Poll: How can a translator use online AI tools when working under NDA? Thread poster: ProZ.com Staff
|
|---|
| | Lieven Malaise
Belgium
Local time: 03:29
Member (2020)
French to Dutch
+ ...
If you use a DeepL subscription, your translations aren't stored or used for AI development. I suppose it will be the same for ChatGPT and the like?
Bottom-line is: you shouldn't use free tools as a professional.
Which brings me to Pastey. Can someone of the Proz staff confirm that your translations are protected if you would subscribe to it and use it? Possibly a dumb question, but I've never read anything about it so far.
| | | | Samuel Murray 
Netherlands
Local time: 03:29
Member (2006)
English to Afrikaans
+ ...
| Pastey is probably not secure | Sep 2 |
Lieven Malaise wrote:
Which brings me to Pastey. Can someone of the ProZ.com staff confirm that your translations are protected if you would subscribe to it and use it?
I think the fact that the first question any sane professional translator would ask is whether the data is secure, and the fact that the Pastey website says nothing about that, is very, very telling.
| | | |
Lieven Malaise wrote:
If you use a DeepL subscription, your translations aren't stored or used for AI development. I suppose it will be the same for ChatGPT and the like?
Bottom-line is: you shouldn't use free tools as a professional.
Exactly! The paid version of ChatGPT doesn't use input for training purposes either.
| | |
|
|
|
| Tin foil hat man speaks | Sep 2 |
I'm not really a tin-foil hat wearer, but I would draw everyone's attention to the idea that just because a company says it isn't doing a thing, doesn't make it true. All the big players have been fined for various breaches of data protection and almost certainly will be again. They can't help themselves (to be fair, I have some dim memory of a case where data was collected inadvertently somehow, but perhaps that's actually worse...).
Just pseudonymise. It only takes a couple of min... See more I'm not really a tin-foil hat wearer, but I would draw everyone's attention to the idea that just because a company says it isn't doing a thing, doesn't make it true. All the big players have been fined for various breaches of data protection and almost certainly will be again. They can't help themselves (to be fair, I have some dim memory of a case where data was collected inadvertently somehow, but perhaps that's actually worse...).
Just pseudonymise. It only takes a couple of minutes of search & replace. And is easier to reverse (for delivery of the translation) than true anonymisation ▲ Collapse
| | | | Lieven Malaise
Belgium
Local time: 03:29
Member (2020)
French to Dutch
+ ...
Charlie Bavington wrote:
I'm not really a tin-foil hat wearer, but I would draw everyone's attention to the idea that just because a company says it isn't doing a thing, doesn't make it true.
It's a written agreement. You can never be sure, but it's the basis of doing business with anybody.
| | | | IrinaN
United States
Local time: 20:29
English to Russian
+ ...
| Sure, and Santa is real | Sep 2 |
Edith van der Have wrote:
Exactly! The paid version of ChatGPT doesn't use input for training purposes either.
| | | |
Lieven Malaise wrote:
Charlie Bavington wrote:
I'm not really a tin-foil hat wearer, but I would draw everyone's attention to the idea that just because a company says it isn't doing a thing, doesn't make it true.
It's a written agreement. You can never be sure, but it's the basis of doing business with anybody.
Yup for sure.
And the chances of a translator or their client even detecting a data breach are tiny.
It would surely only emerge in the context of a larger incident.
But breach of an NDA it would surely be, and open to the relevant sanction.
And the chances of you successfully suing DeepL (as an example) to get redress for that sanction being imposed are, I would suggest, next to nil.
In situations where prevention takes so little effort, I believe it is better than cure.
[Edited at 2024-09-02 13:13 GMT]
| | |
|
|
|
Charlie Bavington wrote:
Just pseudonymise. It only takes a couple of minutes of search & replace. And is easier to reverse (for delivery of the translation) than true anonymisation
The following examples of Pseudonymization and Redaction were obtained using the Gemini API, driven from a home-brew .NET app. YMMV, especially with adjustment of the generationConfiguration parameters, and/or by tweaking the prompts:
Pseudonymization
Prompt: Apply pseudonyms to conceal the true names of all persons in the following text and then translate it into French:
'Jack and Jane went to London to visit the Victoria and Albert Museum with their cousin Harold. Afterwards, Harold took Jack to the Science Museum, while Jane went shopping.'
Gemini-1.5-flash (API)
## Pseudonymized and Translated Text:
**Pseudonyms:**
* Jack: **Ethan**
* Jane: **Olivia**
* Harold: **Arthur**
**English:**
Ethan and Olivia went to London to visit the Victoria and Albert Museum with their cousin Arthur. Afterwards, Arthur took Ethan to the Science Museum, while Olivia went shopping.
**French:**
Ethan et Olivia sont allés à Londres pour visiter le Victoria and Albert Museum avec leur cousin Arthur. Ensuite, Arthur a emmené Ethan au Musée des Sciences, tandis qu'Olivia est allée faire du shopping.
Gemini-pro (API)
**Pseudonymized Text:**
John and Jill went to London to visit the Victoria and Albert Museum with their cousin Henry. Afterwards, Henry took John to the Science Museum, while Jill went shopping.'
**French Translation:**
John et Jill sont allés à Londres pour visiter le Victoria and Albert Museum avec leur cousin Henry. Ensuite, Henry a emmené John au Science Museum, tandis que Jill est allée faire du shopping.
Interesting to note that both models respect the gender of Jane/Jill (... est allée faire ...).
Redaction
Prompt: Redact the names of all persons in the following text to conceal their true identities and then translate it into French:
Gemini-1.5-flash (API)
## French:
[REDACTED] et [REDACTED] sont allés à Londres pour visiter le Victoria and Albert Museum avec leur cousin [REDACTED]. Ensuite, [REDACTED] a emmené [REDACTED] au Musée des Sciences, tandis que [REDACTED] est allée faire du shopping.
Gemini-pro (API)
**English:**
**Redacted Text:**
[Name 1] and [Name 2] went to London to visit the Victoria and Albert Museum with their cousin [Name 3]. Afterwards, [Name 3] took [Name 1] to the Science Museum, while [Name 2] went shopping.
**French Translation:**
[Nom 1] et [Nom 2] sont allés à Londres pour visiter le Victoria and Albert Museum avec leur cousin [Nom 3]. Ensuite, [Nom 3] a emmené [Nom 1] au Science Museum, tandis que [Nom 2] est allé faire du shopping.
The output from the Gemini-1.5-flash model is redacted to the point of being close to meaningless, whereas the Gemini-Pro model keeps track of 'who is who'. However, it seems to be gender-agnostic (... est allé faire ...), which might offer better anonymity but ends up making the text somewhat ambiguous in a strongly-gendered language.
[Edited at 2024-09-02 19:37 GMT]
| | | | | Why should a translator use AI in the first place? | Sep 3 |
Hello everyone,
I must say I am very surprised, and a bit dismayed too, that no one even reacts to the fact of using AI as a professional translator.
And especially if an NDA was signed. I don't think anyone who has their professional translator sign an NDA would consider or agree to them using AI for the translation.
| | | | | Pastey uses Enterprise Licenses | Sep 4 |
Samuel Murray wrote:
Lieven Malaise wrote:
Which brings me to Pastey. Can someone of the ProZ.com staff confirm that your translations are protected if you would subscribe to it and use it?
I think the fact that the first question any sane professional translator would ask is whether the data is secure, and the fact that the Pastey website says nothing about that, is very, very telling.
Hi Samuel,
The whole (economic) point of using Pastey is that you use the different LLMs under our enterprise licenses. That should answer your question.
I was certain I had added that bit of info to the FAQ section. Just checked, and it says:
• Do I need my own LLM license?
No, you do not need your own license. Pastey is equipped with ProZ.com's Enterprise licenses for various LLMs.
| | | | Samuel Murray
Netherlands
Local time: 03:29
Member (2006)
English to Afrikaans
+ ...
Tanya Quintieri wrote:
The whole (economic) point of using Pastey is that you use the different LLMs under our enterprise licenses. That should answer your question.
Yes... no, sorry, I have no idea what that means.
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Poll: How can a translator use online AI tools when working under NDA? | CafeTran Espresso | You've never met a CAT tool this clever!
Translate faster & easier, using a sophisticated CAT tool built by a translator / developer.
Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools.
Download and start using CafeTran Espresso -- for free
Buy now! » |
| | Trados Business Manager Lite | Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
