galaxyrocker wrote:Personally, I think DL's pedagogy just fails in general. You're not really practicing anything truly language-related. You're asked to translate a sentence, usually from TL into NL, and that's it. Given whether it's right or wrong, a correct translation, and then you move on. On the app, it's worse, since, as cainntear said, there's a lot more of just picking the correct words...which are often obvious given capitalization patterns, etc. But, you don't really get any true reading practice, or speaking, or writing (apart from the really rare NL -> TL question).
I think there's a lot that can be achieved with well thought-out translation exercises -- the problem with Duolingo is that the exercises aren't well thought out. Comparing the translation in Duolingo with Michel Thomas, there's a clear difference that you don't often think "what the hell is that supposed to mean???" when doing MT. "I want it, but I don't have it" may not have an unambiguous independent meaning, but you immediately get it as a sentence -- you can imagine saying it. Duolingo, on the other hand gives you things you could never imagine saying, whether it's telling someone "you are a girl" or "I am a man/woman" when you're the other one, or saying that "I have X and Y" for weird combinations of X and Y. OK, so there are circumstances where you might say "you are a girl", and if you think about it you could imagine telling that to a little girl, but you shouldn't have to think about it -- the prompt should immediately have a sensible meaning, otherwise you're thinking about the words and rules, rather than the meaning.
The algorithm also has a tendency to switch arbitrarily between synonyms and alternative forms. For example, I just did the first lesson of Dutch for the first time, and the prompt for "she is a girl" used "ze" while "she is a woman" used "zij". Similarly, I've seen random inconsistency with the alternation between "Führer" and "Leiter" for "leader" in the German course.
Then there's also the fact that the prompts don't respect common orderings of co-ocurring words in languages. When I get asked to translate "a woman and a man" or "butter and bread", I know that this is something I would never say in English (the phrases are "a man and a woman" and "bread and butter" in English) so I'm left unsure as to whether the target language version is equally wrong, or actually the normal ordering in the language. While there is an argument for challenging the inherent sexism in the way we put men first, that is
not the responsibility of a language course (teach the language as it is spoken, not the language you want it to be) the fact that it happens with things like "bread and butter"/"butter and bread" shows that this isn't even an ideological decision anyway -- just poor course design.
The most serious example of non-language in Duolingo that I can remember is "El menino da seu tigre" (or similar). The discussion of the sentence is full of people giving contorted explanations of where the sentence may be used, but the English translation is "the boy gives his tiger", and from the number of people who comment "the boy gives his tiger
what?" it is clear that the default interpretation of the incomplete, ambiguous sentence is the wrong one for this case (the missing information in the Portuguese was
who he gives it
to).
Of course, you do need to start doing more stuff after translations, but in principle, I don't see a problem with a course the length of Duolingo being based primarily on translation, if done right.
You also don't, in my opinion, get any good listening practice (note: mainly for courses with robotic voice, or which don't have complete audio, like Irish). Sure, the robotic voice might help when you speak with others who have a bad accent, but first you need to be able to understand the native speakers. Anecdotally, I've had plenty of people tell me they understood Dl's audio just fine (Spanish), yet couldn't understand a single thing a native speaker said to them. I think it hinders you a lot in that regard. Plus, there's no questions related to actually using the skills, apart from translation/transcribing. No reading comprehension, no listening comprehension, no writing and no speaking.
For one, the Irish audio is terrible, recorded by a non-native and really not worth listening to. (Unless they rerecorded it after all the complaints.)
On a deeper level, there is an argument that text-to-speech might be
better for low-level learners than any other type of audio input. One of the lesser discussed concepts in language learning is that of "salient features" -- the features of language that stick out and are most obvious to native speakers. As non-natives, it is very difficult to reprogram our ideas of what is "salient", so we often look in the wrong place out of habit, and phonology is an area which is very often affected by this.
In phonology, one of the salient features is the "primary distinction" between two related phonemes.
For example, English has a primary distinction of voicing between B and P, G and K, and D and T. (There is a secondary distinction of aspiration.)
Some other languages will have a primary distinction of aspiration. (Possibly accompanied by a secondary distinction of voicing.)
Where languages have two distinctions, it's very easy to check while is primary (high salience) and secondary (low salience). If you take the voicing from your D and add the aspiration from your T, what does a native hear? If they hear T, then aspiration is the primary distinction; if D, voicing. similarly, you can take the unaspirated characteristic of D and add the unvoiced characteristic of T and do the same test.
Secondary characteristics are unreliable, and sometimes disappear in relaxed speech, so non-native speaker who is working from the salience model of their own language will struggle with natural speech in the L2.
There have been studies (no references to hand, sorry) suggesting that exaggerated salience is effective in helping learners notice the salient features of a new language. As was discussed recently in a thread on exaggerated teacher pronunciation, the problem with teacher pronunciations is that they mispronounce rather than exaggerate salient features. However, I recall years ago reading that a principle of voice synthesis is that salient features should be subtly exaggerated in the voice model, as this makes them easier to understand, particularly when delivered in a noisy environment (phone lines, public announcements, on your mobile phone). As voice technology has improved, I'm sure the salient features are exaggerated less, but they are still very clear, and I think that makes it an excellent pronunciation model for beginner learners.
Where it doesn't work so well is when you have a fully grown man saying "je suis une fille". Sounds very, very creepy.
Really, though, this all boils down to what I think DL's main mistake is: they're a for-profit organization trying to offer the most 'effective', 'best', 'language learning' for free. So, they've got to find a way to make money. Their original scheme failed, in part due to the fact that EU laws excluded all of Europe from legally working on the translations; in part because who would buy crowd-sourced translations made by learners who aren't even fluent? Now, they're trying to keep a huge userbase, thus dumbing down the app where the majority of their users are, and still try to find a way to make money. It'll be interesting to see how it goes for them; looks like it's already turned into pretty much an ad-based model, where you can subscribe to remove ads. Now we just need to see what actually helpful features they remove next to keep their revenue high!
They're doing ads now? I hadn't seen any (but I hardly ever log in).