of work for J>E translators. In today's environment, customers/agencies take text files of published Japanese patent applications and run them through machine translation, sometimes with post-editing. This is a good example of technology taking jobs away from human workers.

 

(B) is a totally different beast. This is usually a case where a Japanese customer, usually a company, has already applied for a Japanese patent, but the patent application has not been published yet. The applicant wants to file an application for the same invention in an English-speaking country to become "first to file" before the Japanese application becomes public knowledge (18 months after filing in Japan). A great amount of money rides on whether a patent application will be approved or not. The translated English version is a final product that is to be reviewed by a patent examiner. It doesn't make sense for the applicant to compromise on the quality of translation to save a small fraction of the total cost. Moreover, the applicant cannot afford to risk leaking information by using online services, and in case there is litigation involving the translated application, the applicant may need a human language and subject-matter expert to appear in court.

 

To people who are familiar with CHAT-GPT:

 

I have the following questions.

 

1) Is it free?

2) Can you use it on a PC (Win 10)?

3) Can you upload an entire document or do you need to upload one sentence at a time?

4) Can you upload using copy/paste or do you need some advanced function?

5) How easy is the process (number of steps, waiting time, etc.) compared with something like DeepL or Google Translate?

 

I've never seen CHAT-GPT, and I appreciate your input.

 

John Stroman

----------------

[Geoffrey Trousselot]

To people who are familiar with CHAT-GPT:

 

I have the following questions.

 

1) Is it free?

I think it is best to go to openai.org and check it out yourself. The latest CHAT-GPT is a beta service in which users can use it for free, but Openai is bearing the considerable cost of providing it. I think there are currently 1 million users and it is not feasible to scale that up at the moment.

 

2) Can you use it on a PC (Win 10)?

It is a browser based service, but I think you can set up scripts to communicate with it using bots, providing all the log-in authentication is properly completed and the security of the connection can be established.

 

3) Can you upload an entire document or do you need to upload one sentence at a time?

I don't know. I have just delivered my latest novel, and I have just been using it to offer suggestions for alternative phrasing.

 

4) Can you upload using copy/paste or do you need some advanced function?

You have to ask it to do something. For example: 

Me: 

Please translate the following Japanese into English: "立憲民主党（露: Конституционно‐демократическая партия）は、かつて存在したロシアの自由主義政党。ロシア帝国末期の国会において有力政党として種々の改革を要求し、二月革命後の臨時政府では政権を担った。略称はカデット。「教授の党」とも呼ばれたように、大学教授や弁護士、貴族などを中心とした知識人の政党だった。

「超階級」的な立場を標榜したこの党は、労働者の権利保障、女性参政権の実現、ユダヤ人差別の廃止など、当時のロシアにおいてはかなり先進的な内容の政策を掲げていた。体制に対して非妥協的な政党だったとも逆に「反革命的」な政党だったとも評されるが、いずれにせよ、カデットの構想は、当時のロシアの現実に適さない面があったと考えられている。"

Chat GPT:

The Constitutional Democratic Party (Russian: Конституционно-демократическая партия) was a liberal political party in Russia that existed in the past. It was a powerful party in the Russian Empire's parliament and demanded various reforms, and it held power in the provisional government after the February Revolution. It was also known as the Cadets and was a party of intellectuals, such as professors, lawyers, and nobles, who were sometimes called "the professors' party."

This party, which claimed a "superclass" position, advocated for the protection of workers' rights, the realization of women's suffrage, and the abolition of Jewish discrimination, among other relatively progressive policies for Russia at the time. It was also seen as an uncompromising party towards the system, or alternatively as a "counter-revolutionary" party, but in any case, it is thought that the Cadets' vision did not fit the reality of Russia at the time.

 

5) How easy is the process (number of steps, waiting time, etc.) compared with something like DeepL or Google Translate?

 

I am sure you can automate it. But the automated script would have to use natural language requests (I think)

 

[John Stroman]

Thank you for the detailed information, Goeff,

CHAT-GTP appears to be very impressive, but I will be unable to use it in my work right now for a number of reasons that are  mainly imposed by contracts with my current customers.

John Stroman 

----------------

[Tom Gally]

A bit more about ChatGPT and related software:

Earlier this year, OpenAI created similar waves when they released the text-to-image software DALL-E 2. At first it was free to registered users, but then they started charging for it. I played with it for several weeks and spent about 45 dollars. You can see some of the results I got here:

https://www.youtube.com/playlist?list=PLj4urky_8icRPzgFS_b98Ks_dPDkNOCvc

Not long after DALLE-2 came out, similar software, including Stable Diffusion and Midjourney, was released. Developers have released downloadable versions that you can run on your local computer without connecting to the web. I downloaded and installed a free one called DiffusionBee a few days ago and tried it out. It’s a bit slow on my Mac, but it works. Some of the pictures are pretty good, while many are mediocre or strange. Below is a strange one I just made with the prompt “A translator cat worried about competition from AI. Hard Edge Painting.”

https://www.gally.net/temp/20221231Translatorcat.png

Perhaps even more so than ChapGPT, the quality of the output from these text-to-image models depends a lot on the prompts, and people are only now starting to figure out what kinds of prompts work best for various purposes. A new term, and possible profession, has emerged: prompt engineering.

I don't think OpenAI has announced their plans yet for ChatGPT, but I suspect that they will start charging for it at some point soon. When they do so, they will probably also start offering APIs to customers so that the generation of prompts can be done more efficiently for specific purposes such as translation. It might also be possible for customers to tweak the model itself for specific purposes. The president of OpenAI, Sam Altman, suggested these things in an interview a few months ago:

https://www.youtube.com/watch?v=WHoWGNQRXb0

When software comparable to ChatGPT, either free or commercial, will emerge is unclear, but it almost certainly will before too long. EleutherAI (https://www.eleuther.ai/) is one company already releasing similar large language models, and Google is reported to have something as powerful as ChatGPT that they haven't released yet. OpenAI is also said already to be preparing a more powerful language model than the one used in ChatGPT.

Whether OpenAI will release a downloadable version that might allow translators under confidentiality restrictions to use ChatGPT on their local machines is unclear, but it seems likely that someone will release something similar in the not-too-distant future.

Tom Gally

--
You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CAJVTx96SSbmUQvuf4TLnapcbLbpHYTMLCY%2BRggjQsf8iy34nqg%40mail.gmail.com.

[Jon Johanning]

My bit of wisdom, at the end of another eventful machine translation year (Happy New Year!), is that if a translator can’t do a professional job with the aid of all of the apps besides chatbots that are now available, it’s time to look for another line of work. Perhaps building self-driving vehicles that can really drive themselves, or the next CEO of Southwest Airlines (I think that position will open up pretty soon).

I seriously doubt that software of this type that will be useful for translators will come along very soon, if ever. My chief guru on this subject is Gary Marcus, a leading cognitive scientist at NYU, who has a Substack newsletter “The Road to AI We Can Trust,” and author, with Ernest Davis, of the book Rebooting AI: Building Artificial Intelligence We Can Trust. He calls GPT-type software “the king of pastiche,” very good at “seeming so brilliant one minute and so breathtakingly dumb the next.” He gives many examples of pure nonsense that it can unwittingly push out, including this one: 

 Question: “Which is heavier, 10 kg iron or 10 kg cotton?” 

Answer: “10 kg iron is heavier than 10 kg cotton. Iron is typically heavier than cotton, so the answer is 10 kg iron is heaver.” (https://garymarcus.substack.com/p/how-come-gpt-can-seem-so-brilliant)

(See also: https://garymarcus.substack.com/p/is-chatgpt-really-a-code-red-for and https://garymarcus.substack.com/p/an-epic-ai-debateand-why-everyone.) My inexpert thoughts: https://www.johanning-translations.com/post/what-we-can-learn-from-ai-grappling-with-language.)

 His basic message, echoed by many other experts on the subject, is that these AI things just have no knowledge of the real world at all or any comprehension of language; they are neural networks which are trained on huge collections of stuff swept up from the internet, etc., and constructed so that they can stitch the stuff together in ways that look as though they were written by knowledgeable humans, when in fact they are just tapestries of fragments of the materials they were trained on. And we know how much junk the internet is loaded with.

 AI researchers are still working feverishly after years and years on this problem of enabling computers to learn, not from being fed “training data,” but from observing and living in the real world, which all of us humans do naturally. I doubt that they will solve it very soon. Meanwhile, one thing that translators certainly need to excel in is knowing a lot of things about the real world. At least, if you're going to try out these pieces of software, be sure to ride herd on them very carefully. As always, revise your translations and give them a human touch before turning them in.

 (Another very informative skeptical book on AI is Erik J. Larson: The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do.)

 

Jon Johanning

[Matthew Schlecht]

On Sat, Dec 31, 2022 at 12:51 PM Jon Johanning <jjoha...@igc.org> wrote:

 AI researchers are still working feverishly after years and years on this problem of enabling computers to learn, not from being fed “training data,” but from observing and living in the real world, which all of us humans do naturally. I doubt that they will solve it very soon.

Well, educated guesses are that it took "us" (Homo sapiens) around ~230,000 years to develop that capability, from ~300,000 BCE (earliest H. sapiens fossils) until ~70,000 BCE (rough date of the cognitive revolution, according to Yuval Noah Harari); even longer if you count the time put in by other hominids pre- and co-evolving with us.

At that point in time, a random mutation fostered cognition in our ancestors, and allowed them/us to develop abstract thinking, cooperation among larger groups, and language, among other faculties.

That chance mutation is akin to what AI watchers label the "singularity", when machines and/or software become self-aware.

Matthew Schlecht, PhD
Word Alchemy Translation, Inc.
Newark, DE, USA
wordalchemytranslation.com

[Perry E. Gary]

I’m an admitted amateur.  Seduced by all the hype, I’ve tried to download ChatGPT with no result. Your link gets “This site can’t be reached.” Searches on both Yahoo and Google educe the message, “Your browser is out of date”. (Not true in other searches.)

 

Am I doing something wrong, or is there another way to access ChatGPT?

 

Many TIA,

Perry E. Gary

 

From: Geoffrey Trousselot

Sent: Friday, December 30, 2022 10:21 AM

To: hon...@googlegroups.com

Subject: Re: CHAT-GPT (Reduced Income)

 

 

To people who are familiar with CHAT-GPT:

--

You received this message because you are subscribed to the Google Groups "Honyaku E<>J translation list" group.
To unsubscribe from this group and stop receiving emails from it, send an email to honyaku+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CADJAnY05744o8_PSG2xD0gkaOAe1kmvS5W5aQ77fpCYE84jY-A%40mail.gmail.com.

[Tom Gally]

Perry,

Try the following link:

https://chat.openai.com/chat

You should get a page that says “Welcome to ChatGPT. Log in with your OpenAI account to continue,” with an option to either log in or sign up. The sign-up process for OpenAI should still be free; I'm not sure what information they ask for.

However, for the past few days ChatGPT has sometimes been “at capacity” and unresponsive. They’ve also added some kind of browser security check before the initial log-in, presumably to prevent abuse. If you continue to have problems, you might try again later and maybe use a different browser (Edge, Firefox, Safari, Chrome, etc.).

Also, ChatGPT is not, as far as I know, downloadable at the moment. After you log in, you enter prompts in the browser window and ChatGPT responds in the same window.

Tom Gally

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/546D4292CCAD418B87C106BC55AC7396%40PerryEGaryPC.