Sharp warning about ChatGPT

[Jon Johanning]

https://open.substack.com/pub/thealgorithmicbridge/p/chatgpt-and-the-future-present-were

It's very entertaining, but dangerous, as this reference makes clear. I don't think we should be giving it three cheers.

Jon Johanning

zenn...@icloud.com

[Tom Gally]

Jon,

Thanks for the link to that interesting article.

I don’t think, though, that many people have been giving three cheers to ChatGPT. Most of the reactions I’ve seen, including my own, have been some combination of “amazing” and “frightening.” Many of the other reactions have been “Because ChatGPT can’t do X, we don’t need to worry about it.” I personally think those latter reactions are misguided, and the article you link to seems to confirm that.

I continue to test ChatGTP’s translation ability. For my day job, I am now translating a speech from Japanese to English, and I first ran the source text through ChatGTP, Google Translate, and DeepL. ChatGPT was not remarkably superior to the other two, and at a couple of points it departed significantly from the source text—it seemed to want to write its own speech. That could probably have been prevented by fine-tuning the prompts.

I also have run some tests using literary texts, which are often cited as a genre where MT has no chance of succeeding. The results from Google and DeepL were really bad, but ChatGPT was almost sort of okay. I could see the output being usable for genre fiction after some light human postediting.

My in-progress notes for a video I plan to make about ChatGPT and translation are below, in case anyone is interested:

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

I mentioned previously that one reason I was impressed by ChatGPT’s translation ability is that it did not seem to have been trained specifically for translation. I found out today, though, that OpenAI has in fact been fine-tuning its translation ability:

https://github.com/openai/openai-cookbook#translation

https://arxiv.org/pdf/2110.05448.pdf

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/82eab9ef-a26c-4cb9-a4a7-a4276a4c88c5n%40googlegroups.com.

[Tom Gally]

I decided to try outsourcing the debate to ChatGPT itself. Here is the result:

My prompt:

Create a dialogue between two fictitious freelance translators, Hunter and Kaoru, about the potential for large language models (LLMs) like ChatGPT to reduce the demand for professional human translators. Hunter thinks that LLMs have the potential to replace human translators, while Kaoru believes that they are only passing hype and not a threat to professional translators’ business. Hunter and Kaoru translate technical and business-related texts between Japanese and English, so any examples you give should relate to those fields and to that language pair. Develop both viewpoints with compelling arguments and have the two translators take turns in their argumentative “fight.”

ChatGPT’s response:

Hunter: "Kaoru, I think we need to talk about the potential for large language models like ChatGPT to replace human translators like us. They're getting better and better at understanding and translating technical and business-related texts, and I think it's only a matter of time before they start taking over our jobs."

Kaoru: "Hunter, I understand your concerns, but I don't think we should be so quick to jump to conclusions. Sure, LLMs can translate text quickly and accurately, but they lack the cultural and context-specific knowledge that human translators bring to the table. It's true that they can't replace professional human translator."

Hunter: "But don't you see? That's exactly what the companies want. They want speed and accuracy, and they don't care about cultural nuances. LLMs can provide that, and they're getting better all the time. Just look at how well they're able to translate Japanese to English in technical and business-related texts."

Kaoru: "I understand that they can translate quickly and accurately in some cases, but they still can't replace the human touch. They can't understand the cultural context and intent of the text. For example, in business-related texts, LLMs can't understand the subtle nuances of negotiation and persuasion."

Hunter: "But that's the thing. Companies don't need that human touch when it comes to translating technical documents or contracts. They just want it done quickly and accurately, and LLMs can do that just fine. Plus, they're getting better at understanding context and cultural nuances all the time."

Kaoru: "I see your point, but I still believe that there will always be a need for human translators. We can't rely on LLMs for everything. They can't replace the human touch and expertise that professional translators bring to the table. Plus, in many cases, the cost of using an LLM is less than the cost of hiring a professional translator. And the demand for human translator will always be there."

Hunter: "I understand your perspective, Kaoru. I agree that human translators will always be necessary to a certain degree, but I also think that we need to be aware of the potential for LLMs to take over our jobs. We should be proactive in finding ways to adapt and stay competitive in the industry."

Kaoru: "I agree, Hunter. We should keep an eye on the developments in LLMs and find ways to adapt and stay competitive. But at the same time, we should remember that there is still a place for human translators in the industry."

[Tom Gally]

I just noticed that Kaoru made a point that weakened his/her own argument:

Kaoru: "I see your point, but I still believe that there will always be a need for human translators. We can't rely on LLMs for everything. They can't replace the human touch and expertise that professional translators bring to the table. Plus, in many cases, the cost of using an LLM is less than the cost of hiring a professional translator. And the demand for human translator will always be there."

Yes, ChatGPT is not perfect.

Tom Gally

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

Thanks Tom,

This illustrates that ChatGPT can keep a dialogue going without having a particularly deep insight into the subject matter. It seemed to rely a lot on this nebulous concept the “human touch” and didn’t mention risk management once.

 

(This is my response to Dan Lucas from last year, but then I report on my experience with ChatGPT and ingress to my business by MC)

 

“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.”

 

I would counter by calling this the category error. Why is “performing translation” defined so narrowly? When every translator with whom I have ever discussed these matters has revealed after some pushing that of course they too get second opinions, use the insights of other people, outsource varying kinds of task, just unofficially. The hardest part about being in business is realising it is happening.

 

“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…”

 

Not suggesting a panacea. That you think it is a “switch” and not an evolutionary step is another reason I suggest you are committing a category error.

 

“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.”

 

It is whatever industry the survivors say it is. How does “performing” exclude “managing”?

I mean I know so many people who would argue that it does, and they are the most vulnerable and unhappy “translators”.

 

Dan made several other points worth debating but I want to move so I will sum up simply that if you adamantly stick to your definition of your job as digging ditches but only with a shovel, don’t complain when along comes a backhoe.

Creating a category “freelancer” who only “performs” translation is the very sort of concept you need to challenge in order to survive. Better to re-evaluate your core skill set and redefine your job as “providing solutions to language barrier problems that people will pay for” (for example). That’s what I did. I didn’t take the “agency approach”. I did what I just said. I don’t accept schemes of categories produced by a world of ignorant people like “agency” vs “freelancer”.

I guess I agree with your assessment of where good freelancers go. And a 20 year old carpenter who is any good will own a building firm when he is 40. The others will be complaining and threatened by mechanisation.

 

Now, yes Perry it is hard to get on because all the buzz has led to their servers crashing. But I opened an account eventually. I haven’t tried translating anything but just asking a bunch of highly inappropriate questions, just for the lulz. And oh boy did I get a bunch of long-winded, hedging, non-commital, moralistic bullshit! In response to basically yes or no questions. Yes ChatGPT racism causes untold harm and women are not objects for sexual gratification and religion is complex blah blah blah. Couldn’t get it to shut up!

I think before we need to worry about “artificial intelligence” we should probably check what definition of “intelligence” they are employing. I mean it is certainly fast, grammar and spelling is good, but it is only as intelligent as a pompous first year student at some American Christian College.

 

Now more seriously, I just got off a conference call with a Japanese auto manufacturer and their US lawyers, and inter alia they mentioned a large amount of discovery material that they need translated and I am not QUITE so inappropriate as to start a sales pitch in the middle of an interpreting job but they will be getting this company

https://www.epiqglobal.com/en-au

 

To do “machine translation” with the Japanese editing by hand of anything the lawyers notice is odd. Japanese automanufacturers. Not Japanese translators. And only the ones they notice…

I will be putting in a bit of work getting this valued client back to where I can apply the human touch.

 

 

Chris Poole Translation

24 Greenwood St, Abbotsford VIC 3067

T     0412 287 487

W1 www.pooletranslation.com.au

W2 https://translator-interpreter.com.au/

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CACnS3NfSL2HepHhm-UXdx9eE11fvFs0pOUhCK_T-%3DybOzBMH3g%40mail.gmail.com.

[Dan Lucas]

On Fri, 13 Jan 2023, at 05:33, cpta...@ozemail.com.au wrote:

It is whatever industry the survivors say it is. How does “performing” exclude “managing”?

I mean I know so many people who would argue that it does, and they are the most vulnerable and unhappy “translators”.

Chris, I don't doubt your sincerity, but your definitions of these words simply do not reflect reality. I'm happy that you've apparently been successful in doing your thing, but what you describe as the way forward is the supervision, coordination, and allocation of resources and people (the shuffling around of 人、物、金), which is management.

A person in an office who oversees somebody in a workshop performing the work of, say, draining a car's engine oil and replacing the filter is a manager, not a mechanic, even though she works in the auto industry and may know a good deal about the work performed by mechanics.

A person who oversees somebody performing the work of converting text from one language to another is a manager, not a translator, even though he works in the linguistic services industry and may know a good deal about the work performed by translators.

This is neither linguistic nit-picking, nor a trivial matter of diction. These are materially dissimilar roles, requiring different skill sets and arguably different temperaments as well. To go back to the point in the previous email, they are different categories. Even though a small minority of people may succeed in straddling those categories, most will not because different people have different competencies. Surely this is not a controversial assertion?

My argument is that most people are simply not competent to tackle both translation and management and that, if they force themselves into a role to which they are not suited, they will not achieve fulfilment or happiness. Given the scale of the challenge involved in such a transition, most of them may as well retrain in a different business entirely and be done with it.

For this and other reasons, turning themselves into small agencies - which I think is the approach you are proposing and that has worked for you - is not a solution that will work for the vast majority of freelancers.

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/003b01d92710%2491fedfa0%24b5fc9ee0%24%40ozemail.com.au.