Hacker News Reader

Comment by crazygringo | Hacker News Reader

Agreed. I speak Portuguese fluently as a second language, and wanted to see what the Duolingo material was like. First warning sign: I couldn't pass the test to study intermediate content, because a significant proportion of the multiple-choice answers were just wrong. Then, looking at the beginner content, a shocking number of answers were wrong as well. It was kind of funny, since each piece of material had user comments associated, you'd see 70+ comments from people all complaining about wrong answers.

That immediately destroyed any trust with them. When you put out educational material, it has to be correct. That has to be the foundation.

View on HN

I'm a Turkish native speaker and I couldn't pass their intermediate Turkish material as well.

EDIT: Also it felt like the whole game was written from an "Indo-European" perspective. For example, in Turkish there is a word "bir" (read [biɾ]) which literally means "one" which is sometimes used akin to indefinite articles (a/an). But they're not exactly indefinite articles, and sentence usually makes sense without them and when you speak fast you omit them. But in Duolingo they taught as if you should translate "the X" without "bir" and "a/an X" with "bir". That's not even remotely close to the truth (and made me get all "article" questions wrong). In Turkish definiteness is mostly denoted with accusative case. There is no single language construct similar to "article", you infer it from other kinds of information. For example "dolaba bira koy" : "put A beer in the fridge" vs "dolaba biraYI koy" : "put THE beer in the fridge". You add +I accusative suffix to "bira" (beer). This was not even mentioned in Duolingo. It also makes sense that they don't mention this sicne this is probably super advanced nuanced speaking, but then be consistent and don't teach that "bir" distinction to beginners as well.

Funnily enough, the most ancient Indo-European languages have an incredibly complicated case system.

Having only learned Indo-European languages before, when to use "bir" was hard to get right in Turkish for me. But our teacher insisted.

That's not to excuse Duolingo obviously.

>You add +I accusative suffix to [imply definiteness]

Very interesting! In Russian, the accusative suffix would play exactly the opposite role. While there's no real way to say the beer without additional context, you can say:

поставь пиво в холодильник / put the beer in the fridge

поставь пива в холодильник / put SOME beer into the fridge

I think "пива" is the partitive case, though.

That's exactly it! [1]

I'm a native speaker, and I wasn't even aware of partitive case. They never taught us about it, even though a mind-numbing amount of time was spent cramming the name of the usual six cases into our heads.

I only got more interested in linguistics after moving to the US, and, sadly, only took one class in college on it.

Thanks for noticing this, I just learned something!

As a side note, I also mixed up genitive and accusative cases.

[1]https://en.wikipedia.org/wiki/Partitive_case#Russian

I think the definitive and most authoritative review of the Russian grammatical cases was done by Zaliznyak: http://inslav.ru/images/stories/pdf/2002_Zalizniak_RIS_i_sta..., in Russian, pages 36-55. Fascinating reading.

пива (as partitive gen.) here sounds really weird to me, like you're pouring it out of a can into something else, and putting that in the fridge. If it's a can you can emphasise it with a number e.g. я поставил одну банку пива в холодильник :)

not a native speaker, just living and studying russian in moscow

The partitive here is to express that the speaker is not specifying in what form or shape the beer in the fridge should appear. For a native speaker it sounds like you are talking about some quantity of beer but omitted the quantity so it's literally just some unspecified quantity of beer.

So you would say that though? Like I've just never heard my хозяйка квартиры say that is all. I hear it often in things like "would you like some tea/beer?" etc or "i'm going to buy some tea/beer" but never like that. My impression was that outside of those contexts it's not popular. Mind sharing some examples? Would be really useful! :)

You probably would never hear somebody saying "Поставь пива в холодильник." since, while a correct sentence, it would need some rare circumstances to be said. Поставь here is much more concrete than "put", it's more like "place" and I could imagine saying this only if I had been managing the fridge at some party and noticed it's running out of beer so I requested somebody to add more beer there. Much more common would be "Принеси/достань/возьми пива из холодильника" (get some beer from the fridge).

>You probably would never hear somebody saying "Поставь пива в холодильник." since, while a correct sentence, it would need some rare circumstances to be said. [...] I could imagine saying this only if I had been managing the fridge at some party and noticed it's running out of beer

Yup, pretty much that. Or like that one time when we went binge-shopping on craft beers, but there was only enough space in the fridge for a couple of bottles; so that was a normal request about an hour before we were to watch a movie. Which is pretty much the exact situation you described :)

It was noted below that this is properly called the partitive case; the Wikipedia articles have good examples in the vein of "А не испить ли нам чаю":[1][2]

[1]https://en.wikipedia.org/wiki/Partitive_case#Russian

[2]https://ru.wikipedia.org/wiki/%D0%9F%D0%B0%D1%80%D1%82%D0%B8...

Articles are more of a regional feature, that is restricted to just a subset of indoeuropean languages. Articles work quite differently in Romance languages than in Germanic languages and even there are outliers like Icelandic which does not use indefinite articles.

In my opinion you could go even further an say that they have an English bias.

It's not even just "indo-european" bias, it's just that word-for-word translation of sentences without any context only gets you so far. Even for something as "simple" as English-to-French it breaks down when the languages don't map 1:1 with each other.

For instance there's no single way to translate "I ran" in French, it could be "j'ai courru" or "je courrais" depending on context and depending no the lesson Duolingo will favor one or the other, leading English speakers to write puzzled comments like "I thought the imperfect was 'I used to run' or 'I was running', why is it 'I ran' here?" And the answer is "it's more complicated than that, but Duolingo won't teach you that".

Which is fine. It is about being understood. You are never going to get fluent from a course. Language changes, and even native speakers are learning new things.

I disagree. The nuances that change are on a different level. I have learned a number of languages from books and this is taught pretty easily. If you can't conjugate a verb, you don't know much about that language.

That's besides the point. My point is "definiteness" is not a concrete concept in Turkish and if you're academically analyzing a sentence you'd have to use other types of information (accusative case, the word 'bir', context, etc...) to judge if a noun is definite or indefinite (or maybe something in the middle?). But Duolingo teaches as if this distinction is as strict in Turkish as it is in English. In English "a bear" and "the bear" are different things even without context whereas in Turkish "ayi" [bear] and "bir ayi" [one bear] mean more or less the same thing outside of a sentence and I'd just use "bear", but in some contexts they'll mean different things. "Ayi cikabilir" roughly means "caution, bear in area" and as you see you don't use "bir" but it is still indefinite. Duolingo would reject this answer. "Bir ayi cikabilir" would be Duolingo's answer, but it sounds awkward, no one would say that.

And there are no hard rules. In that sentence you can't just put +I to make it definite as I described above "ayiyi cikabilir" << this sentence doesn't mean anything. I think this is something your brain has to pick up implicitly while you're speaking/listening, I'm not sure if someone can discretely provide you a set of rules that always works. Just like in German you cannot translate "the" > "der". This is like Duolingo saying 'the'='der' and rejecting 'das' and 'die'.

So, it is not fine, it is wrong. And the wrong thing is not that Duolingo has a bug, it is that it has made-up grammar rules, which users still have to learn. If they never learned this rule, they'd be as well understood.

(I know you were replying to commenter discussing French, but I'm guessing something similar is going on)

Sure but my point was that Duolingo basing all of its courses on translating from one language to the other seriously limits its ability to actually teach you a language. My native language isn't English but as I'm writing these words I'm not translating, I think in English directly. If I switch to "translating" mode I'm a lot slower and I end up with unidiomatic English.

Any decent language learning course should try to immerse you in the target language as soon as possible, forcing you to actually think in the language instead of your own. There are many ways to do this, for instance making you answer questions. Instead of telling you "translate 'the cat is black' in French" they'd show you a picture of a black cat and ask you "Quelle est la couleur du chat?" and you'd have to answer "Le chat est noir". No English involved, like in real life. The problem of course is that such an approach is hard to correct by a dumb algorithm, especially as the concepts being taught become harder.

I really don't understand that-- compared to the cost of developing and marketing the app itself, surely the cost to hire someone whose only qualification is that they be bilingual is trivial, and you would only need them for long enough for them to run through all the questions to double check they are correct.

That‘s probably what they thought, too. If you just pick a random bilingual person and let them proofread your learning material you are going to end up with crappy learning materials.

If you want good content, you need good editors, and they are neither cheap, nor easy to find.

The reason why companies try to save money with editors is that they don‘t „scale“ (one dev and a designer can program an interface that can be used for 20.000 different questions, but you need a lot of people to come up with those questions)

It's not hard to hire someone to notice that the comments are reporting errors.

Not only good editors, but good editors who are also bilingual.

I'm not sure they've even gone that far. There are _plenty_ of long-standing complaints (on the Japanese course) that there are errors in the _English_ sentences, such as that they are grammatically wrong, or that you've translated something one way but later don't accept that as a correct answer. Just checking the comments, or proofreading the English sentences, would have caught most of them.

Reading this thread, I'm lead to wonder if there's any open source, Wikipedia style (public editing) apps / sites / courses being developed? This would certainly combat the scaling issue, as Duolingo's comments seem to imply.

Why do you need 20,000 questions, though, when a typical user isn't going to need that many?

Engineer's disease... "why hire an expert when we're already Engineers!"

I'm Brazilian, and I was curious. So I just did the placement test (in English), completed all the questions correctly and was placed at level 11, with a few gray bubbles at the end of the tree.

I got all the questions correctly, but:

- Given capitalization I was forced to use word order that didn't come naturally. Mind you, what the app requested isn't _wrong_ but it's certainly not the only way to do it. If they were enforcing a rigid wrong order they were incorrectly marking as failures usage that I would argue is more natural, at least given my particular background (Southern). I went with it, but I was adjusting to the app.

- They repeated the sentences and some vocabulary a lot, so if you had any gaps or couldn't remember one particular word you might get three questions wrong, which would have a large impact on your placement.

- There's something inherently weird about the vocabulary, Duolingo-style. This doesn't go well with the previous point.

Overall it felt rigid and random. In any case Duolingo does not correlate at all with fluency. Duolingo correlates well with being trained in Duolingo.

I heard Portuguese from South America and from Portugal are different - maybe that could be a reason?

> I heard Portuguese from South America and from Portugal are different - maybe that could be a reason?

I came here to say the same thing. Spanish from one country is "wrong" in another one. A lot of people doesn´t understand that it´s just not your flavor of the language. It´s almost the same that If I, that speak Spanish, commented how wrong is Portuguese because is similar to Spanish but not the same.

Maybe the mistake is to announce the course as Portuguese instead of, for example, "Brazilian Portuguese".

Did you check if other language learning apps were wrong as well?

Why would this matter, at all?

It would matter to those of us trying to learn a language, because if another app was better, we could use it instead of duolingo.

At the moment I'm supplementing duolingo by nattering away to my father in French and asking the half-French guy at work how to say various things, but if the app is teaching me things that are plain wrong, I'll have to either find another app or start taking real life classes, which is why it would be useful to know what else is broken.

I've been trying a few of these lately. I passed French DELF A2 late last year and going for B1 in a week.

My chronological progression of picking up French via apps/courses (lots of trial-and-error):

1. Duolingo

2. Michel Thomas audio course - this has its critics but it gave me a much more solid, if limited, foundation on grammar.

3. Lingvist - I was lucky to find this early. Added around 2000 words to my vocabulary (yes they claimed ~5000 - that was inaccurate; conjugations and plurals are counted separately). Today you need to pay $23 / month to get the same number of words.

4. Clozemaster - it offers a lot more words, so I got a paid subscription. Got disappointed really fast though because its automated method to offer mass sentences really shows its flaws quickly: there are many errors and nonsensical cloze placement. I got tired of flagging.

5. Assimil - seems good, but without any gamification or anything I can play on my phone while waiting, commuting, etc., this soon dropped off the radar.

6. Speechling - this one should not have the problem of Clozemaster or Duolingo, because they have a human teacher correcting your speech. I imagine their sentences are more carefully curated as well. I started the trial, but dropped after a while because I found I just couldn't allocate the time to sit in a quiet place and record myself for an extended amount of time. Yes, it's a genuine "it's not you, it's me" thing. I really like their recent features and will probably try this again during a less hectic life period. Hongyu (the CEO) is super responsive too, which is great.

7. Glossika - this is what I'm currently on. It's not perfect; their main thing is you hear someone saying X in English and a moment later the equivalent in French. They offer mass repetitions as well, with the same class of problems: occasional inaccurate translations. The reason I picked this one instead of Speechling is not quality: it allows me to listen to mass sentences while doing mindless boring stuff that I have to do anyway (e.g.: doing the dishes, walking to/from the subway station, etc.). This allows me to get French exposure every day, even if not perfect, which ends up meaning more exposure than what Speechling can give me.

8. I'm a subscriber of one more app, I can't believe I forgot to add this one the first time: Kwiziq. They have a very specific target: teach you grammar from A0 (i.e.: the level below A1) to C1. My main problem is that they don't have an "aging" feature. Something I learned fully in A1 9 months ago will be forever marked as "mastered", despite my having forgotten it completely. Nevertheless I continue to be a happy subscriber. It's fun to finish a quick grammar quiz here and there as you're waiting for food, queueing for stuff, etc.

I did the Michel Thomas Spanish course and I found it to be excellent. It can be a lot harder to stay engaged than a gamified app, but you come away knowing a lot more than you would expect.

Yes, plus I can put it on while doing the chores, which is ideal, same as Glossika as my previous post. It's naturally there embedded into my daily workflow.

Everything else e.g.: Assimil is a workflow interruption, a new habit I have to assimilate, so it becomes a lot harder to do consistently over a long period of time.

Thank you for mentioning speechling. I'm going to give it a try.

Hey!

CEO of Speechling here. We'll work on (7). It's trivial to implement. We want the functionality of (7) to be better and free.

Hongyu

Besides what the sibling comment said, if all apps are generally not great, it speaks to a basic limitation of app-driven study of languages, at least, that indicates one needs to supplement. My plan was to use duolingo to get to the point that I could start reading French newspapers and watching French TV, which I assumed would add a dimension to my learning experience that I couldn't get from DuoLingo.

After this thread, I plan to move that step up as a check on DuoLingo.

Reality check: after finishing the Duolingo French tree I tried to read Tintin. It was quite a wake up call, to say the least.

Since then by my estimate I know roughly about 4000 more French words, and reading lemonde.fr is still a struggle. Doing much better with Tintin though.

Tintin is a fantastic way to learn french. Just wanted to stop by to say I found it super helpful as well :) I remember the first book I finished in french was tintin in russia

Reply on news.ycombinator.com