ATA blogger's perspective on future of translation

[John Stroman]

Fellow 'yakkers,

My wife came across a rather longish but well-written article on the future of translation that was posted on the ATA blog "The Savvy Newcomer."

https://www.atanet.org/business-strategies/is-there-a-future-in-freelance-translation/

Just to clarify, I am just providing information that may be useful, and I am not advocating any opinion held by the author. 

John Stroman

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

[Jon Johanning]

I would like to thank John for posting this article from the ATA blog “The Savvy Newcomer.” I think he was wise not to advocate any opinion contained in it, because I have some criticisms of it.

 

The author ends with this very positive claim: “If you’re reading this article, things have already started to change!” She doesn’t tell us what her optimism is based on, but I have to say that, in my opinion, since the article was published in 2021, things have changed, all right, but for the worse.

 

She points out a number of problems that freelancers have to deal with, and I think she is right about them. But the situation has deteriorated greatly, especially in a couple of respects.

 

One is the trend in technological development. When the article was written two years ago, “chatbots” or large language models were just starting their very rapid rise to their present widespread employment. They are now joining conventional computer-aided translation programs and machine translation software, making it even easier for people with very little translation skills, or none at all, to pump out simulated translations, which have a surface appearance of “hand-made” work by translators with real skills. 

 

Given the fact, which Ms. Maginot notes, that monolingual translation customers – the great majority in the U.S., if not the rest of the world – have no way of judging the quality of the “translations” that are churned out this way, and the obvious fact that this stuff can be pumped out by any halfway-competent computer user for free and instantly, it’s pretty clear that we human translators (freelance, at least) are probably doomed.

 

The answer to this fatal trend that she offers us is large-scale client education: “possibly the cornerstone of change; the one thing that may change consumer perception and revalue the profession in the eyes of the general public.” And she quotes ATA Treasurer John Milan: “Together we can educate the public and ensure that our consumers value us more like diamonds and less like water.”

 

The ATA and other associations are working mightily to do this education, but as any educator knows, offering education on a subject to someone who doesn’t understand what the subject is all about in the first place, doesn’t care about getting the education into their brains, and isn’t willing to spend the time and money that would be required from a properly educated person, is essentially useless. How many people looking for a way of getting their precious communications transferred to languages they know nothing about are actually being educated in the ways they should be? Not many, I’m afraid. As the old Zen expression puts it, you’re trying to sell water by the river.

 

Her comparison with the vanilla industry doesn’t help her argument, as far as I can see. “Nutrition education and consumer demand for all-natural foods started turning [the vanilla industry] around,” perhaps, but translation education seems to be having very little impact, as I have said, and the synthetic product is tasting more and more like “all-natural” translation (translation by us Homo sapiens folks).

 

Ms. Maginot’s cheerful closing sentence may be her way of following the advice writing teachers always give: don’t end your essay with a downer that leaves the reader depressed. But I would rather keep my attention on the reality: in fact, it’s just harder and harder to make a living this way, and may soon be impossible.

Jon Johanning

zenn...@icloud.com

[Matthew Schlecht]

On Sun, Jun 25, 2023 at 2:48 PM Jon Johanning <joha...@johanning-translations.com> wrote:

 

One is the trend in technological development. When the article was written two years ago, “chatbots” or large language models were just starting their very rapid rise to their present widespread employment. They are now joining conventional computer-aided translation programs and machine translation software, making it even easier for people with very little translation skills, or none at all, to pump out simulated translations, which have a surface appearance of “hand-made” work by translators with real skills.

I haven't read the blog posting, but will add it to my reading list.

I just wanted to respond to the point mentioned above.

Inexpensive, poorly-done translations have always been available, in those past cases done by inexperienced and/or inept freelancers. To some extent, those have always subtracted from the number of projects available to established competent freelance translators. What is different now?

Possibly because the inferior product was easier for the end clients to identify in the past since it was obviously bad, but now the presentation and appearance have improved. The crap has been dressed up to look like crêpes, and thus a more thorough taste testing is required.

Also, end clients get caught up in the hype. A few weeks ago, I received in the late afternoon a project offering of a dissertation to translate. ~15,000 words in a scratchy dead PDF, and the client wanted it back in two days. I guess this sort of thing isn't really new, but I sense there is an expectation is that it is indeed routinely doable, since AI/MT can do anything.

As others have pointed out, there is (sadly) a huge market in which mediocre translations are good enough, especially pretty ones, but it will take a case or two in which mediocre translation product causes a huge financial liability for expectations to be readjusted.

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

[Dale Ponte]

Jon Johanning wrote:

 “hand-made” work by translators with real skills. 

"Artisanal" sounds about right too.    

Dale Ponte

[Avantika Srivastava]

Hello Fellow 'yakkers, hope everyone is in good spirits and health. 

I was recently approached by an agency to provide rates for translating medical records from Japanese to English. They use Japanese source text for estimating the total word count of the project, and the Japanese word count is done using tokenized word count system with fuzzy matches. They want me to provide my rates for that. 

I have never calculated rates for tokenized word count. I would greatly appreciate it if someone could guide me on how to do it. 

With so much experience and knowledge here, my mind straightway thought of this group for asking this question. 

Thanks in advance,

Avantika Srivastava Ganjoo

Silent 'Yakker since I can remember... 

[Tom Gally]

I'm not sure what a “tokenized word count system with fuzzy matches” is. If they are referring to the tokens into which text is converted for processing by LLMs, then you need to ask them what particular toxenization system they are using, as the tokenization system varies depending on the language model.

You can get token and character counts for the GPT-3 tokenizer here:

https://platform.openai.com/tokenizer

Here are two sample results:

Text:

富士通は29日夜、子会社が運営するマイナンバーカードを利用した証明書交付サービスで新たに誤発行が発生したと発表した。同社は5月下旬以降に全国123自治体のシステムを一時停止して点検していたが、今回の不具合は修正できていなかった。同社はシステムを利用する全自治体を対象に、システムを再び止めて点検するという。

　誤発行は福岡県宗像市で28日に起きた。市民の女性が市役所の窓口で住所変更の手続きをした後、マイナンバーカードを使って証明書交付サービスで住民票の写しを発行しようとしたところ、他人のものが発行されたという。

tokens: 390

characters: 258

tokens/characters = 1.51

Text:

最近よく耳にするアクティビストは、もの言う株主を指す。投資先の企業の株価が上がるように積極的に働きかける。その戦術などを表現する用語には、戦争に由来する言葉が多いそうだ。企業買収に詳しい太田洋（よう）弁護士の近著『敵対的買収とアクティビスト』で知った▼たとえば、ウルフパック（群狼〈ぐんろう〉）は、複数のアクティビストなどが別々に株式を買い進めて一気に要求を通す行為を意味する。もとは第２次世界大戦中、ナチス・ドイツの海軍提督が考案した戦術だ。Ｕボートを多数配置し、英国の通商ルートを破壊した▼調べるとほかにもある。取引開始直後に大量に買う「暁の急襲」。市場から抜けるために利益を放棄して投げ売りする「降伏」。投資と無縁の身にはただ恐ろしげに響く軍事用語が、特別な意味を持つ世界がある▼今年も株主総会の季節が来た。ピークのきのうは６００社近い企業が開いた。株主提案が過去最多になったのは、経営のあり方についてモノを言いたい株主が増えたからだ。目先の利益還元の要求から中長期的な提言まで、内容も多岐にわたるという▼かつて株主総会には、金品目当てで介入する総会屋がつきものだった。同一日に集中したのは出席を避けるためだったが、法改正や規制強化などで激減した▼太田氏は、アクティビズムやその対応で何が「良い」かの正解はないと書いている。もの言わぬ株主でも株主でなくても、会社はだれのものかを改めて考えたい。できれば平和的な言葉で。

tokens: 963
characters: 612

tokens/characters = 1.57

The token/character ratio might be quite different for text that contains a different kanji/kana ratio or a lot of numbers or romaji.

For reference, here are OpenAI’s per-token prices:

https://openai.com/pricing

The most expensive is $0.06 input + $0.12 output for per 1,000 tokens using GPT-4 and 32,000-token context.

Let’s see how much that second text above would cost to translate through OpenAI’s API using GPT-4 and the largest context. I use the token counts from the GPT-3 tokenizer.

Input text (including prompt):

Translate the following text into English: 最近よく耳にするアクティビストは、もの言う株主を指す。投資先の企業の株価が上がるように積極的に働きかける。その戦術などを表現する用語には、戦争に由来する言葉が多いそうだ。企業買収に詳しい太田洋（よう）弁護士の近著『敵対的買収とアクティビスト』で知った▼たとえば、ウルフパック（群狼〈ぐんろう〉）は、複数のアクティビストなどが別々に株式を買い進めて一気に要求を通す行為を意味する。もとは第２次世界大戦中、ナチス・ドイツの海軍提督が考案した戦術だ。Ｕボートを多数配置し、英国の通商ルートを破壊した▼調べるとほかにもある。取引開始直後に大量に買う「暁の急襲」。市場から抜けるために利益を放棄して投げ売りする「降伏」。投資と無縁の身にはただ恐ろしげに響く軍事用語が、特別な意味を持つ世界がある▼今年も株主総会の季節が来た。ピークのきのうは６００社近い企業が開いた。株主提案が過去最多になったのは、経営のあり方についてモノを言いたい株主が増えたからだ。目先の利益還元の要求から中長期的な提言まで、内容も多岐にわたるという▼かつて株主総会には、金品目当てで介入する総会屋がつきものだった。同一日に集中したのは出席を避けるためだったが、法改正や規制強化などで激減した▼太田氏は、アクティビズムやその対応で何が「良い」かの正解はないと書いている。もの言わぬ株主でも株主でなくても、会社はだれのものかを改めて考えたい。できれば平和的な言葉で。

tokens: 971

characters: 655

Output text:

The term "activist", which we often hear lately, refers to a vocal shareholder. They proactively strive to raise the share price of the companies they invest in. Apparently, many of the terms expressing their tactics originate from warfare. I learned this from the recent book "Hostile Takeovers and Activists" by Yo Ota, a lawyer well-versed in corporate acquisitions.

For example, a "wolf pack" refers to a practice where multiple activists buy shares separately and then collectively assert their demands. This strategy originated during World War II, conceived by an admiral of Nazi Germany's navy. He deployed a large number of submarines, which disrupted Britain's trade routes.

On researching, I found other terms too. A "dawn raid" is buying a large quantity of shares right after trading begins. A "surrender" is selling off shares by forgoing profits in order to exit the market. For those unrelated to investment, these seemingly intimidating military terms exist in a world with special meanings.

It's now the season for annual shareholders' meetings. Yesterday, which was the peak of the season, nearly 600 companies held their meetings. The reason for the record number of shareholder proposals is an increase in shareholders who want to voice their opinions on corporate management. The proposals range widely in content, from immediate profit returns to long-term suggestions.

In the past, there used to be 'sokaiya' or corporate racketeers who would intervene in shareholders' meetings for monetary gain. Concentrating the meetings on the same day was a way to avoid their attendance, but legislative amendments and strengthened regulations have significantly reduced their presence.

Mr. Ota writes that there's no correct answer to what is "good" regarding activism and how to deal with it. Whether a shareholder is silent or not a shareholder at all, we should reconsider to whom the company belongs. Ideally, we should do this with peaceful words.

tokens: 396

characters: 1972

The cost of this GPT-4 translation would be, if my calculation is correct, about $0.11.

Obviously no human translator can compete on price or speed. We will need to compete on the quality of our translations and the value of other services that we provide.

Warren Smith posted a forceful statement in that regard here on June 8.

Tom Gally

Yokohama, Japan

--
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/CAMuw29GHTruaHHSWDeKD_GkyZTz1Qb4As3K-Uxq7mRfVqONdkg%40mail.gmail.com.

[Avantika Srivastava]

Thanks for the detailed reply, Tom. I am still going through it trying to understand this thing about tokens. 

The agency told me that they use SDL Trados for translation and the tokenized word count is a feature it comes with. 

What I understand from your explanation and previous threads is that machine translation is not really worth it. I have not done machine translation before, though I see that agencies are paying lesser and lesser for academic and other human translations also. I have been doing translations part-time, but the amount of effort that goes in is not being compensated enough anymore. I thought it was only me but I realise that it has become the trend with the new technologies coming up. 

Thank you for your time and the valuable information. 

Avantika 

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CACnS3NdzZsedYdC5%2BUKaZncrJM6D3%3D6td2n%3DTCUS-19sPrfs4g%40mail.gmail.com.

[Kevin Johnson]

Hi Avantika,

Tokens in this case don't have anything to do with machine translation. With respect, I would ignore Tom's reply because it has nothing to do with your question. Check out Studio's use of tokens here: 
https://gateway.sdl.com/apex/communityknowledge?articleName=000012678
https://gateway.sdl.com/apex/communityknowledge?articleName=000014534

Let's set aside the word "tokenized" for a minute, as this is just jargon and not especially important.

Fuzzy matches don't have anything to do with MT (machine translation) and are instead related to TMs (translation memories). A fuzzy match is the extent to which a segment (usually a sentence) imperfectly matches a pre-existing segment in the translation memory, or in other words, the extent to which the new sentence matches an old sentence that has already been translated.

These old translations are usually translations done by humans and are being reapplied to new texts. The translator (you) is expected to adapt the old translations to the new sentence. 

A "low fuzzy" (medium-percentage match like 70%) would typically get about 50% your full rate, while a "high fuzzy" (very high-percentage match, like 90%+) gets more like 30%. Anything below a 70% fuzzy match would generally get full rates.

This has been pretty standard since CAT (computer-assisted translation) became standard. Anyone who is not familiar with these concepts probably hasn't done professional translation in a long time -- which is fine! But this has been really normal stuff for the last decade or longer.

Returning to the concept of tokens, it sounds like what they're trying to do is use Studio's calculation of "Japanese words" (tokens) to determine a rate. Studio attempts to recognize "words" within spaceless Asian languages and convert them into "tokens" so that instead of charging by character, one can charge by "word" (token). This is a bit of a pain and I would try to negotiate a character-based rate using fuzzy thresholds instead (refer back to my concept of high and low fuzzies). Personally, I don't have a token rate and I always manage to get a character-based rate instead.

Again, this has nothing to do with machine translation at all. And although the industry has changed dramatically in the last 4 years or so, with old-fashioned companies (and translators) getting swallowed up or downsizing, MT just ain't good enough. There's definitely still a place for professional translators with professional skill sets and professional salaries. We're the ones working every day rather than doomposting about it :)

P.S. The yen is so cheap right now that I doubt overseas translators can compete. Those of us in Japan feel like we have the pick of the litter.

Kevin Johnson

Osaka

[Dan Lucas]

Kevin, the observation in your PS may have been partly tongue-in-cheek but I feel bound to point out that this overseas freelancer is still competing quite effectively. I completed a record number of projects in May and both June and July have been more than usually busy, in fact.

Yes, the yen is weak, but this too will pass and in the meantime I make a comfortable living. The pendulum will swing back, eventually - I play the long game. On a positive note, the recent depreciation of the yen has made the forthcoming month-long holiday in Japan for my wife and children rather cheaper than it would otherwise have been...

As for the rest of your comments, I agree. I too am busy working rather than catastrophising about the industry. Indeed,  for the past three weeks I have been cursing myself for taking on too much...

Paul Filkin of SDL/RWS wrote a useful article about how Trados approaches word counts and why those counts differ from those of, say, Word. It is old but the concepts are still valid and so it may be useful to Avantika.

https://multifarious.filkin.com/2012/11/13/wordcount/

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/CAMk1sch0D2F2mu05OiRGNU%3Do3WBdsq9us-JM4q47p_GjUR7Pfw%40mail.gmail.com.

[Avantika Srivastava]

Thanks, Kevin and Dan. I will definitely go through the links you have sent. 

It is heartening to know that there is still work available for Japanese translators and interpreters. 

I have been registered with only a few agencies who have been sending me old-fashioned word/excel files for translation. 

Now I have entered a new phase of life with children off to university, and I can devote more time to my first love of Japanese language, planning to take cautious baby steps. 

Bests,

Avantika 

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/f29ffeeb-e322-4a6b-9a0a-2f6ab46e82f1%40app.fastmail.com.

[Tom Gally]

Kevin,

> Tokens in this case don't have anything to do with machine translation.

Thanks for the clarification!

Yes, Avantika, please ignore my earlier reply.

My ignorance about the meaning of “tokens” in the context of TMs was due to the fact that I left freelance translation in 2005 and never used the then-new TMs. I have now returned part-time to translation, but in an in-house position where the nature of the work means that I am not threatened by TMs, MT, or LLMs.

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/CAMk1sch0D2F2mu05OiRGNU%3Do3WBdsq9us-JM4q47p_GjUR7Pfw%40mail.gmail.com.