Immersion – How is it supposed to converge to the optimal translation?

I’ve been using Duolingo for nearly a year and seen various incarnations of the “translate the web” interface. I’ve just returned to try out the latest – Immersion.

I could understand what Duolingo was trying to achieve by crowd-sourcing translations, and made various suggestions at the time about how the scoring needed to be optimized to cause the best translation(s) to “rise to the surface”.

This model seems to have been abandoned completely in the Immersion model. All I see are Duolinguists putting forward their latest idea (good or bad) about how to translate a sentence and that then becomes the de facto translation, until someone else comes along and changes it again – either to something different or back to how it was before it was “hacked”.

Can someone explain how this is ever going to converge or stabilize into anything like a reasonable translation, or is the final translation going to be the result of the most dogged Duolinguist’s determination to have it their way?

April 29, 2013


It won't: crowd sourcing a professional skill will never result in professional standards. There is a lack of Translation Theory (needed to set the tone for each document). Basically refusing to use people who have spent many years getting credentials and qualifications to perform correct translations understanding the pragmatics and above sentence level semantics leaves you with this sort of mess.

April 29, 2013

I certainly agree with that, unless you carefully moderate the contributions, as they have done with Wikipedia. Perhaps I should have said: "converge towards the optimum translation" (as written by a professional).

April 30, 2013

I know that the Cult of the Amateur wants to believe they can pull off anything (in this case crowd source and then sell that product--like the recaptua paradigm)--but like trying to crowd source plastic surgery--you only end with a hack job because the people have not been through the actual professional process of learning the skill.

May 3, 2013

Translation is still under development and the team is running constant studies of its effectiveness. So far, they find Immersion concept the best of everything that has been tried.

I don't know if the translations are ever published. In any case, if they are, I'm sure they get proofread.

I'd suggest you to pay your main attention to the quality of your own translations. If somebody edits your translation and makes it better, this is an opportunity for you to learn. If he makes it worse and insists on keeping his translations, let it be, "edit wars" are not worth wasting your time.

April 29, 2013

But there are not only the one author of the good/proper translation and the one author of the last, whatever the quality, translation. There are others, who see the last translation not the best (how many people would use the option to see the translation history?). And those people may have insufficient knowledge to see that the last translation is bad. And, obviously, bad translation is misleading for those who try to learn. So it is not about "edit wars" or jealousy or something like that. Showing the bad translation when you have the good one is at least strange.

And for the translator himself there is no way now to see if his variant is good or bad. One can only see that someone else "revised" his translation, but for better or for worse is unknown. Obviously, we need some method of scoring. Scoring both correctness of the translation, and of the translator.

April 29, 2013

I think kosmozhuk has a point (if I've understood him/her correctly).

For myself, the only purpose of me doing translations is to keep my French up to date and learn the occasional technical word. I read French well enough to understand what it means in French without translation. The only advantage to me of reading the English translations of others is to see how I could have put my English translation more elegantly by using an alternative word or phrasing. So I'm certainly not going to waste my time entering into "edit wars" with others re-correcting plainly bad translations.

But if you are a beginner and even working (hopefully) with a dictionary, you may still accept the translation of others at face value, without realising that what you are reading is incorrect or has missed an important nuance.

For instance, one contributor translated "Gare St. Lazard" as "St Lazard Park". Sure, "gare" does mean "park" - as in to park your car! It can also mean "take care!" or "look out!", but the only correct translation is railway station.

So unless you know French well enough to know all the possible meanings of a word (it is particularly difficult in French to recognise the root of a conjugated verb like "garer" above), and then choose the correct one for the context, just seeing a bunch of alterative howlers isn't really much help at all when it comes to learning French.

As you say Olimo, the only real benefit is to attempt the translation yourself first, and then see what others have made of it. So I don't understand why Duolingo now only allows one user to benefit from doing the initial translation, irrespective of what is going on in the background.

April 29, 2013

I'd say people learn by trying to translate themselves, by using their understanding of grammar and dictionaries. This is a useful practice even if you can't tell for sure if you're right. Then you look at others' translations and most often when you were initially incorrect, you get that feeling of "aha! that's what it means!", and then you go check again in the dictionary. This way you know if you are right. You can also post a discussion to get the explanations from native speakers.

And, obviously, bad translation is misleading for those who try to learn.

Everyone here knows that all the translations are made by users of different knowledge and experience, so one should be prepared to wrong translations and not take the existing translations as 100% correct. I don't think absolute beginners read the articles in the Immersion section and try to learn from the translations. People usually go to Immersion when they are able to read and understand most sentences with a dictionary. Anyway, the most useful piece of learning comes from reading the original text and not the translation. I have tried to read a lot of texts in foreign languages. Most of these texts were without any translations at all. Does it mean I have not learned anything from them? Does that mean I made any harm to my learning if I understood something wrong? I don't think so. Even if a user understands something wrong, it will only be temporary. With practice this mistake will be corrected.

As for "scoring", who will give those scores? If you think that users can't tell if a translation is good or bad, then obviously it won't be users. Who then? Some paid experts? Than what about all this 100% free concept? Users who are native speakers? Why would they want to work as proofreaders if they are not paid?

I'd like to stress once again that I don't find the current system perfect, but I also know that the developers are working on it and look at the results of our translation closely. Suggestions and ideas are always welcome, but you should understand that translations have to remain crowdsourced and not rely on some external scoring of translations and/or translators.

April 29, 2013

OK. To reasonably discuss if the Immersion does what it suppose to do, we need to know what exactly it suppose to do. Can we somehow see the goals of Immersion clearly defined? What is the basis of your assumptions about who, when and for what reasons goes to Immersion? Can we see some statistics? Can they really show us something? Have YOU seen any statistics?

As for scoring, I did mean the internal scoring. Some previous version of Immersion had a mandatory choice of "best translation" for every sentence, for example. Every user has his own proficiency score measured in skill points, level etc. This can be added to the formula, as well as info on the user's native language (obviously native German proficient in English should be given more credibility in translating from German to English, than native Spanish speaker with only basic understanding of English). So there are ways to score without any external (and, god forbid, paid) help.

Users who are native speakers? Why would they want to work as proofreaders if they are not paid?

Why would they want to answer stupid questions in Discussions if they are not paid? Why would they want to help those in need if they are not paid? Why do we discuss this if we are not paid? It's a mystery! But it does happen now and then. Why not utilize that?

Not to be rude, but why are we discussing this topic with you? Do you really know how Immersion suppose to work and/or do you have a say in how it would be developed? Maybe we're just bothering a completely innocent person (you) with details, reasons, arguments and suggestions?

April 29, 2013

I know that Immersion was tested for a while on a group of people and then rolled on to everybody instead of previous system. That means it proved better. I also asked Kristine about the translation quality and she answered it was "quite good". This is all the statistics I have.

I don't have any say in how Immersion would be developed. At least, no more than any other user. Why are you discussing this with me? Well, it just happened so that you replied to my comment and I replied to yours in return. No harm done, I hope. Maybe our discussion will be useful for those who actually develop the site.

April 29, 2013

Thanks. It's always helpful to know who we are and where we stand.

I answered and then asked about your status in this discussion, because I see a star on you userpic. Definitely means something, said I.

And since your role in the discussion is the same as mine (as opposed to a developer, for example) I shouldn't have any objection to "self-centered-ness" (real or imagined).

But still, having a list of well-defined goals for every instrument available to user is significant. Could that be done as a start, do you think?

April 29, 2013

The green star means a person is a community moderator. Our task is to watch the discussions, help people with their questions and concerns and also delete spam. Apart from this, we are users like anybody else and can express our own opinions.

Yes, a set of well-defined goals, recommendations and some statistics would be helpful. For now, we only have the developers' announcements and our best guesses as to how to use every tool most effectively. I hope to get more information on the team's view on Immersion to share with you soon.

April 30, 2013

The next step for you guys must be translating English into the foreign language. Keep up the good work!

April 30, 2013

Having just read the revised translations of a document I spent time and effort translating into comprehensible English (it is my fist language) I won't be visiting immersion again. The old system although more complicated and perhaps not perfect was far superior in producing a sensible document. Frustrating to see your efforts reduced to complete nonsense.

May 2, 2013

I've looked at your last translated article about nettle ( It was indeed edited, but with all due respect, I see no "complete nonsense" there. I can't judge the style as I am not a native English speaker, but the meaning of the edited translation seems correct. For example:

...personnellement je n'ai pas goûté ce mêt

Your translation: Personally I don't like the taste

Edited translation: Personally, I have not tasted this dish

This is just one example of a fully justified edit, but there are also other ones that made the translation more accurate (again, I can say nothing about style, but accuracy should come first anyway). Aren't you exaggerating a bit when you call the edits "complete nonsense"? Or maybe you are talking about a different article?

In any case, your effort was not wasted because you learned something from this translation and practiced reading in French. You are not obliged to monitor all the edits made to your translations. If they frustrate you, you can just turn the notifications off. This is what I did when I became tired to get messages about corrections, many of which were just preferential. Enjoy reading and translating, get your points and move to the next article. However, if you are interested in making the translation perfect, you can discuss the versions with other users or ask someone fluent in French and English to have a look at it.

May 2, 2013

I can't say I agree.

In a similar vain, a while ago I worked pretty hard to translate an article about strawberries. I looked things up, tried to make sense of synonyms, idioms, etc, only to find out my translations were replaced in every instance with the first word in the dictionary popup (e.g. "pied" was made into "foot" even though it's clearly an instruction about spacing the strawberry seedlings: "Place your strawberries in line, with a space of 60 cm between them and within each foot at least 40 cm" - this is not just a mistranslation, it's semantically vacuous).

It's very discouraging, and there's no easy way in the interface to recover the previous translation from the article itself, or even to point out these mistakes (the discussion feature notwithstanding; it's a bit too much for fixes like this).

I'll still keep using immersion, it's a great learning tool, it doesn't feel as forced as the lessons and there is real information to unlock by doing it, but it's a real shame that the last edit wins and that's that when there's a clear incentive for doing shoddy work. I don't think it matters if people get points for shoddy work, that's their problem, but it certainly doesn't feel like I'm helping get anything translated in the long run.

My translations were definitely pretty bad. People improved some of my bad english and corrected translation mistakes and small gaffs. This was both nice to see and perhaps more importantly to me, helped with my learning. But if there's no effective way to moderate and make a distinction between such edits and mindless point racking, what good is it that the UI is simpler? useful information just gets lost in the usual background noise of the internet...

May 2, 2013

i should clarify - i don't consider the revision selection a solution, since jumping back to a previous revision is not a way of providing meaningful feedback, and therefore is likely to result in edit wars.

there's a need for something in between the discussions and reversion IMHO.

May 2, 2013

Olimo, There was one sentence towards the end that was a much better translation than mine and it would have been great as in the previous system if I could have voted for it. I also have no desire for my translations to be used if they too are not the best. Unfortunately this isn't an option..

Apart from that one sentence the English elsewhere was poor and one sentence had been repeated. As you say you can't judge style as you are not an English speaker,

I loved the old system, I actually liked translating and then being able to see a selection of better and worse efforts than mine. I also felt satisfaction that I had contributed to a document in a positive way, regardless of whether my translation was used or not, by using the voting system. I am not doing it for points, I need to learn French pure and simple. This system doesn't help anyone.

May 3, 2013

I can see some other phrases where the corrections made the translation more accurate even if not better styled, I just did not aim at discussing every word and phrase. As for the style, you can correct it if you care about this translation, keeping the justified corrections in place.

The old system was just less transparent. You did not get notifications about the translations selected as final, but only notifications about suggested corrections of your own translations. Now you get all the notifications if you don't turn them off, and this leads to more frustration. With the old system you would probably never notice that a worse translation was upvoted and shown as best instead of yours, but this happened, too.

May 3, 2013

It is not always about style it is also, as nothingmuch is trying to explain with his strawberries, about understanding subject matter. I love gardening so I chose to translate those documents both for practice and to gain factual information. The translations need to make sense. Accuracy is impossible in translation surely the whole point of immersion, as compared to the practice sections of Duolingo, is that you shouldn't be translating on a word by word basis.

Thank you for removing the name I have deleted my paragraph in kind.

May 3, 2013

I'll answer in your stream to save this thread from being flooded with discussion of particular phrases.

I did not post all this to discourage you, but it was hard to restrain from comments when quite valid corrections were called "complete nonsense". Some of the corrections were good, some perhaps not so good, but this is not the case where someone posted nonsense. For real vandalism, you can use to report those who replace normal translations with Google Translate stuff or spam.

May 3, 2013

I did originally try to reply directly to your post but for some reason it wouldn't work. I guess these streams are another complication.

I stand by my statement the article was a work in process the revised translations in the main did not make sense pure and simple. It wasn't vandalism it was somebodies attempt at translation. Just as some of my attempts might come across as nonsense and I would rather they didn't affect the actual document. I certainly would not take it personally and for the life of me I can't see why you are when you weren't even involved.

May 3, 2013

@ Olimo and Rachel66E

I am not going to comment on whether Rachel’s translations were better or worse than the ones that replaced them. I have my own views on that, and I would share those with Rachel66E, but not if I have to do it by dissecting her translations in open forum. That is not an effective way to help someone learn who might be sensitive about their work, and one reason why I miss the old one-to-one edit/comment, accept/reject approach.

Rachel is now feeling demotivated, because the time and effort she expended to come up with her translation has been replaced, but not acknowledged one way or another – not even by the DuoOwl. She has received an arbitrary allocation of points for typing something into the box. No one has taught her anything about how her translations might be improved. And she now says she will not return to use Immersion again, which I think is a shame given the standard of her attempt. She clearly has the aptitude to learn and improve her French.

Still, I think your debate is a perfect example of what my Discussion point is about.

Rachel has worked long and hard to produce what she thinks is a good translation. Like any person who takes pride in their work, she also thinks it is good, unless someone or some system can show otherwise. And there is nothing better than instant feedback to reinforce learning, rather than to find your efforts have been "hacked" without explanation after several days.

The person who replaced Rachel’s translations may think their translations are better. (So does Olimo, apparently – but you don’t get a vote unless you review the article yourself.)

If Rachel was pig-headed she could just return to the article and revert her translations. And what would that teach her?

Worst of all there is no arbitration and no protection against deliberate damage. The last editor is judge and jury, and for some it may end in tears and frustration, because the current system feels neither objective, democratic, nor fair - whether or not it delivers its design objective.

May 3, 2013

@ 1km @ olimo. Hi, This is an important point. I must say, that I liked the old system much better, perhaps because it was less transparent. Every one of the languages I am using here is a foreign one for me. That nonwithstanding I tried translations. Of course my English was not perfect, sometimes I did not understand the text, but so many persons butchered the text (referring to the plastic surgery). That was all right in the old system because I was not confronted with that. In the new system I saw translations being kicked up and down. There were real good translators, correcting small or big mistakes I made, but mostly the corrections did not make sense, people missing important points in the text. All in all, I dropped the translation part of Duolingo.

May 3, 2013

Thanks for this comment, it really helped me to look at the problem from a different angle. It is even better than your original post :-)

May 3, 2013

@Olimo - Thank you - it's my teaching experience. No great insight.

May 3, 2013

Thank you 1KM, for your sums up how I was feeling.

Besides not being pigheaded I really don't have the time to go back and redo obvious errors that have now been placed into a document especially as that does not help me improve my French. Also having your translations instantly accepted puts a huge amount of responsibility on a person. I was rather shocked when I got to the end of the article and realised that this is what had happened.

Out of interest 1KM how did you find my translation attempt because all I can find is the newest version?

May 3, 2013

@Rachel66E - Thank you for the feedback.

Click sentence "Edit" icon. Then click "Show revisions". Optionally, in the "Revision history", you can select and alternative translation and click "Revert to selected ...".

"I was rather shocked when I got to the end of the article and realised that this is what had happened." So I think you can see how easy it is for another DL to make the same mistake and replace your translations. Not always deliberate, so don't take it too personally, although I would always try to leave a comment if I change an earlier attempt.

May 3, 2013

Thank you for that information. I think the problem for me was that there had only been one translation attempt previously so there were no previous revisions to compare with.

I agree about it 'not being deliberate' as I myself did a bit of a hatchet job on the original translation without realising until it was too late.

May 3, 2013
Learn a language in just 5 minutes a day. For free.