Literotica policy on using AI for translation from non-English languages

[Bramblethorn]

Mais, Madame, il ne m'appartient pas de la convaincre de son erreur ; le grand Voltaire ne disait-il pas que nous pouvons et devons tous avoir des opinions différentes ? La liberté d'expression et le droit de la faire connaître par la parole et par l'écrit, c'est l'un des grands acquis de la démocratie. Ce qui était bon et juste pour les Grecs de l'Antiquité l'est toujours pour nous.
Pour paraphraser Evelyn Beatrice Hall : « Je désapprouve ce que vous dites, mais je défendrai jusqu'à la mort votre droit de le dire ».

I suspect you're interpreting as an ethical judgement something that might have been intended as a statement about technical possibility.

 

[Meekly_Anna]

I suspect you're interpreting as an ethical judgement something that might have been intended as a statement about technical possibility.

Not really.
I just think it's wrong to believe that machine translations are always recognised as such. Especially not if the writer has at least a basic command of the language and proofreads the text afterwards.
Admittedly, most people are probably too lazy to do the latter

 

[debaucher]

Why not just publish in your native language? Im sure those who speak/ read it would be appreciative

Already doing that. But trying to widen my audience, that's all.

 

[debaucher]

I just did a thing and went off and did something I've been resisting for a long time - which was to do some research on how LLMs work under the hood - since it's becoming of increasing importance in both my real life and this second home.

I can see why they seem almost magical as tools. To someone with fair to middling ability, they feel like they bring a massive leap in the upper bound of what one can accomplish.

Under the layers of gilt and rococo, though, they're just a probability machine that answers the question "given this stream of tokens, what is the most likely next set of tokens"

they're phenomenal achievements. But as an engineer and writer I could never in good conscience trust the unmoderated output of one. And maybe I'm a dinosaur who's destined to go out with a bang (hur hur) but I don't see myself ever willingly using one.

As a software dev myself, I find them useful in a lot of situations. But what you said about unmoderated output is true. Puns especially never translate well. Anyway I've decided to not take the LLM translation route. Still, I have a tendency to use em dashes, semicolons, and paranthseses (even when writing initial outline drafts) — all of which are these days treated as AI red flags, so I worry a bit that even my own translations might get flagged.

 

[StillStunned]

Not really.
I just think it's wrong to believe that machine translations are always recognised as such. Especially not if the writer has at least a basic command of the language and proofreads the text afterwards.
Admittedly, most people are probably too lazy to do the latter

A "basic command" isn't nearly enough to turn a machine translation or AI translation into something usable, unless all you want to do is understand what it says. If you want it to be a real translation, you need a real editor. Preferably, you need a real translator to do the work in the first place.

 

[AwkwardlySet]

@Meekly_Anna

I see two distinct points of disagreement in this discussion.

First, you seem to think that translating a work of fiction doesn't involve actual creativity and should thus be left to AI without any ethical qualms. As you can see, most of us here disagree about that.
Second, putting all the disagreements aside, even if Laurel recognized the validity of your claim, how is she to distinguish between a work of fiction fully generated by the AI and one merely translated by the AI? They will read almost exactly the same, and trigger the same kind of response from whatever algorithm she is using.
So, how would you go about proving your claim? By submitting the original text in French and expecting Laurel to check that as well? Is she supposed to be fluent in French? Or have an algorithm that works for French just as well as it does for English?

 

[AwkwardlySet]

And @debaucher would you please edit and correct the title of the thread? The typo short-circuits my brain.

 

[XerXesXu]

I tried it out. Some of you may not be aware that LLMs were born from a machine translation model.

I've translated many samples from my own work using Google Translate and fed the translation into AI detector. The translations were faithful to the original; all 0% AI.

I did the same with a story recently posted here, in the forum, that on a visual scan read as AI. It was detected as 100% AI in both English and French.

That's pretty much what you'd expect. Translation is playing to the strengths of LLMs.

Has anyone else carried out a similar experiment using their own work and known AI generated work?

 

[Bramblethorn]

Not really.
I just think it's wrong to believe that machine translations are always recognised as such. Especially not if the writer has at least a basic command of the language and proofreads the text afterwards.

I'm sure they're not. But I expect the same is true for LLM-generated stories, when the writer makes some effort to massage the content into shape.

 

[Jorunn]

It is interesting to compare this thread with the one, “do you write for yourself or for readers”. Clearly, using AI translations are not ‘for yourself’, regardless of category or content. English is pervasive in many ways and places, and I speak and read it often. But I write here specifically to improve my writing skills in English (which I otherwise use infrequently or quite briefly).

I feel I have gotten better as I write on Lit in English, and I use dictionaries, a thesaurus and a rhyming site. I spend time doing research on different story ideas and the history of places. I enjoy that part.

I encourage ESL writers to write in English. You will get better the more you do.

 

[debaucher]

And @debaucher would you please edit and correct the title of the thread? The typo short-circuits my brain.

Yep, noticed that just as I saw the notification for this comment. Fixed!

 

[Meekly_Anna]

@Meekly_Anna

I see two distinct points of disagreement in this discussion.

First, you seem to think that translating a work of fiction doesn't involve actual creativity and should thus be left to AI without any ethical qualms. As you can see, most of us here disagree about that.
Second, putting all the disagreements aside, even if Laurel recognized the validity of your claim, how is she to distinguish between a work of fiction fully generated by the AI and one merely translated by the AI? They will read almost exactly the same, and trigger the same kind of response from whatever algorithm she is using.
So, how would you go about proving your claim? By submitting the original text in French and expecting Laurel to check that as well? Is she supposed to be fluent in French? Or have an algorithm that works for French just as well as it does for English?

As a general rule, you should leave discussions in which you are told what you supposedly believe. I do not.
Writing and translating are both creative processes. I have never claimed otherwise. I am merely saying that ai can be a help.
To your second point, I don't give a shit how she wants to test it. I think the request not to post machine-generated stories here is a good one. But to simply ban everything that an algorithm (!) claims was created by an algorithm? That's equal parts funny and ridiculous.

 

[AwkwardMD]

To your second point, I don't give a shit how she wants to test it.

AI-assisted translations have repeatedly been rejected by the system. At least one AH regular tried to machine translate her native English story, that was already posted on Lit, into a Spanish language version for Spanish readers, and it was rejected. The system cannot differentiate between prompt-generated AI writing and AI translations. They are indistinguishable.

I have sympathy for writers whose native language is not English, and in fact recently started editing for one who had previously used AI translation tools to get some phrases they weren't fluent with (and received a rejection). Lit does not make exceptions for this. If you need help, seek out a volunteer editor.

You are wrong about how it works. It is not an algorithm to spot algorithms. It has no history or pattern of false positives. At this point, between the authors who participate in Story Feedback, the AH, and the Editor's Forum, we would be seeing more rejections from within our own ranks.

It's not the Blackwall. It's something else.