https://www.duolingo.com/Doctor-John

Duolingo Feedback accepting a translation suggestion

Doctor-John
  • 25
  • 24
  • 11
  • 179

I sometimes get emails from Duolingo like the following:

Hi Doctor-John,

You suggested “It is certainly my elephant.” as a translation for “Es seguramente mi elefante.” We now accept this translation. :)

Thanks for the contribution, please keep it up!

  • Duolingo

This was in response to my reporting that my translation should be accepted on a question from learning Spanish as an English-speaker.

Yesterday, however, Duolingo sent me the following email:

Hallo Doctor-John,

Du hast “Generalmente yo no leo por la noche.” als Übersetzung für “Generally I do not read at night.” vorgeschlagen. Wir akzeptieren diese Übersetzung jetzt.

Danke für deinen Beitrag. Mach weiter so!

  • Duolingo

This email is also about accepting a translation in learning Spanish as an English-speaker. So why is this email in German? Is this a mistake or a deliberate choice by Duolingo?

I really appreciate the acknowledgements. I know Duo aggregates suggestions and acknowledges everyone who makes a suggestion, if and when it's accepted. You don't have to be the first person to make the suggestion.

Naturally, when it's a suggestion on learning Spanish from English, I expect the acknowledgement to be in English. So did Duolingo make a mistake about my native language (English)? Or was this their way of acknowledging my progress in German, despite the fact that the suggestion was about Spanish and English and had nothing to do with German?

You see, I'm actually pursuing three courses: German from English, English from German, and Spanish from English.

In case it's relevant to what happened, here's my current status:

German (from English): level 22, next level 404, fluency 66%.

English (from German) level 13, next level 450, fluency 70%.

Spanish (from English) level 10, next level 370, fluency 66% (down from 68%).

It would be nice to think Duo didn't make a mistake and is patting me on the back. Still, an acknowledgement in German regarding an English-to-Spanish translation seems very odd.

If Duolingo has the means and can deliberately do this, when they have evidence that the recipient will understand the message in a third language, I think this is a great idea!

So I'm not complaining. I'm puzzled.

If it was an accident, I'd very much like to know how this can happen.

Also, if it's an accident, I'd like to suggest doing it deliberately, when appropriate, but perhaps indicate in some way that it's not a mistake. :-)

Thanks, Vielen Dank! Muchas gracias.

6 months ago

11 Comments


https://www.duolingo.com/garpike
garpike
  • 25
  • 24
  • 22
  • 20
  • 19
  • 19
  • 18
  • 17
  • 17
  • 17
  • 17
  • 16
  • 16
  • 15
  • 15
  • 14
  • 14
  • 14
  • 14
  • 13
  • 13
  • 12
  • 11
  • 11
  • 11
  • 10
  • 10
  • 9
  • 8
  • 8
  • 6
  • 6
  • 6
  • 5
  • 5
  • 1158

The e-mail was sent whilst you were using German as an interface language (presumably doing the reverse course). Any time you are learning Y from X, Duolingo's e-mails will be in X until such time as you change to another base language, regardless of the languages relevant to the course to which the suggestion was made.

6 months ago

https://www.duolingo.com/Doctor-John
Doctor-John
  • 25
  • 24
  • 11
  • 179

Lol! That makes perfect sense. So a lot of people doing reverse trees will get this kind of surprise!

Thanks for the reply 12 minutes after I asked!

Related questions:

So when I'm not logged onto Duolingo, new email from Duo will be in whatever interface language I used last. It won't default back to English.

Does Duolingo even know that my native language is English? (If so, how?) Or do they just know that I'm doing German from English, English from German, and Spanish from English?

Thanks so much! I love Duo.

6 months ago

https://www.duolingo.com/Thomas.Heiss
Thomas.Heiss
  • 23
  • 15
  • 10
  • 697

No, they don't.

  • there is no 3rd language interface / e-mail setting
  • no 3rd language setting for "tips and notes" (if you do laddering trees)
  • no native language / 2nd language option in the user settings
6 months ago

https://www.duolingo.com/Doctor-John
Doctor-John
  • 25
  • 24
  • 11
  • 179

Thanks, those are good points. Of course, anyone who looked at my levels in German from English 22, English from German 13, and Spanish from English 10, could easily guess that I'm a native English-speaker who finished the German tree, then decided to do the reverse English from German, and while still pursuing that, decided to start Spanish (first from my native language English). But that's a reasonable deduction, not something explicitly encoded in Duolingo's data.

Note: I had to google "laddering in Duolingo" to discover what you meant by the term. I was familiar with the concept, but didn't know what it was called. So thanks for that. When I finish the English from German and Spanish from English trees, I'll have to try laddering Spanish from German and German from Spanish. Even now, when doing the Spanish from English, I often ask myself "How do I say that Spanish sentence in German?" Whether Spanish from German or vice versa will help me bypass thinking in English remains to be seen. But even now, I find I know German words I didn't know I knew until I need them, and the right Wortstellung (word order) is usually automatic.

6 months ago

https://www.duolingo.com/piguy3
piguy3
  • 25
  • 25
  • 25
  • 25
  • 21
  • 21
  • 18
  • 17
  • 17
  • 15
  • 15
  • 14
  • 14
  • 12
  • 12
  • 12
  • 10
  • 9
  • 6
  • 6
  • 6
  • 6
  • 43

anyone who looked at my levels in German from English 22, English from German 13, and Spanish from English 10, could easily guess that I'm a native English-speaker

It would be a reasonable guess, but plenty of non-native English speakers have a similar profile if their native language doesn't happen to have many courses from it.

There are plenty of Romanians, Finns, Swedes, Czechs, Bulgarians, etc, etc who fit this category, not to mention the people who do have options to learn their target language from their native language but just start out with the from-English courses anyway. I don't fully understand this in every instance (in some cases the from-English course really is better developed, but I don't know how you'd know this in advance), but some pretty active forum participants have fit in this category.

6 months ago

https://www.duolingo.com/Doctor-John
Doctor-John
  • 25
  • 24
  • 11
  • 179

Your point that in absolute numbers there are a lot of people learning language A from B, whose native language is C is well taken. And yes, the choices people make range from obvious to puzzling.

Still, we agree that with my profile the best guess is that my native language is English. The only question would be with what probability.

German is currently available form 5 languages: English, Spanish, Russian, Portuguese, and French (in decreasing order of enrollment).

English is currently available from a whopping 19 languages. It thus seems safe to say that most people whose first Duolingo course is English will be taking it from their native language.

Spanish is currently available from 4 languages: English, Portuguese, French, and German (in decreasing order).

The percentage of people learning German from English whose native language is English might be as low as 80%, but the probability that a non-native English-speaker would choose this as their first course isn't 20%. It might generously be 10%.

The percentage of people learning English from German, whose native language is English, may be only 10%. But the probability that a person who first pursues German from English to level 22, then adds German from English to level 10 is a native English-speaker is conservatively 95%. Thus, this behavior suggests that the probability of not being a native English speaker is generously 5%.

The percentage of people learning Spanish from English, whose native language is English might be only 80%. But the probability that one would choose German from English to level 22, then English from German to level 13, and then start Spanish from English to level 10 must increase the previously estimated 95% probability of a native English-speaker to about 99%.

In short, the absolute number of people with my profile whose native language is not English may be impressive—a fact or likelihood I had failed to acknowledge. But it's not impressive as a percentage. On the contrary, I think my profile combined with its chronology suggests a native English-speaker with an extremely high probability. :-)

6 months ago

https://www.duolingo.com/piguy3
piguy3
  • 25
  • 25
  • 25
  • 25
  • 21
  • 21
  • 18
  • 17
  • 17
  • 15
  • 15
  • 14
  • 14
  • 12
  • 12
  • 12
  • 10
  • 9
  • 6
  • 6
  • 6
  • 6
  • 43

Since you seem to be up for shooting-the-breeze statistical discussion, I'll point out that your back-of-the-envelope statistical reasoning has some issues.

Primarily, you seem to assume that all these probabilities are independent. They're not. For example, I don't think that considering the German reverse tree likely alters the percentage estimations for the first part much at all.

I don't have an a priori reason to assume that non-native English speakers who first worked on the German-from-English tree differ much from native English speakers in their likelihood of choosing the corresponding reverse tree. If they chose to learn German from English and they choose to do a reverse tree, it's overwhelmingly likely to be the English from German one.

If a non-native English speaker decided to learn German from English and they also decide to learn Spanish, then they're almost assuredly going to learn it from English as well (unless they laddered from German, but a native English speaker has that option equally). Off the top of my head, every language you can learn Spanish from on Duolingo you can also learn German from with the exception of Chinese, and the Spanish from Chinese course is quite small so unlikely to alter conclusions.

6 months ago

https://www.duolingo.com/Doctor-John
Doctor-John
  • 25
  • 24
  • 11
  • 179

The probabilities are neither completely independent nor completely dependent. Each of the three courses and the order in which they began increases the probability that the person is a native English-speaker. Only our estimates of by how much differ (by more than I expected). Most telling, actually, is the English from German, because English can be pursued from 19 different languages.

Since German is currently available only from 5 languages (English, Spanish, Russian, Portuguese, and French), many learners of German may take it from English, not because that's their native language, but because they know some English, presumably more than they initially know German. I agree that someone who completes German from English is somewhat likely to pursue English from German, even if English is not their native language. But the percentage of non-native English-speakers who's first course is German from English and who moreover complete the tree is small compared to native English-speakers with this pattern.

I'm guessing that the number of people who Complete the A from B and B from A trees whose native language is C is quite small compared with the number of native C-speakers who Start A from B.

IF and when I ever start and complete the Spanish from German and German from Spanish trees, the probability that my native language is English will rightly appear to be less than it appears now. :-)

Laddering is quite intriguing and challenging. To what extent does it help the learner bypass his or her native language? No doubt, it helps some more than others. But I'll bet that Duolinguists who complete a double ladder are blips on the radar picking up all the other traffic.

6 months ago

https://www.duolingo.com/garpike
garpike
  • 25
  • 24
  • 22
  • 20
  • 19
  • 19
  • 18
  • 17
  • 17
  • 17
  • 17
  • 16
  • 16
  • 15
  • 15
  • 14
  • 14
  • 14
  • 14
  • 13
  • 13
  • 12
  • 11
  • 11
  • 11
  • 10
  • 10
  • 9
  • 8
  • 8
  • 6
  • 6
  • 6
  • 5
  • 5
  • 1158

It would be a reasonable guess, but plenty of non-native English speakers have a similar profile if their native language doesn't happen to have many courses from it.

I easily guessed. His post was quite long, and in entirely natural English; chances are that a non-native speaker would have introduced at least some trace of a wonky phrase or otherwise slightly unidiomatic usage which I'd have readily noticed. This greatly sharpens Occam's razor when applied to the levels. Even many the very best non-native contributors on DL leave these little clues when they write long posts in English; it is remarkably hard to pass this 'English Turing test', so you really ought to factor this into your statistical analysis.

6 months ago

https://www.duolingo.com/piguy3
piguy3
  • 25
  • 25
  • 25
  • 25
  • 21
  • 21
  • 18
  • 17
  • 17
  • 15
  • 15
  • 14
  • 14
  • 12
  • 12
  • 12
  • 10
  • 9
  • 6
  • 6
  • 6
  • 6
  • 43

@ garpike

All true. I'll admit to knowing a lot more about statistics than I do about natural language processing. The question about the probabilities of native speaker status conditional on Duolingo course selection history is an empirical one that would require a computer to crunch all the data. To add the "English Turing test" systematically, then, one would need have it be even more Turing and have it computerized. Of course such a system is conceptually feasible. I don't know how good such systems are in practice. This appears to be a relevant effort from not too long ago: https://arxiv.org/pdf/1211.0498.pdf I think it's saying their system got to 74% accuracy.

6 months ago

https://www.duolingo.com/Doctor-John
Doctor-John
  • 25
  • 24
  • 11
  • 179

The question of identifying a person's native language takes an interesting turn with garpike's adding the analysis of samples of the person's natural-language writing. I agree that only a native English-speaker or bilingual could produce extensive, "flawless," complex English prose (without assistance) such as we see here.

The original question was with what degree of confidence Duo could identify my native language based on the levels and chronology of my 3 courses. Someone engaged in solely German from English at level 22 (now 23), English from German at 13, and Spanish from English at 10 (now 11) is probably already quite a bit more than 50% likely to be a native English-speaker (imo). I've felt that piguy3 underestimates the significance of the chronology. Admittedly, the probability of being a native English-speaker would be even higher, if German and Spanish were available from more languages.

As for Duo's deducing someone's native language, garpike's including text samples from the learner is totally appropriate. The learner's texts are as much a part of the available data as the chronology and levels of his or her courses. A human analyst could be 99% certain (assuming the learner writes with no assistance). A purely software/AI analysis might have to claim a little less confidence, depending on the current state of the art.

Garpike writes: "Even many the very best non-native contributors on DL leave these little clues" (a wonky expression or two). I agree. Garpike's native language is surely English. Writing "many the very best" instead of "many of the very best" comes from writing rapidly in your native language and confidently not taking the time to proofread. :-)

Adding the textual data to the mix should enable a human or software to surmise still more about the person, such as country of origin, educational level, etc.

Textual analysis by humans has been used to narrow the search for terrorists and other criminals leaving samples of their writing. And could be used to debunk, if not confirm, the authenticity of manuscripts.

6 months ago
Learn Spanish in just 5 minutes a day. For free.