Had a frustrating experience with a client and don't know what to do.

[nikk...@gmail.com]

Hello. I've been subscribed to this mailing list for a very long time, but haven't contributed. I actually found work through this group years ago. Somebody was looking for translators, I reached out, and got work doing a few independent manga, and that was it. A year or so later, I noticed the same person was looking for more translators for a "big project," so I reached out again and was brought into the team. I worked with them doing a lot of more popular manga (they were working directly with Kodansha). I had been working with them on this project for over a year doing various titles. There was one title in particular that was new at the time, and I asked to be the translator for that title. I became the sole translator for that title from the beginning, and I was rather proud of that. That manga ended gaining popularity, eventually branching out into anime, dramas, and a movie.

 At first it was full books, which we were given quite a lot of time to do, but when we caught up, it started turning into simultaneous releases, and we had to keep up with the new chapters week by week (or sometimes days). I kept going, and this is where some problems started happening. The team of translators, letterers, etc... were all remote and all over the world, and the management of schedules and deadlines wasn't exactly clear-cut. There were multiple databases, online servers, scheduling systems, and chat programs, all independent of each other. Sometimes a file was on server A, and sometimes it was on Server B. Sometimes the app they use to communicate with me would say the new chapter data will be available tomorrow, but it was in fact uploaded a day ago. I worked around the inconsistencies, and managed to get everything done. One of the biggest problems was how they presented the deadlines. The app that they used to communicate with me would give me a deadline in my local time (US central time). Because I really liked the manga I was working on and wanted to know what happened next, I usually got each chapter done early, I never learned that the deadlines they were giving me in my local time were in fact in Japanese time (even though it was showing up as local time for me). For example, I would get a notification telling me a script was due on February 1st at noon my time, but it was in fact February 1st at noon in Japan time. Like I said, I usually worked on them early, so it never seemed like a problem on my end, but apparently some people were getting frustrated that my parts weren't coming in earlier enough for their liking (even though they were still on time even by their deadlines). I didn't learn about this until recently when I started getting busier at home and was submitting work closer to the deadline I was given. I was finally told that the deadlines were in Japanese time and I've been cutting it too close. I explained that my notifications are giving me deadlines in my local time and asked to adjust or fix their notification system so I get proper dates and don't get confused. They did not do this, and I started getting spoken to more harshly in the work chats (i.e. supervisors/managers would specifically thank every member of team by name except me). This started to irritate me because it was all "good work" and "your translations are really good" and "we increased your pay rate because of your excellent performance." I tried to defend myself from the complaints once by reminding them that I'm still getting my work in on time, just not as early as I used to.  

At about the same time, they started having financial trouble and became unable to pay us on time, giving multiple excuses (at one time it was that their Japanese clients hadn't paid them, then they said someone stole their bank account info and they had to deal with that first). Even before then, they had issues with checks bouncing and whatnot. I hadn't been paid in over 4 months, but I kept working mostly because I liked the manga I was working on, and it was drawing to a close and I wanted to keep going until it was finished. A few days ago, I had just finished the latest chapter (and early I might add), and when I tried to upload it, I found that I'd been kicked out everything, and I was met with an email telling me I was terminated. They promised to pay me, but I am still waiting on that.

This really sucks and I don't know what to do. I'm bummed about the thousands of dollars I'm still owned and might not get, but I'm really pissed that after over 200 chapters of a manga I was personally invested in and proud to be the translator for, I got unceremoniously booted and now I have no recourse. I really want to get back into manga translation, but I don't know if I could find anyone would take me on (I ran searches and applied for what I could find, but I don't have any resources other than google). 

Sorry this was so long, but I really wanted to just vent. I don't know if this story is a common occurrence, but if anyone as anything that could help, or at least cheer me up, I'd love to hear it.

-Nicholas Plante 

[Tom Gally]

Nicholas,

That sounds terrible. You have my complete sympathy.

As others on this list probably also know from hard experience, “they started having financial trouble and became unable to pay us on time” and “they had issues with checks bouncing” are usually signs to stop taking work from a client, as the client is likely to fail soon. That can be hard to do, though, when one is owed a lot of money or one has a personal investment in the work, as you did with the manga you were translating.

I know this won’t be much consolation, but you might consider sharing your story more widely so that others can learn from it. As more freelancers—not only translators—work on globally dispersed projects, more are likely to run into similar problems—both minor ones, like misunderstandings about time zones, and major ones, like clients in other countries missing payments, cutting off contact, or disappearing entirely.

If you do post your story to a blog, other translators’ forums, or social media, you might want to anonymize it a bit more: don’t mention Kodansha by name, and make sure that people cannot identify your client from your name or from the manga you translated. You don’t want to expose yourself to a defamation lawsuit.

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/90418f42-ad4c-40e8-a064-76058dd07b67n%40googlegroups.com.

[Bill Lise]

Thank you for that detailed description of what appears to be happening to numerous people working in games, anime, and manga.

Other fora are filled with horror stories of agency clients going bankrupt or just going somewhere and leaving numerous translators unpaid.

You might not have any recourse, particularly if you are working cross-border.

Do you know the physical location of your client? That isn't a given, it appears, since numerous agents do not disclose their actual physical location or just disclose a virtual (AKA, fake) address.

Perhaps the lesson to be learned is to attempt to deal with clients that have nothing to lose by treating their vendors poorly or unethically. That points to direct clients, but my guess is that, in the fields noted above, that isn't possible for most translators, particularly with a large number of translators willing to take high risks to work in those fields.

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

And this was after I refined my prompt half a dozen times…

[Tom Gally]

Chris,

LLMs are not knowledge databases or search engines, and people who expect them to perform that way will be disappointed. While some of them can do web searches in response to prompts, their performance is not great, as they rely on the results of the web searches, which, as we all know, are often not reliable.

LLMs are more useful when, for example, they are used interactively for open-ended tasks that you are unsure how to tackle. For example:

https://chat.openai.com/share/a49aecc2-ff62-4c4b-86b1-e25016a49f2d

I didn't try running any of ChatGPT 4’s code. In my experience, its code often fails on the first try, but by feeding back the error messages to it I can usually get it running in a few minutes.

If I were to set out to write the code for a task like that by myself, I would get it running never.

Tom Gally

On Sat, Apr 13, 2024 at 10:30 AM <cpta...@ozemail.com.au> wrote:

And this was after I refined my prompt half a dozen times…

--

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/015b01da8d42%24240694e0%246c13bea0%24%40ozemail.com.au.

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

Tom you are a legend, and I a fool.

Why this is exactly how I used to treat my kids: “Go and do X. And if you don’t know how to do X find out how to do X, and if you don’t know how to find out, find out how to find out..” and so on. I must remember to be more meta.

Thanks

Chris

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CACnS3NdneV19Nm%3D2gfT98JTtTX7qtU%3DKkzjgHh3TY703qyaBXQ%40mail.gmail.com.

[John Stroman]

Chris,

If it's any comfort, I did the same kind of thing recently (shown below) and actually wrote to Tom about it. Now I realize I was a fool. Worse yet, my memory may have been playing tricks on me.

*****

Input:

I recall a Japanese samurai film in which a slain leader is strapped to his horse and rides on a hilltop behind the battlefield to rally his troops. What is the name of that film?

 

Output:

The film you are referring to is "Ran," directed by Akira Kurosawa. In "Ran," the character Lord Hidetora is strapped to his horse and rides on a hilltop to rally his troops after being slain in battle. It's a powerful scene in the film.

 

Wrong. No such scene appears in Ran.

 

Then I decided to provide more detail:

 

Input:

When I lived in Japan, I watched a Japanese film in a theater directed by Kurosawa. I have a clear memory of a scene in which a wounded or dead samurai lord is strapped to his horse and rides along a hilltop behind his troops during a battle to rally them and deceive his enemies that he is still healthy and unharmed.  The actor who played the samurai lord was Tatsuya Nakadai. The film was released no later than 1983, the year I left Japan. It is possible that this scene was cut from the DVD version of the film. What is the name of the film?

 

Output:

The film you're describing, directed by Akira Kurosawa and featuring Tatsuya Nakadai as a samurai lord who is strapped to his horse to deceive enemies and rally his troops, is "Kagemusha" (1980). In "Kagemusha," Nakadai plays both the warlord Takeda Shingen and his "shadow warrior" (or double), who takes over after Shingen dies, to maintain the illusion that Shingen is still alive. The scene you mentioned, where the lord is mounted on a horse to inspire his troops, is a memorable part of the film. "Kagemusha" was acclaimed for its powerful narrative, visual style, and depth, marking a significant point in Kurosawa's career and Japanese cinema.

 

Correct, based on my memory, but CHAT may be guessing by repeating much of the information that I included in the prompt. Ran was released in 1985, so that fact alone excluded it from CHAT's output.

 

I also gave the following prompt to CHAT:

Input:

What is the name of the actor who played the jester in the Japanese film Ran?

 

Reply: The actor who played the jester in the Japanese film "Ran" is Tatsuya Nakadai.

 

Wrong. Tatsuya Nakadai played Hidetora (King Lear). The actor who was fantastic in the role of court jester was ピーター.

*****

I vividly remember the scene I described to CHAT, but now I'm beginning to doubt my own memory. Maybe I'm the one who has been hallucinating.

John Stroman

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

[Kevin Johnson]

All of these questions have been easily answered by the top results of Google's algorithm for about a decade, so I guess it's not a great sign in any case that ChatGPT is not able to handle them (yet?).

Assuming Kagemusha is indeed the film you were thinking of, Google gives it as the second result on my end with much fewer words:

https://www.google.com/search?q=Japanese+samurai+film+wounded+leader+strapped+to+his+horse+rides+on+a+hilltop+to+rally+troops&sca_esv=89d0d0f18241ff13&rlz=1C1FQRR_jaJP946JP946&sxsrf=ACQVn0_Pxkz5j6uVQAR-jsmRyAva8QH7vg%3A1713012797079&ei=PYAaZr6-BMCMvr0PhYeK0A0&ved=0ahUKEwj-0I_6nb-FAxVAhq8BHYWDAtoQ4dUDCBA&uact=5&oq=Japanese+samurai+film+wounded+leader+strapped+to+his+horse+rides+on+a+hilltop+to+rally+troops&gs_lp=Egxnd3Mtd2l6LXNlcnAiXUphcGFuZXNlIHNhbXVyYWkgZmlsbSB3b3VuZGVkIGxlYWRlciBzdHJhcHBlZCB0byBoaXMgaG9yc2UgcmlkZXMgb24gYSBoaWxsdG9wIHRvIHJhbGx5IHRyb29wc0gAUABYAHAAeACQAQCYAQCgAQCqAQC4AQPIAQD4AQGYAgCgAgCYAwCSBwCgBwA&sclient=gws-wiz-serp

 Google has nothing to worry about on that front.

We're seeing a lot more articles these days about the unfulfilled hype of generative AI.

Anyway, I'm sure it's good at something, just not the Everything we anticipated. Companies are famously still hunting for exactly what it can be used for with little success.

Kevin Johnson

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

Wow, so wrong. Should have called me. I love Ran.

Then again my first thought was El Cid..but that was Charlton Heston..

 

Chris

 

From: hon...@googlegroups.com <hon...@googlegroups.com> On Behalf Of John Stroman
Sent: Saturday, April 13, 2024 9:30 PM
To: hon...@googlegroups.com
Subject: Re: ChatGPT

 

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/CAJVTx96FZ36-EU%3DF%3DjwbSriGFNBELGhH-%3D5M%2B-6bHpWWn-UZzg%40mail.gmail.com.

[John Stroman]

Kevin,

Thanks for the link. Unfortunately, the Google search with the link you provided does not come up with Kagemusha on my screen.

Chris,

Ah, El Cid. I have watched Kagemusha, Ran, The Last Samurai, and El Cid. Now I wonder if Kurosawa borrowed an idea from El Cid, or if I'm on another planet.

I do not remember Watanabe Ken being strapped to his horse after seppuku and leading his followers against the Gatling guns of the Imperial Army. That scene would have been entirely different. 

The scene I do remember is the silhouette of Nakudai Testsuya wearing the helmet of Takeda Shingen and riding the horse on a hilltop behind his troops.

I guess my mind must be conflating all of these films. Thanks to this discussion I'm convinced that regardless of how fluent they are, CHAT and other AI engines simply make conjectures from the content of the prompt about information that is not included in their corpora. A real human would have simply told me I'm wrong, or I must be making things up ( i.e., hallucinating). Final score: Homo sapiens 1, AI 0.

John Stroman 

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

[Anish Krishnamurthy]

"Conjecture" implies processing information and drawing a conclusion. A language model does not make conjectures because it is not a thinking entity and has no concept of a samurai, a horse, or a film. It's assembling an output that mimics the kind of response you might get from a question that sounds like that, but it's not doing it because it has any ideas of what kind of movie you're looking for, real or otherwise, only a model of what kind of text could be in proximity to what you fed into it.

-- Anish Krishnamurthy

--

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/CAJVTx9405vH5J62BcGcMHc_YFzAje7VLyAEtv5cCr%3D2t7624Vg%40mail.gmail.com.

[John Stroman]

Anish,

Thank you for the correction. I admit I was endowing CHAT with a human quality. My vocabulary seems to be limited to human behavior. If it is not guessing or making things up, then how do we describe its behavior in non-technical human language? Does this lie under the umbrella of hallucination as it is technically used in discussing AI?

John Storman

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

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CAGVW6a%3D6hnsCQfMc73%3DFui%2BC%2Bu0DGz-s-rpC%2Bdn5_3C5mx9Gqw%40mail.gmail.com.

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

Apart from including some more appropriate terminology, this video raises the issue that AI might not change much because we won’t be able to afford the power bill.

https://www.youtube.com/watch?v=0ZraZPFVr-U

Chris

 

From: hon...@googlegroups.com <hon...@googlegroups.com> On Behalf Of John Stroman
Sent: Sunday, April 14, 2024 5:19 AM
To: hon...@googlegroups.com
Subject: Re: ChatGPT

 

Anish,

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CAJVTx975aMNyfpoJCSVWc8Qptjs6XqFrbFGC2c0gs%3DsUABu-uw%40mail.gmail.com.

[Herman]

On 4/13/24 12:18, John Stroman wrote:
> Anish,
>
> Thank you for the correction. I admit I was endowing CHAT with a human
> quality. My vocabulary seems to be limited to human behavior. If it is
> not guessing or making things up, then how do we describe its behavior
> in non-technical human language? Does this lie under the umbrella of
> hallucination as it is technically used in discussing AI?
>

The term "inference" is typically used rather than conjecture to refer
to the process whereby a generative LLM generates particular output in
response to an input, with hallucination being an
incorrect/inappropriate/irrelevant inference that is not supported by
the training data. So whether the specific output you encountered would
qualify as hallucination, would depend on how it relates to the training
data.

Herman Kahn

> John Storman
>
>
> ----------------
>
>
> On Sat, Apr 13, 2024 at 2:55 PM Anish Krishnamurthy
> <anishk...@gmail.com <mailto:anishk...@gmail.com>> wrote:
>
> "Conjecture" implies processing information and drawing a
> conclusion. A language model does not make conjectures because it is
> not a thinking entity and has no concept of a samurai, a horse, or a
> film. It's assembling an output that mimics the kind of response you
> might get from a question that sounds like that, but it's not doing
> it because it has any ideas of what kind of movie you're looking

> for, real or otherwise, only a /model/ of what kind of text could be

> in proximity to what you fed into it.
> -- Anish Krishnamurthy
>
>
> On Sat, Apr 13, 2024 at 6:57 AM John Stroman

> <stromana...@gmail.com <mailto:stromana...@gmail.com>>

> <mailto:honyaku+u...@googlegroups.com>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/honyaku/CAJVTx9405vH5J62BcGcMHc_YFzAje7VLyAEtv5cCr%3D2t7624Vg%40mail.gmail.com <https://groups.google.com/d/msgid/honyaku/CAJVTx9405vH5J62BcGcMHc_YFzAje7VLyAEtv5cCr%3D2t7624Vg%40mail.gmail.com?utm_medium=email&utm_source=footer>.

>
> --
> 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

> <mailto:honyaku+u...@googlegroups.com>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/honyaku/CAGVW6a%3D6hnsCQfMc73%3DFui%2BC%2Bu0DGz-s-rpC%2Bdn5_3C5mx9Gqw%40mail.gmail.com <https://groups.google.com/d/msgid/honyaku/CAGVW6a%3D6hnsCQfMc73%3DFui%2BC%2Bu0DGz-s-rpC%2Bdn5_3C5mx9Gqw%40mail.gmail.com?utm_medium=email&utm_source=footer>.

>
> --
> 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

> <mailto:honyaku+u...@googlegroups.com>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/honyaku/CAJVTx975aMNyfpoJCSVWc8Qptjs6XqFrbFGC2c0gs%3DsUABu-uw%40mail.gmail.com <https://groups.google.com/d/msgid/honyaku/CAJVTx975aMNyfpoJCSVWc8Qptjs6XqFrbFGC2c0gs%3DsUABu-uw%40mail.gmail.com?utm_medium=email&utm_source=footer>.

[John Stroman]

Chris,

Thank you for the link to the video. Very interesting.

Herman,

Thank you for the clarification. So if I understand correctly, inference is the term I should have used, and hallucination refers to an incorrect/inappropriate/irrelevant inference.

John Stroman

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

[Anish Krishnamurthy]

First of all, it's not wrong in principle to anthropomorphize technology to some extent. People do it all the time. Even engineers talk about their device being finicky or in a mood. Besides, we often ascribe verbs like "write" and "listen" and "tell" to software out of necessity. However, I think it's especially fraught in relation to generative models.

Fundamentally, a model is a complex, opaque piece of software that generates semi-random outputs in response to inputs. Input->output is just the definition of a program. A dictionary may contain the sentence "horse (n): an equine," but no matter what medium that sentence is on, it doesn't mean it knows what a horse is. It just encodes that information to be understood by a human when presented.

For another example, I do a lot of automation and analytics scripting in support of my translation work. I might have to divide a document by scenes and parse out some information from each one. The script doesn't "know" what a scene is or why one is separate from another, it only knows that some sequence of returns and punctuation means it should log its data and reset the meter, because I programmed it to.

The thing about these tools is that we understand - perhaps not to the atom, but the logical mechanism - how they work and why they produce the information we're looking for, which is what makes them trustworthy. Generative algorithms obfuscate their working, and I say that to some extent, the obfuscation is deliberate so that their manufacturers can maintain the illusion that the programs are something more than they are: the kind of AI you'd see in scifi, or something adjacent to it, instead of an intricate top-down statistical simulator. Even the word "AI" in branding is meant to get you to think of it as a movie robot with a functioning abstract brain.

Bluntly, I don't ascribe any semantic meaning to anything that comes out of generative algorithms. It looks superficially like real content because of the corpus, but it has not been created with intent or comprehension, so I can't trust it not to be noise from the abyss. If you asked me about a movie with a mounted samurai, I would know what a mounted samurai looks like and search my memory for that, and probably tell you nothing matched what (I think, with my theory of mind) you wanted. A model will look for words that go with "Japanese movie" and "man on horse." Hopefully the one you want is dredged up with them.

There are uses for this extruded content. Tom has talked about some of them, and also amply demonstrated the care with which they must be handled, especially because the corpus is big enough that the generated output will resemble something a person might write, and the machine is designed to mimic conversations. But fluency has nothing to do with accuracy. "Accuracy" is not what the machine is for. It may arise as a byproduct of the process, but more important to the peddlers is that you assume that it is what they're for.

(And personally, I think a lot of these uses are still better handled by judicious use of smaller but more precise tools, or else experts. I still get my word choice suggestions from thesaurus.com.)

-- Anish Krishnamurthy

To view this discussion on the web visit https://groups.google.com/d/msgid/honyaku/CAJVTx975aMNyfpoJCSVWc8Qptjs6XqFrbFGC2c0gs%3DsUABu-uw%40mail.gmail.com.

[Herman]

On 4/13/24 15:55, John Stroman wrote:
>
> Herman,
>
> Thank you for the clarification. So if I understand correctly,
> inference is the term I should have used, and hallucination refers to an
> incorrect/inappropriate/irrelevant inference.
>

I would say the term "conjecture" probably makes more sense given what
you were trying to say, but "inference" is the more standard and neutral
term.

Herman Kahn

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

Hey is Geoff on channel? Can someone put me in touch with him urgently?

Thanks

Chris