J-E translations done by Japanese people with native checks

[Michael Hughes]

One of Bill's posts on LinkedIn got me thinking about a question I've always wondered about - roughly how much of remunerated J-E translation is still carried out by native Japanese?

I get the impression it's a lot less than it used to be in decades gone by, and definitely the exception rather than the norm in creative fields. It still seems pretty common, if not dominant, in technical fields though. Japan seems to be a bit of an outlier in that regard when it comes to following the industry convention of only translating into one's native tongue. According to the latest JTF white paper, around two thirds of the demand/revenue in the Japanese<>English pair is for the J>E direction. So demand likely exceeds what native English translators with the appropriate subject knowledge​ can supply. 

One of the reason rates seem to be 30-50% higher in the J-E direction also seems to be (along with skill level) the speed at which native Japanese can realistically do it - some surveys I've seen suggest no more than 600 chars an hour at the top end (whereas highly experienced English natives seem to claim somewhere in the region of 1000 - admittedly my sample for that is small).

I suppose where you have a native Japanese field expert with excellent written English you are probably safer going down that route (+ native check?) than hiring an English native translator with little to no field knowledge. 

Still not sure I fully get "native check" though to be honest - if the checker doesn't read Japanese and can't understand the ST then how can they correct anything but the most glaring grammatical errors....?

Michael Hughes

[Bill Lise]

I don't have hard figures, but a look at the materials produced in English by and for Japanese entities still clearly points to a large portion of JA>EN translation in Japan being done by Japanese natives. This is evident even when it is proofread <sic> by native English speakers. Raw JA>EN by NJS translators without checking, and editing/correction is much rarer these days than when I started in this business.

I agree that subject-intensive industrial or other technical JA>EN translation is probably more heavilly populated by NJS than NES translators. Japanese companies are more likely these days to have an in-house foreigner check things before they go live online or into print.

I agree that there would not be enough capable NES translators to hand the demand. The boom in Japanese learning some decades ago seems to have settled down, albeit with a possible surge anew among young people interested in Japanese popular culture. But they are not going to be front-line translators any time soon, and perhaps never, the way things are going.

And one area I think which is a bastion of NJS-produced JA>EN translations is scientific papers, including for medical journals. I have asked people who write for such publications why researchers translate their own paper into English and the answers vary. Some say that they cannot trust an arms-length translator (or agency) to be able to understand what they wrote. Several said that the getting the paper published in English was essential for them but that they couldn't get funding and couldn't personally afford a translation service. I believe there are foreigners here and elsewhere making their living editing papers for Journals.

The terms native check (from the Japanese ネイティブチェック) offers the katakana advantage; it sounds meaningful and yet doesn't come with much definition of the required or recommended qualifications of the "natives" enlisted in such work. One of my former colleague Russian intercept operators was pretty much useless in Japanese when we were here and never learned any more afterwards, told me he gets English to correct sent to him in the US.

Side comment [1[ on demand, if the JTF estimate is that two thirds of the demand/revenue in the Japanese<>English pair is for the J>E direction, they are surely talking about Japan. In places like the US, it would be nearly all of the demand, because the market is dominated by reader-driven work for English readers. Most writer-driven work is done in Japan, making the repair work more important.

Side comment [2] on the ability of NES translators to translate, when I chaired the JAT board many years ago, I attended a JTF event one time and exchanged business cards with a Japanese bureaucrat somehow connected the JTF's recognized status and he apologized for not having an English business card. When I told him it was not a problem, he asked me whether I could read even kanji. I trust the government people the JTF asks to their events are a bit more clued-in these days.

Bill @ rainy Yokohama

[Tom Gally]

Bill wrote:

And one area I think which is a bastion of NJS-produced JA>EN translations is scientific papers, including for medical journals. I have asked people who write for such publications why researchers translate their own paper into English and the answers vary. Some say that they cannot trust an arms-length translator (or agency) to be able to understand what they wrote. Several said that getting the paper published in English was essential for them but that they couldn't get funding and couldn't personally afford a translation service. I believe there are foreigners here and elsewhere making their living editing papers for Journals.

That’s exactly correct.

During my two-decade interim in academia, I was involved in many projects aimed at improving the English academic writing skills of Japanese researchers, including developing and managing a large undergraduate academic writing program (which is still on-going, with over three thousand students a year), giving seminars on research writing to graduate students and researchers, supervising research on academic writing, and serving as chair of the 学術英語学会.

The focus of all of those projects was to try to enable Japanese researchers to write their papers in English, not to translate them from Japanese. While some researchers do have their papers translated, many fields are so specialized that it is difficult or impossible to find competent translators. Japanese research papers also use different organizational and rhetorical strategies, and papers that are directly translated from Japanese—even very skillfully—will often be rejected by international journals because they seem weirdly structured or make their arguments in unconventional ways.

It’s much better to try to teach the researchers to learn and follow the argumentative and rhetorical strategies accepted internationally in their fields. Also, especially at the undergraduate and graduate levels, an additional goal of academic English education is to enable the researchers to take part actively in the international scholarly community, which includes attending conferences, giving presentations, working in internationally staffed laboratories, etc. Researchers whose English is so poor that they have to write their papers first in Japanese and then translate them (or have them translated) into English are unlikely to become internationally productive scholars.

Of course, the papers written in English by nonnative speakers still have language problems, but those can mostly be fixed by native checkers who can't read Japanese. There are some native-English checkers in Japan who make their living doing that, but most of the business has shifted overseas. I did some consulting work years ago for a company based in Mumbai that has a large number of freelance checkers located in India and elsewhere as well as a well-designed system for processing manuscripts and sending inquiries to the authors. The Japanese researchers I know who have had their papers checked by the company have been satisfied with the results (i.e., their theses were accepted and their papers got published).

One exception to this focus on writing-first-in-English is the publication of scholarly books that have been translated from Japanese. The Japanese government and some private foundations provide funding to support such publication, including translation costs, and I know at least one small publisher that is able to pay the bills by publishing those books. There might very well be people on this list who have translated such books. Most of the translated books are in the humanities or social sciences; researchers in the natural sciences do not write books much even in Japanese, and they are the ones under the most pressure to publish papers in English.

In my semiretirement, I still get asked to lead seminars on research-paper writing. In the next couple of months, I'm going to be giving one to a university in Kansai and another to a social-science research institute near Tokyo. Needless to say, advances in AI tools are changing the landscape drastically, and I'm trying to decide how to show the researchers how they can use ChatGPT, Claude, etc. to write their papers without going over the—still quite unclear—ethical boundaries.

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/8cec55e9-a8b0-417e-ac5e-e2c2bfb74ca3n%40googlegroups.com.

[Dan Lucas]

This has been an interesting discussion, thank you everybody.

In my niche - financial documents for publication, or internal documents of a sensitive nature - the norm appears to be translation by native speakers of English and checking by native speakers of Japanese with a good grasp of English. On the whole, this works well. I get the occasional exasperating question that betrays a lack of understanding of English on the part of the checker, but it is occasional.

On the other hand, even in this area I still see signs of translation by Japanese and/or unqualified translators. One of the things that strikes me is that some native speakers of Japanese still seem to believe that if they really, really want it to happen, English will magically take on the meaning that they would like it to have.

Case in point: yesterday I found that somebody has (again) translated the heading for a time series of interim (i.e. half-yearly) sales as "Quarterly" because 四半期 appears in the Japanese. I had to point out (as I always do) that a period of six months cannot be referred to as "quarterly", and provide links to various dictionary definitions for "quarter" (as usual).

I went on to note the 累計 in the source and demonstrate how this can be added to the English (e.g. "cumulative sales for the first two quarters") to arrive at a compromise. I am willing to bet that nothing will change and that in six months' time I will be faced with a chart showing six-monthly earnings and headed "Quarterly consolidated results".

Why? Because somebody, somewhere along the editing chain, would rather have demonstrably incorrect English published to the entire world, at considerable expense, than admit that they were wrong.

This is what tends to happen in a deeply hierarchical and risk-averse society. Nobody wants to lose face. Nobody wants to go their boss and say "Bucho, I'm sorry, but this really isn't correct. We need to change it." There are good reasons for the very long and dishonourable tradition of poor English among Japan's companies, and the roots go down deep into the national psyche.

There is no easy solution. The cultural landscape massively raises the cost of making mistakes and of failure, and this affects things like startup formation as well. It's less of an issue in a stable economic milieu where growth requires doing more of the same, but a little bit better. It's a potentially disastrous factor in a fast-changing business environment where growth (and possibly survival) requires doing something completely different and untried.

With regard to academic papers, a semi-retired acquaintance of mine has for many, many years made a living by editing and rewriting documents for Japanese academics. He says that demand has plummeted recently because his clients have found that ChatGPT and its ilk seem to do a good job of rewriting poor English. Maybe the software makes mistakes, but his clients seem to think the trade-off is worth it.

Regards,

Dan Lucas

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CACnS3NeE1DX9jy3fS29-TW%3D7E4xOPMO9pXPkiTZfccz6wk%3DqOw%40mail.gmail.com.

[Tom Gally]

Dan wrote,

With regard to academic papers, a semi-retired acquaintance of mine has for many, many years made a living by editing and rewriting documents for Japanese academics. He says that demand has plummeted recently because his clients have found that ChatGPT and its ilk seem to do a good job of rewriting poor English. Maybe the software makes mistakes, but his clients seem to think the trade-off is worth it.

Very interesting. I hadn't tested ChatGPT for that purpose yet, so I did so now. The results are here:

https://www.gally.net/temp/20231107gptassistanttest/test4.html

That’s an excellent rewriting job, if you ask me. It would have taken me hours to produce anything close to that quality. GPT-4 spat it out in about twenty seconds.

Tom Gally

[Dan Lucas]

I find that these models write well, but with a distinct style that is usually easily identified.

I don't see that being a problem for academic papers, however.

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/CACnS3NfQ5K%3Du7DWFvdfMiGky%2B4OUNSMp%2BU2WxxM8CXCpyON%2BhA%40mail.gmail.com.

[John Stroman]

To respond to Dan's comment "I don't see that being a problem for academic papers, however."

The scientific journal Nature has published a series of articles on the potentially adverse effects of using AI to generate research papers submitted for publication.

Here is a link to a short article entitled; ‘ChatGPT detector’ catches AI-generated papers with unprecedented accuracy

https://www.nature.com/articles/d41586-023-03479-4

By scrolling down the page you can see the titles of several related articles on the Nature website although I believe they are all behind a paywall.

John Stroman

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

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/0b1b1ba1-20db-4fe3-a69f-4d5ec561984e%40app.fastmail.com.

[Dan Lucas]

AI detecting AI.

As the saying goes, it takes a thief to catch a thief...

Thanks to John for that.

Dan Lucas

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CAJVTx951UyJm%2B%2Bb9Hx8hqK8w4Y1_dgLMvFVVsmTkMeYO4FZfhQ%40mail.gmail.com.

[Tom Gally]

Dan wrote:

I find that these models write well, but with a distinct style that is usually easily identified.

That distinct style is their default style, as most people don't prompt the models for anything else. After ChatGPT was first released, and before I realized that it could be prompted to write differently, I thought it must have been trained mainly on high-school composition textbooks, as most of what it produced sounded like a model five-paragraph essay. GPT-4’s default style is somewhat better, though it will still sometimes produce series of paragraphs beginning “First, ... Second, ... Third, ... , In conclusion, ...,” just like in those high-school textbooks. Almost nobody writes that way in real life.

Just for fun, I had GPT-4 write summaries of a text in the following ten styles:

(1) A straightforward summary in standard English
(2) A sarcastic summary in Indian English
(3) A slangy genz text message
(4) A excessively formal and pretentious version as if written by a snobbish university professor
(5) An eight-year-old’s note to a friend
(6) A mistake-filled version as if written by a native speaker of Japanese with only limited English skills
(7) A mistake-filled version as if written by a native speaker of Spanish with only limited English skills
(8) A text message written only in emoji
(9) Elizabethan English
(10) A sloppily written series of tweets by a native English speaker

The full prompt and the (decidedly mixed) results are here:

https://www.gally.net/temp/20231107gptassistanttest/test5.html

For me, the biggest giveaway of LLM-produced text on the web is the absence of mistakes. As shown by (6), (7), and (10) above, GPT-4 cannot convincingly imitate human error or sloppiness. (2), (3), and (9) would probably seem phony to natives of those dialects, too.

I was impressed by (8), though.

Tom Gally

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/3da2e5b7-299b-4cea-a5a2-b01ab589b525%40app.fastmail.com.

[Oroszlany Balazs]

" The scientific journal Nature has published a series of articles on the potentially adverse effects of using AI to generate research papers submitted for publication."

This is a completely different issue.

Non-native English speakers* have a double disadvantage if they want to publish in English. The first is obvious: they should be able to write in English.
They should also be able to convince reviewers that they can write in English. Note the asterisk on non-native English speakers - it is entirely anecdotal, but during my graduate years we more often received comments on the grammar of a paper if the first author had a non-native English sounding name - even if all papers went through the same internal (and often external, e.g. the aforementioned [see? "aforementioned" is completely DeepL's fault... :)] native English reviewer) processes, or if the first author was actually a native speaker with a "foreign" name.

Recent advances on the language front - and I am not talking about ChatGPT - have helped a lot in this regard - as I type this reply in gmail, any misspelled words are automatically corrected. And I could have installed Grammarly, or run this through DeepL's write. These tools are spell checkers on steroids, and can improve both the text produced and the writer's English.

Relying too much on ChatGPT, and not tweaking the results, could result in poor quality papers - but then again, we already have paper mills to produce those, and the journals that accept them...

(I also think that the average English proficiency of the younger generation of scientists, not only in Japan, is improving - but that is another issue...)

Balazs

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/3da2e5b7-299b-4cea-a5a2-b01ab589b525%40app.fastmail.com.

[John Stroman]

Although I generally agree with Balazs' comments, I do not think it is a completely different issue.

First, we should distinguish reviewers from editors. Unless things have changed, reviewers are blinded to the authors' names and institutions although they may be able to guess from the content of the submission and the references.  When I worked in a research lab in the early 1970s, our lead author had a Czech surname, and none of his submissions were rejected, but the institution was also very prestigious. Reviewers make comments on the content of a submission and submit questions for clarification to the authors. The difficulties that Balaz refers to were most likely decisions made by editors, who unfortunately have prejudices just like the rest of us.

Researchers find themselves under intense pressure to publish because their positions and funding correlate with their output. And there are plenty of disreputable paper mills and journals that are willing to publish almost anything for a fee.

That being said, I sometimes work with an agency that prepares Japanese medical research articles for publication. The agency describes the jobs as "editing" although I seldom receive a file in English. My job is not only to translate the content, but also rewrite it to target an English speaking readership based on each journal's guidelines for authors. I actually enjoy the rewriting more than the translating.

AI and LLMs can serve as useful tools for researchers who need to submit papers, but at this point I think it is unclear whether CHAT and other AI engines can rearrange Japanese sufficiently into English without intervention by a native English speaker since the traditional ways of presenting a convincing argument differ so greatly between the two cultures. As Balaz has pointed out, I agree that the English proficiency of younger Japanese scientists is improving, but most likely this is  thanks to the work of people like Tom Galley who teach seminars to budding scientists on how to write good scientific English.

John Stroman 

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

On Fri, Nov 10, 2023 at 7:20 AM Oroszlany Balazs <rori...@gmail.com> wrote:

" The scientific journal Nature has published a series of articles on the potentially adverse effects of using AI to generate research papers submitted for publication."

This is a completely different issue.

Non-native English speakers* have a double disadvantage if they want to publish in English. The first is obvious: they should be able to write in English.
They should also be able to convince reviewers that they can write in English. Note the asterisk on non-native English speakers - it is entirely anecdotal, but during my graduate years we more often received comments on the grammar of a paper if the first author had a non-native English sounding name - even if all papers went through the same internal (and often external, e.g. the aforementioned [see? "aforementioned" is completely DeepL's fault... :)] native English reviewer) processes, or if the first author was actually a native speaker with a "foreign" name.

Recent advances on the language front - and I am not talking about ChatGPT - have helped a lot in this regard - as I type this reply in gmail, any misspelled words are automatically corrected. And I could have installed Grammarly, or run this through DeepL's write. These tools are spell checkers on steroids, and can improve both the text produced and the writer's English.

Relying too much on ChatGPT, and not tweaking the results, could result in poor quality papers - but then again, we already have paper mills to produce those, and the journals that accept them...

(I also think that the average English proficiency of the younger generation of scientists, not only in Japan, is improving - but that is another issue...)

Balazs

--

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/CANqOQsBDi4eBSshpaW0UQd0cQ9SFAGd2QpCnzTyPEHTpjDQ28A%40mail.gmail.com.

[Tom Gally]

A couple of tangential comments.

As essay appeared in Nikkei yesterday by a former commissioner of the Japan Patent Office in which she mentions using generative AI to rewrite her English: https://www.gally.net/temp/20231107gptassistanttest/nikkeiessay.html

After I switched from freelance translation to academic work twenty years ago, I gradually became aware that attitudes toward native English ability in academia are more complex than I had realized. While some academics believe that, as a matter of course, native English speakers are essential to support research and education in international contexts, others strongly disagree.

The disagreements come from a variety of positions, including the political (objection to perceived hegemony of certain English-speaking countries or undue privileges that seem to be given to native speakers), the practical (many researchers are able to perform at a high level in international contexts despite being nonnative), and the personal (resentment at being passed over for jobs and other opportunities because of nonnative status).

I tried to maintain a neutral position in those debates, though my status as a white male American native speaker of English with a tenured position at a high-ranked university sometimes complicated matters. If anyone is interested in my thoughts on these issues, I have written them up in the following book:

https://books.kenkyusha.co.jp/book/978-4-327-49026-3.html

Tom Gally

[Jon Johanning]

Hello all,

Some of what I'm doing these days is editing medical Japanese docs, mostly case reports, translated J>E by folks who are apparently Japanese speakers who know medicine pretty well but with a much more imperfect command of English. This is evident from the fact that they make all the common mistakes that native Japanese speakers with poor English make. 

The agency I'm getting these jobs from is therefore hiring non-native-English-speaking translators to do J>E, contrary to what I used to think was the pretty firm rule in the translation game. Perhaps they are doing that rather than giving the jobs to me because that's cheaper for them; I don't know what other reasons they might have. I am also getting some small J>E jobs from them which have rather complex formatting, perhaps because the LLMs can't handle that formatting -- things with lots of tables or other quirks. But I might be wrong about that.

Happy Thanksgiving to observers of that festival!

Jon Johanning

zenn...@icloud.com

[Jon Johanning]

This is a subject that is directly relevant to my work now. Much of it is revising translations of Japanese medical case reports, etc., that were obviously done by native Japanese speakers with less-than-perfect English writing ability. I say this because they are full of the typical mistakes that these writers do in writing English: punctuation, word order in sentences, unusual vocabulary that native English speakers would not use, etc.

I am getting this work from an agency who is giving much less translation work than they used to (apparently because the Japanese speaker translators charge lower rates than mine). It is an American agency; where the Japanese speaker translators are located I don't know, but they do seem to have good medical knowledge (better than their English, for sure). (Of course, they also make generic translators' mistakes that need fixing, such as omitting words or clauses by mistake. Many of them seem to be using raw DeepL without correcting its mistakes.) The end client for these translations is an American medical company, so I try hard to make the English sound right. The fact that it is an American agency bypasses, I think, the aspect of how well Japanese companies understand the whole process of J>E technical translation. 

I throw this information out as one person's experience; I don't know whether it proves much of an answer to Michael's original question: "how much of remunerated J-E translation is still carried out by native Japanese."

I still hold firmly to the old principle that the best translation usually is done by native speakers of the target language, since that is the language that readers of the translation will be reading, and clumsy mistakes in that language are very off-putting.

Jon Johanning

zenn...@icloud.com

[Bill Lise]

" The agency I'm getting these jobs from is therefore hiring non-native-English-speaking translators to do J>E, contrary to what I used to think was the pretty firm rule in the translation game."

That is certainly still common sense and a "firm rule" among language professionals native to the target language, but I am confident that it is "firmly ignored" in the majority of cases, particularly when you consider all the JA-EN translation done in Japan.

Lower rates and greater availability of translators (actual or 自称) are at work, plus the lingering notion that people without Japanese DNA couldn't possibly understand the source text. The only thing that MT brings is that it has neither language as native. That said, the English produced by some MT systems is much better than that produced by many NJS translators.

[TCM Lounge Sendai]

This is a very interesting topic although I haven't gone through all the messages.  The following phrase was very funny, "native Japanese speakers with less-than-perfect English writing ability." 

I hope I will not be such a person in the future after studying medical writing at UC San Diego. Thanks for sharing.

Best regards,

Masatoshi Shoji 

2023年11月17日(金) 7:56 Jon Johanning <joha...@johanning-translations.com>:

--
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/93aef50e-85f0-46f1-b8ef-da0e05e0717cn%40googlegroups.com.

[Jon Johanning]

@Masatoshi

I apologize for my less-than-perfect statement that you quoted. Of course, there are plenty of native English speakers with less-than-perfect English writing abilities, also. What I should have said was that, like all languages, English is one that non-natives have a lot of trouble picking up, and it would be good for translators into English, however they picked it up, to be able to write English in a correct and idiomatic way, which is .

All of this reminds me of a magazine article I read when I first went to Japan long ago. The writer claimed to be an expert on the English language, and insisted that Japanese was much better at expressing subtle emotional overtones. As I remember, he had a theory that the DNA of Japanese persons' brains made them much more sensitive to these subtleties, as well as to bird songs and other beauties of nature. But what he betrayed, of course, was how limited his "expert" comprehension of English was. I don't think he was much of an expert on human neurology.

Jon Johanning

[TCM Lounge Sendai]

Hi Jon,

Thanks for sharing your episode with a self-proclaimed expert.

A true gem is always hidden, I believe. 

It's been a hard path to be a translator from Japanese to English as a non-English native speaker.

BTW, I met Professor Jody Noguchi at the 2023 Japan Association for Language Teachers conference.

She was very impressive as usual.

Regards,

Masatoshi Shoi

2023年11月26日(日) 2:27 Jon Johanning <joha...@johanning-translations.com>:

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/d8034147-4168-49d8-97c0-7f3f874c3797n%40googlegroups.com.