AI-generated report on state of translation market

82 views
Skip to first unread message

Tom Gally

unread,
Dec 11, 2024, 9:45:32 PM12/11/24
to hon...@googlegroups.com, swe...@groups.io
A few hours ago, Google announced a bevy of new AI products:


One is a Gemini Advanced feature called Deep Research:


I tried it with this prompt:

I want to get an overview on how advances in large language models and AI are affecting the work and careers of professional translators. Please prepare a comprehensive report on trends within the last year or so. I am especially interested in the effect on the translation market and translators’ careers in Japan for translators working between Japanese and English. In your research survey, access information on this topic in both English and Japanese.

After about five minutes, it produced the following report:


There’s nothing particularly new for us in this report, but I can see how this tool might be useful for compiling background information about subjects one is unfamiliar with.

The citations in the above report are all in English, but during its research process Gemini showed a much longer list of links that it accessed. About half were in Japanese.

Tom Gally

cpta...@ozemail.com.au

unread,
Dec 11, 2024, 9:49:15 PM12/11/24
to hon...@googlegroups.com, swe...@groups.io

“ Japan is a highly diverse country, with a variety of regional dialects and cultural norms.”

 

Whut?

--
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 visit https://groups.google.com/d/msgid/honyaku/CACnS3NdwpYq9kxbSz0Tet6MbZKxUCecKwD2zA2RMkCAPhbMpfQ%40mail.gmail.com.

Bill Lise

unread,
Dec 11, 2024, 10:23:24 PM12/11/24
to hon...@googlegroups.com
The report appears to have largely learned from promoters of AI translation. It several times mentions "new opportunities" for translators, referring to post-editing, and it does mention downward pressure on prices. They are probably talking about prices for _translation_. And it appears they miss what is actually happening, which is that translators are being asked to do post-editing for rates that are 1/3 to 1/5 of what even translators working for agencies could make before MT arrived.
Maybe AI can be asked more directly whether it would recommend that a young person learn Japanese in order to become a translator. I know my answer would be a resounding no.
Freelance translating as a lucrative career is quickly coming to an end for most translators, a reality missed by a lot of people, including those running JAT, who seem to be holding an increased number of diversionary socializing events lately, but nothing much aimed at the real problems faced by working translators. The most serious and career-destroying problem is not the problem of how to adopt or adapt to AI, it's more a problem of how to find clients when you've been accustomed to just sitting in front of your computer and waiting for a "project" to arrive so you can "collaborate." Those days are quickly coming to an end. I wonder what AI would say about that.
*********************************************
リゼ ウィリアム
Bill Lise
*********************************************


--

Tom Gally

unread,
Dec 12, 2024, 2:17:29 AM12/12/24
to hon...@googlegroups.com, swe...@groups.io
I continued the Gemini thread with the following prompt:

I shared your report with a group of professional translators.

One person responded “Whuut?” to your statement that “Japan is a highly diverse country, with a variety of regional dialects and cultural norms.”

Another person responded as follows [NB: This comment was sent only to the Honyaku ML]: “The report appears to have largely learned from promoters of AI translation. It several times mentions "new opportunities" for translators, referring to post-editing, and it does mention downward pressure on prices. They are probably talking about prices for _translation_. And it appears they miss what is actually happening, which is that translators are being asked to do post-editing for rates that are 1/3 to 1/5 of what even translators working for agencies could make before MT arrived. Maybe AI can be asked more directly whether it would recommend that a young person learn Japanese in order to become a translator. I know my answer would be a resounding no. Freelance translating as a lucrative career is quickly coming to an end for most translators, a reality missed by a lot of people, including those running JAT, who seem to be holding an increased number of diversionary socializing events lately, but nothing much aimed at the real problems faced by working translators. The most serious and career-destroying problem is not the problem of how to adopt or adapt to AI, it's more a problem of how to find clients when you've been accustomed to just sitting in front of your computer and waiting for a "project" to arrive so you can "collaborate." Those days are quickly coming to an end. I wonder what AI would say about that.”

 What are your responses to these two comments?

It responded as follows:

Thank you for sharing these valuable comments with me. It's important to address the concerns of professional translators and incorporate their perspectives into my understanding of the field.

Response to "Whuut?"

The comment "Whuut?" regarding the statement "Japan is a highly diverse country, with a variety of regional dialects and cultural norms" indicates a misunderstanding of the linguistic and cultural landscape of Japan. While Japan is often perceived as a homogeneous society, it actually has a rich diversity of regional dialects and cultural practices. These variations can be significant, affecting not only pronunciation and vocabulary but also communication styles and social norms. For example, the Kansai dialect, spoken in the Osaka region, has distinct grammatical structures and vocabulary compared to standard Japanese. Similarly, cultural norms regarding politeness, formality, and social etiquette can vary across different regions and social groups.

It's important for translators, especially those working between Japanese and English, to be aware of these regional and cultural nuances to ensure accurate and culturally appropriate translations. Failing to consider these variations can lead to misunderstandings or misinterpretations, potentially impacting the effectiveness of the translation.

Response to the Second Comment

The second comment raises several crucial points about the impact of AI on the translation market and the challenges faced by professional translators in Japan. I acknowledge that my previous report may have presented an overly optimistic view of the "new opportunities" created by AI, without fully addressing the concerns about declining translation rates and increased competition.

The commenter rightly points out that while AI is creating new roles for translators, such as post-editing, these roles often come with significantly lower pay compared to traditional translation work. This is a valid concern, and it reflects the downward pressure on prices that AI is exerting on the translation market. As MT engines become more prevalent, clients may be less willing to pay premium rates for human translation, even for high-value content.

Furthermore, the commenter highlights the changing dynamics of the translation market, where translators can no longer rely on simply "waiting for a project to arrive." The increasing use of MT and the rise of online platforms are changing the way translators find clients and compete for projects. Translators need to be more proactive in marketing their services, building their online presence, and specializing in niche areas to remain competitive.

The commenter also raises a critical question about the future of the translation profession and whether it's still a viable career path for young people. This is a complex question with no easy answers. While AI is undoubtedly changing the landscape of the translation industry, human expertise and skills are still essential for many types of translation work. Translators who can adapt to the changing market, develop new skills, and specialize in areas where human expertise is still valued will likely continue to find rewarding careers.

I appreciate the commenter's insights and suggestions. I will strive to incorporate these perspectives into my future reports and provide a more balanced and nuanced view of the impact of AI on the translation industry in Japan.

Addressing the Specific Points Raised

  • Post-editing rates: I acknowledge that post-editing rates are often significantly lower than traditional translation rates. This is a concerning trend that needs to be further investigated and addressed.
  • Finding clients: I agree that translators need to be more proactive in finding clients and marketing their services in the age of AI. This includes building an online presence, networking with potential clients, and specializing in niche areas.
  • JAT's role: I will research the Japan Association of Translators (JAT) and their activities to better understand their response to the challenges faced by working translators.
  • AI's recommendation for young people: I will attempt to gather information and insights on whether AI would recommend translation as a career path for young people, considering the current trends and challenges in the industry.

I am committed to providing accurate and insightful information about the impact of AI on the translation industry. I will continue to learn and refine my understanding of this complex and evolving field.

As far as I know, the promise to “incorporate these perspectives into my future reports” is hallucinated, unless Google has developed a way for LLMs to update their knowledge continuously.

The Deep Research function released yesterday uses Gemini 1.5 Pro, which was released back in February, many dog years ago in AI time. I put the same prompt to the just-released Gemini 2.0 Flash Experimental, changing only “for translators working between Japanese and English” to “for freelance translators working between Japanese and English.” To save bandwidth, I put the new full prompt and the response here:


Opinions will vary, but this report looks better to me than the previous one.

Note that Gemini 2.0 Flash Experimental, like some other recent LLMs, can do web searches to support its responses. ChatGPT can do that, too, though Claude cannot. I don’t know yet how web-search capability affects the quality of LLM translations. In principle it should make them better, as it becomes possible to confirm the readings and translations of proper names (which LLMs can be bad at for J to E) as well as to find translations for technical terms. But it might take some sophisticated prompt engineering to get those web searches working well.

Tom Gally

Bill Lise

unread,
Dec 12, 2024, 2:57:56 AM12/12/24
to hon...@googlegroups.com

Wow, what a total failure it made on the dialect and culture diversity problem. It doesn’t realize why someone said whut snd tried to defend a comment but made that is irrelevant to commercial translation.
It’s going to check your see what JAT is doing? If it learns from the JAT website, it will get a picture that is completely different from the real world. Most of the mission statement items of JAT are not being fulfilled or are off-target in these times.
I think it’s still better to ask a human about these things. 
*********************************************
リゼ ウィリアム
Bill Lise
*********************************************

Tom Gally

unread,
Dec 12, 2024, 3:57:23 AM12/12/24
to hon...@googlegroups.com
It doesn’t realize why someone said whut snd tried to defend a comment but made that is irrelevant to commercial translation.

I agree that that diversity statement by Gemini doesn’t apply to most translators; I suspect that it emerged from its U.S.-centric alignment. Google AI models have often been excessively cautious, sometimes to comical degrees (another AI generated report: https://www.perplexity.ai/page/google-ai-s-inappropriate-imag-cI51_xCRQLu8lRvAuekOLQ).

But, to give it a break, my prompt didn’t specify the type of translation. An interpreter working in some parts of Kyushu, Tohoku, and Okinawa would need to be aware of that cultural and linguistic diversity, as would many literary translators. I’m now reading a novel written in the Taisho period and set in Osaka. While I’m having no trouble following the story, if you asked me to translate it into English I would struggle with a lot of the dialogue, which is rendered in working-class Osaka-ben of that period.

I think it’s still better to ask a human about these things. 

Obviously humans are better for many things, but what one human would be able to produce such a report in ten seconds, as Gemini did? And how well would that one human be able to present various viewpoints?

Tom Gally

Bill Lise

unread,
Dec 12, 2024, 5:23:02 AM12/12/24
to hon...@googlegroups.com
I certainly wouldn’t have come up with something as detailed as what the AI did, and it certainly wouldn’t be as fast. There was a lot of boilerplate descriptions of things that anybody familiar with the business knows. I guess it did not assume knowledge that is, well, common knowledge.

*********************************************
リゼ ウィリアム
Bill Lise
*********************************************
Reply all
Reply to author
Forward
0 new messages