Thoughts on automated AI translation

[tvavrda (Tomas Vavrda)]

EDIT: This topic started as a kind of a feature request that would let me, as an author, have more control over how my contributions (posts) are handled. It resulted in a much deeper thinking about how content AI translation could end up and if it is meaningful and possible at all in a context of this forum (and similar). Follows my original post:

---

I want to continue on what I realized when discussing the pitfalls of automatic translations. Most of the discussion was about how the Discourse Admin can set up the site so it translates (or doesn’t) nicely and when it does, how should the users be informed on what they read (what is the original language, if it is a translation etc.).

But I have another point of view:

@tvavrda

And now I actually realised that what I wrote just now was translated automatically into other languages. I have no idea what the translation is. And people speaking those languages might not even realise I am not speaking that language. What do I say there? I feel strange. Someone is creating an image of myself in another language, another culture and I have no way to influence it.

There was an argument that proxies do some translation in order to bring traffic:

@nat

Just a quick note here – Google already translates our posts even before this feature was implemented. When a related query is searched for and unsupported, Google puts the forum through their proxy, e.g.

But I want to talk about the source of information, meaning when you finally get to the post the author wrote, what will you see?

So let’s take Meta as an example and have a look from these different points of view and aims:

• Site owner (CDCK): We want to talk to as many people as possible, so we turn on AI localization
• Reader: I want to read easily everything in the languages I am confident with (my mother language and whatever others I learned I want to read the originals, otherwise auto-translate into my mother language)
• Writer: Now there are two options
  A) I write in any language and I am fine with you translating it to anything you want
  B) I am very careful on what I write and how I write it in the language I have chosen (here on Meta it would be English as the original forum’s master language) and I want to opt out from automatic translations that you offer.

So B is my point. Me as a contributor, I want to have some control over what you do here with my contributions, no matter how you imagine the forum to work. I am fine that some people will not be able to understand what I say. If I write English I may expect people also learn English in order to understand me—I had to learn English myself in the first place. And I don’t want to be misunderstood just for the sake of being heard by as many people as possible.

So this feature is about the author to be able to control the level of translation of their posts.

And if you would imagine implementing it, of course it can get more complicated—if I speak multiple languages I may go as far as localizing my post manually. Not likely, but in theory, yes.

[Jagster (Jakke Lehtonen)]

@tvavrda

If I write English I may expect people also learn English in order to understand me

That is… something. Are you really demanding that from every living person in the Earth, just because you’ve learned English

Sure. It could be done. Then every posts where author says what a reader must can, should be hidden from every single translated topics. Because that would then be just fair.

This will be interesting world when every italian, spanish speaking, portugese, russians, chinese, vietnamise… will follow your idea.

[tvavrda (Tomas Vavrda)]

Are we discussing the technical and philosophical aspects of translations or each other’s taste and life experience here?

[Jagster (Jakke Lehtonen)]

Why do you think those are un-related? You just told why you want to get something using very philosophical reason and now some reason is not suitable?

I really dislike double standards. But it is not your place to say what I can or cannot do. You make your choise to write or not, nothing more.

I’m off.

[tvavrda (Tomas Vavrda)]

@Jagster

This will be interesting world when every italian, spanish speaking, portugese, russians, chinese, vietnamise… will follow your idea.

It will be a very sad world if people stop trying to learn each other’s language. A global war in less than 5 years. Even without AI being evil. And I am being optimistic here.

[tvavrda (Tomas Vavrda)]

@Jagster

You make your choise to write or not, nothing more.

Exactly and this post is all about my choice and CDCK’s choice to let me in or not.

[tvavrda (Tomas Vavrda)]

@Jagster

That is… something. Are you really demanding that from every living person in the Earth, just because you’ve learned English

Thanks for your feedback but I guess it might be even against the community rules to react on a topic just because you don’t share the author’s point of view. There are many Discourse features I wouldn’t use and I don’t announce that on each of them, much less telling the authors how stupid they are.

If you, as an author, don’t care how you are presented—I am totally fine. I am not and I didn’t come here to hear I was a wrong person and I think it’s not even a goal of this forum. I am more than fine hearing that it won’t be done, though or that there is another solution to my problem. A civil and constructive discussion. Thank you.

---

And as for your question: Do I demand everyone to learn English? No. I don’t need everyone to read what I wrote. But if they do, I want them to see only the original. And if they want to see translation, I want them to use their own tool so they are conscious when translating something. So they know it is not perfect. I am afraid, that, as the AI auto-translation is implemented and will be used, will make people lazy to study what the original was. And lazy to talk back to me in the language I understand. And I want to opt-out from that as an author on this site. I was very happy with English speaking community from around the world and I have a strong feeling that the content would degrade if everyone of us would start talking in our languages, while letting AI auto-translate to everyone else’s. Not with very technical topics, but definitely with deeper topics like this one.

I think auto-translation should be enabled for categories like Support or Bug but wisely considered for any deeper categories like Feature or Site feedback and enabled but redacted manually in e.g. Documentation or Community wiki. If not, even as a reader I may start to be worried about the content quality.

Basically Discourse now switched from the business of 100% super-high-quality UGC (User Generated Content) to Semi-AIGC (one post is UGC—the original, all the translations are basically AI generated, although everyone expects them to be correct—but still— they are totally AI generated and no human will read through all of them to make sure something didn’t go wrong). And since people will start responding to AIGC, it might get very messy.

It’s a big step.

And my question is: Do I want to be part of that? Or at least to have an option to stay away?

[Architect]

This is a good topic, I agree having writing being automatically translated without that being clear to readers can be a problem. Annother possible solution is there being an alert banner to notify readers what the original language is they are reading and what system was used to translate that.

English is one of the most bizzare languages that has ever existed, wherin a few minor alterations to sentence structure or grammer can drastically change the meaning of a sentence or way readers may interpret that.

[tvavrda (Tomas Vavrda)]

Now when you say it like this, maybe it would not even solve my problem.

Let’s say I am English speaking (but not native). Now we have a topic that goes like this:

1. Master Topic: SomeLanguage
2. Response: EnglishLanguage
3. Response: SomeLanguage
4. Response: TotallyDifferentLanguage

Everything gets translated into English but only the 1st response is actually from an English speaker.

Now what language is this thread in? No warnings will make it easier to me. What is the quality of such a thread? Basically everyone is responding to AI’s translation. If the original topic was not high quality already, it will get worse and worse… And I think me responding “English-Only” will not fix that.

Maybe my whole point can be reduced really. Can a meaningful discussion be maintained in a mix of languages bridged by a robot? Will opening to anyone with any language skills and lowering the barrier to the all-time-low really improve the discussion?

I have to emphasize: I am really happy with this community being full of people that invest their energy into:

1. Learning a new technology
2. Learning a new language (if applicable)
3. Learning how to discuss in the first place

One of those will not be necessary now. But not even that. It will be replaced by a robot. Win or lose?

Telephone game

It can also be used for older or adult learners of a foreign language, where the challenge of speaking comprehensibly, and understanding, is more difficult because of the low volume, and hence a greater mastery of the fine points of pronunciation is required.^[21]

[tvavrda (Tomas Vavrda)]

Zkusím si tu něco (toto píši česky) – dnes jsem od učitele dostal pětku.

[tvavrda (Tomas Vavrda)]

Now this is interesting. I wrote this sentence in Czech. We have grades numbered 1–5 where 1 is the best and 5 is the worst at school.

I wrote exactly:

Today I’ve got 5 from my teacher.

In English it translated actually pretty well, preserving the meaning (although not the very fact). I briefly checked other languages and it is not that great.

In German it said “Fünf” but they have 6 grades in Germany. So 5 means failing—but they have something even worse than that (imagine!).

But if a German speaking Swiss person reads it, they have it vice versa (5 is the second after the best)!

In Japanese, it still mentions number 5, which means actually “Best” in Japan.

Now imagine Czech, German, Swiss and Japanese go to the bar…

Oh, and I checked Finland and there it is even funnier!

[Moin]

(post deleted by author)

[sam (Sam Saffron)]

One sec…

What feature can we add to Discourse to ban “cut and paste” round trips into Google Translate???

Should we provide end users with the ability to disable “copy” by overriding CTRL-C on the keyboard???

Much of this discussion feels like an intellectual exercise. Part of TOS of the vast majority of forums out there is that “you are giving rights to the forum operators, in return for the value they provide you by giving you an online community”

---

Where I do stand strongly on is “labeling” we want it to be clear that content is translated and ensure it is always easy to reach to the original, I certainly support iterating on our UI.

“originally written in German - see original by clicking here”

Is a bit too much noise to add to every single post, but I am sure we can come up with a friendly and clear way of labeling stuff.

Also … the “I speak Lang 1 / 2 / 3 so show me original in those languages” is valuable. I have this very problem on X cause I am multilingual. Showing the languages I speak also is handy at giving better nuance to the community when cross translation is happening.

[Moin]

@sam

What feature can we add to Discourse to ban “cut and paste” round trips into Google Translate???

I don’t think anyone suggested that you prevent that. I had the impression that this was preferred because it creates awareness that it is a translation.

@tvavrda

And if they want to see a translation, I want them to use their own tool so they are conscious when translating something. So they know it is not perfect.

And I also have the impression that currently not all users are aware of that. That leads to misunderstandings like in Edit of hidden entry blocked period setting where the site setting name I provided was translated, and the user didn’t notice so it didn’t help them.

I can understand the feeling of wanting to opt out of getting one’s posts translated in order to avoid misunderstandings. As I said in the other topic: I wonder what would have happened if I had written my post that caused misunderstandings in German and AI had translated it the way I posted it. Would my intentions have been interpreted differently then? Or would the reaction have been the same?

[sam (Sam Saffron)]

@Moin

I can understand the feeling of wanting to opt out of getting one’s posts translated in order to avoid misunderstandings.

I want to take a step back here.

If it was

1. Crystal clear it is a translation
2. Somehow easier to get at original (it is already 1 click)

What problem remains?

[tvavrda (Tomas Vavrda)]

This thread was supposed to be longer but simple feedback and a feature suggestion. While discussing it I realized it is something much more complicated and it may not have a solution at all or it will take a long time before finding it.

There are many subtle issues that may appear only over some time, frustrating one use after another. I deeply value CDCK’s main motivation to create Discourse—to support civil discussions. I loved that from the first time I installed it 10 years ago. It was the selling point for me, no matter how good or bad the product was. I believed that with this mission you would create a great product. And I think I was right.

@sam

What problem remains?

Quite some.

Product Localization Issues

I localized a few software packages, Discourse is one of 2 big ones. I also created several large software packages. It’s a lot of work with the language if you want to do it properly. Naming something inconsistently frustrates the users. Being it in the app or in the documentation. Sometimes I find a better Czech word for what you have a simple English term (but we don’t) and I have to go through the whole Discourse translations in order to check it is being used consistently.

So then you see @Moin‘s trouble with just a small bit—a quoted setting. If you support Czech one day and every Czech person will be able to read all Meta in their language, I am worried. Because I bet that 50 % will be translated badly. All the site settings etc. will have incorrect names. Because it was not straightforward to find words that have similar meanings to what you mean in another language. There were several options and I had to choose. Sometimes against the translation AI suggestions.

Deeper Discussions

Then there are discussions like this one. It is a complicated discussion from my point of view. It is critical, so I really have to think a lot and write things with some cultural perspectives in mind. I kind of know to whom I write because I read your profiles and know the countries you are from. The fact that we all know English already means something. It means we came closer and share some bits of a common culture already. We kind of know what sounds funny in English so we don’t have to always say things like “fyi that’s a Czech joke”.

We already know the obstacles in finding words for what we want to say, we made mistakes in the past and now we try not to repeat them. See this test I just made here

@tvavrda

Now this is interesting. I wrote this sentence in Czech. We have grades numbered 1–5 where 1 is the best and 5 is the worst at school.

If I am using a foreign language the probability is high that I am also aware of not just translation problems. It might be that I learned (maybe hard way) that grades in school are not same around the world. So I don’t tell you which grade I’ve got when in USA you have letters. AI knows as well so it translated “failing grade” to English. But failed miserably with other languages and just gave you the number. But this is exactly what I would write if I would be communicating in a Czech forum with Czech people (you all would seem Czech suddenly). So I would not worry at all that you don’t understand. And the conversation would get very weird from the 3rd post on.

And this is a very simple problem. Imagine that you start losing meaning one word after another throughout an important discussion.

Like in the Telephone game I mentioned. AI translation works as the “noise” here. You start with something and end up with something completely different.

Me, The Author

So what I wanted to suggest was that I started feeling bad as an author here. I want to contribute but I want to have more control. Like a book author who has to agree on translations of their books. The translation will never ever sound anything like the original even with the greatest care. Authors know it. So it’s not like anyone can translate their books anytime. They would choose the translator, at least, and check how the book sounds in the new language.

I have no control over it here. I want to be understood, I really care about it. Maybe 95 % of the posts here are simple technical things, be it. But I would be discouraged to start anything deeper and more meaningful if I knew I am carelessly (AI) translated into multiple languages and cultures. No amount of warning would solve it. It’s out there and people will use it because it is easy. No one will ever read the original. And people will start responding to my thread based on those bad translations and I will read bad translations of what they wrote. I would be very unhappy.

[Jagster (Jakke Lehtonen)]

@Moin

I can understand the feeling of wanting to opt out of getting one’s posts translated in order to avoid misunderstandings.

Well, if you are writing in Germany I can guarantee we don’t have misunderstandings without translations. Same thing if I use Finnish.

How is this any kind win-situation?

This is not matter of translation issues that are easily solved just by asking. This is matter of making communication as strongly behind borders as possible and force to use lingua franca — because earlier was no other solution, because there wasn’t any translators. Now we have such tool.

And I ask for what reason someone feels it’s his/hers/its right to dictate what language I use to reading — and if I’m guessing right that was Sam’s point about copy&paste.

There is already tools if someone want to read original language. There is a mark to show when something is federated (oh dear, even there someone can use a translator and misunderstand things ) and hopefully everyone is using somekind mark when an AI is participating. And we have a mark for translators.

But that isn’t the point now. It is to give writers rights to say when translations are made or not. It is not about tags and buttons.

[tvavrda (Tomas Vavrda)]

@Jagster

But that isn’t the point now. It is to give writers rights to say when translations are made or not. It is not about tags and buttons.

That was exactly where I started. And to answer your question “who am I to say?” The answer is simple: The author of the content here. The content that powers the forum. The actual product “sold” here.

But now I am asking really a more sophisticated question: Will multilinguality kill meta? Will lowering the entry barriers bring more quality? That’s more of my concern now. I know you think “Yes of course! Everyone should join!”. While I whisper “I am not sure, maybe the language barrier was a good one to filter all the noise…”.

I remember the days when we had content farms. So many websites where people would generate tons of texts in order to trick SEO. And the websites using that quickly degraded in search results. Because we all know it is not a real content, people were paid to generate it. And I have a feeling (not knowledge, cannot prove it) that auto-translations would somehow fall into that space. Not a real content.

And I would feel like being the person that supports the content farm. I write something here and 20 other versions automatically emerge. And people start writing back nonsense… And there is nothing to do about it.

Actually I also fear that meta could end up like those cheap Q&A sites that we used before StackOverflow appeared with its great curated content. Which I know has a shared history with Discourse, of course.

[sam (Sam Saffron)]

@tvavrda

But now I am asking really a more sophisticated question: Will multilinguality kill Meta? Will lowering the entry barriers bring more quality? That’s more of my concern now. I know you think “Yes of course! Everyone should join!”. While I whisper “I am not sure, maybe the language barrier was a good one to filter all the noise…”.

Meta is primarily a discussion forum for talking about Discourse features / bugs / UX niggles and so on.

If a bug is carefully described in German and then I read it in English - odds are reasonably high I will understand what the bug is. With the right prompting, we can also get AI to present that bug to a wider audience so they are clear that it was a bug and was fixed.

@tvavrda

Will multilinguality kill Meta?

Another question worth wrestling with… only 8% of Japanese people speak English, compared to say 45% of Czech people.

English as a filter for “people capable of civilized Discourse online” may hold as some sort of litmus test in some countries where all the “high quality people speak fantastic English” but certainly not in all countries.

Yes, some nuanced Discussion is very hard to translate. But it is important to note that AI translation today is the worst it will ever be. As time passed we can capture better nuance and better voicing. In fact I just read your post in Hebrew and it was pretty faithful, all your major points were captured solid 8 out of 10.

[tvavrda (Tomas Vavrda)]

But this specific use case will not happen anymore. You see we keep the thread in English while the translation can go on in the context of the whole communication (I don’t know if it does but it could). Translating one way English → other languages. So we have a stable 8/10 translation of the whole thread. That’s very good!

Let’s say Czech-Hebrew is 7/10.

So I would start in Czech and you would read 7/10 in Hebrew.

Then you would reply in Hebrew and I would read 7/10 Czech translation.

To make it funnier, now Japanese would join in with 7/10 version (but different 7/10 – some special language quirks other than Czech-Hebrew).

And let’s say @Jagster would be also willing to contribute with his very special language.

English would not be involved. At all.

People talking in their own boxes, nothing in common, not even a sense of it. And there is something out of their will that tries to make them understand each other. (Actually not—just doing its job—translation. The job is not “let them understand each other”. AI will never care like a human translator who would ask “Did you mean X or Y?” before going on with the translation.)

Are you really positive this could work in any circumstances? Even human translators would refuse to serve this, I believe. It’s not about AI being good enough. If it is 9.8/10 in a thread of 100 posts I believe it would just fail no matter what.

---

@Jagster

because earlier was no other solution, because there wasn’t any translators. Now we have such tool.

I imagine it like a campfire. People sitting together at night, talking, singing, experiencing the warmth of the fire. Now you say “We can stream the fire to the whole world, now we have the tools!”. So from now on everyone will make their own small campfire at home. You know, it’s just easier, you don’t have to travel anywhere and all that hard stuff. And we will stream the campfire to everyone’s place. Solved. You have fire? Check. You see everyone else? Check. Even more people can join! (that was not the request, but Check). So you are together at a campfire with your friends! No.

Now that reminded me the Covid parties. We actually lived through this.