Luis Von Ahn AMA answers a few of our questions
People have been angry since Immersion was removed and questioned what Duolingo is even doing, but after a quick read through this: https://www.reddit.com/r/duolingo/comments/5pr3sv/i_am_luis_von_ahn_ceo_and_cofounder_of_duolingo/ I saw that the Duolingo team is doing a lot.
Here are some things I noted from Luis' responses:
- Japanese and Mandarin should be in (or out of if they work fast I guess :P) the incubator this year
- Adding new languages takes a lot of resources on their end and the process isn't very organized, so they have to priortize. That's why Latin, Finnish, Japanese, Mandarin, etc. haven't been introduced
- Luis plans on keeping Duolingo and its education free
- Luis does not think a Castilian course should be made as it's decently similar to Spanish (a dialect actually)
- Only .05% of the people on here used Immersion (still less than 1% for those with access to it) and since it was making things slow Luis decided to cut it
- Lingots will be used for more and based off of a picture Luis posted it seems as if we'll have customizable avatars this year. I'm not sure if it's just for mobile though. http://imgur.com/a/IhBQ0
- There is less out for Android as there are more iOS developers than Android ones in the Duolingo team
- Luis plans on adding a lot this year
- Luis also plans on rewriting Duolingo from the bottom up so it's faster.
- They plan on making things more standardized; e.g. Tinycards/Clubs on website, etc.
- Luis definitely wants to bring Tips & Notes to the app
- They'll be working on 4-5 smaller languages per year, but he said since there's over 6,000 languages, they can't do a whole lot.
- Swahili should come out this year, doesn't know anything about other African languages. So sorry people who want Afrikaans :/
- Arabic should be in the incubator this year
- Trying to improve TTS doesn't result in good metrics
- Luis really wants to add more sentences where you translate from English to [blank]
- They're working on adding more exercises that teach listening and reading comprehension of longer passages.
EDIT: I might've messed up a few words as I couldn't remember exactly what he said; e.g. '...since it was making things slow Luis..'
EDIT 2: Added a few bullet points. Change some things (not much)
EDIT 3: Said 'Catalan' instead of 'Castilian'
EDIT 4: Small bit about Castilian
Your post really needs another edit. Castillian -is- a dialect of Spanish, which is why they will not be including a course. As Gerardd, and Dr. von Ahn, and many others have said, it would be like having a separate course for British English.
Also, it's not improving TTS that causes large metric drops, it was trying to replace it entirely with human voices, for a variety of reasons.
Edit: A dialect is distinguished by its vocabulary, grammar, and pronunciation (phonology, including prosody). Where a distinction can be made only in terms of pronunciation (including prosody, or just prosody itself), the term accent may be preferred over dialect. Other types of speech varieties include jargons, which are characterized by differences in lexicon (vocabulary); slang; patois; pidgins; and argots.
Castillian is classified as a a dialect of Spanish -a dialect of Spanish , a variant, family; category: Spanish(es), by linguists. Wether you like it or not. I really think, like CHarrell that you don't understand the definition of "dialect", even if you paste it.
American and English English are not 2 dialects. If Castillian is even more similar to Spanish than American English is similar to English from UK, so Castillain doesn't exist, they are the same. It's not about what you see, it's more about how linguistics defined the words.
Is there any plans to change the robotic voice used on courses to a real spoken voice, this is one of my biggest issues with Duolingo.
We've tried doing this and it reduces usage metrics :( That is, people stick with Duolingo for less time when we present real human voices rather than our synthesized speech. Not sure why this happened, but given that it hurt and that having real human voices requires quite a bit of effort, we stopped working on it.
That's really bad to hear. The robotic voice in the courses is one of my biggest issues with Duolingo, too — human audio is incomparably better...
While I agree about the horror of anyone sounding like the English voice the solution using human voices is worse as not having audio for the single words is a big setback. The Hebrew course is useless without using the memrise course which has audio for all the words along side it. The Irish course was incredibly difficult because of the lack of audio too. I'll opt for the tts everytime. The benefit of the slow speed for people new to language learning shouldn't be underestimated either. I no longer use it but at first I really really needed it. I think duolingo actually teaches listening skills really well, and the tts is why.
There would be a solution to that I believe, but only for mobile devices: Let us use Duolingo TTS with our own voices. I bought several voices in several languages on Google play for only 2 or 3€, and they are perfect. If we could use them with Duolingo app, Duolingo wouldn't need to add so much resources.
I actually don't understand why they just didn't record the words themselves. The voice recording is challenging for sentences, but for words - it's much easier. It's challenging for sentences, because if you add any new sentences they would have to get a new recording. But we cannot add new words, so you record them only once and you're done. And there are not that many words in the course, let's take Hebrew for English speakers, 2833 words (actually less than that as some are repeated) and at least thrice as many sentences as you have to have at least 3 sentences per word.
And at the end of the day, why not have a robot to pronounce separate words. Actually you don't even need that, you can buy libraries of separate words pre-recorded by natives, they are used in many dictionaries, that's often the case even for languages for which there are no good robotic voices. Because again, getting records of separate words is much easier than of sentences.
Those avatars are going to look really cool! I'm glad Duolingo seems to be focusing on expanding, even if it did mean cutting immersion completely away from the site. Looking forward to Arabic, Mandarin and Japanese for sure! I hope to see them in the incubator soon!
True, but if we need to pay lingots for them, that will not only boost value of an otherwise worthless currency, it could give people who like customizing things motivation, which will lead to more use of Duolingo. It's pretty much a marketing strategy to make everything more appealing (it works well, too). I do agree on your point of hoping it doesn't become a focus on the developer's end, since this is a site for learning.
Frankly, I don't get the vehemence against that idea. The sound systems are incredibly different. (In Spanish and Portuguese the sounds and the most basic verb conjugations are different.) Kids in school might well be learning British English, and they come here and discover something entirely different. Switching accents isn't easy. You have to literally relearn how to pronounce almost every word.
I don't think I get your point. You think it's a good idea to make a British English Duolingo course from American English to learn the accent (probably with some TTS) and a few spelling/semantic subtelties? Of course the differences between dialects are significant and important but there are some good ways to learn them (or about them) and making a sentence-translating course isn't one of them.
Well, if I wanted to be cheeky, I could cite the publicity generated when Harry Potter was "translated" into American ;)
If people think Klingon and High Valyrian are are all about publicity...
But, no I just meant a TTS option — perils of not mentioning the base language one has in mind ;)
I don't see the point. They tell us that they have great problems to add more audio. Like a lot of course doesn't have all the audio (Vietnamese: 55% only, and they are not allowed to have more), so there's a problem, I don't think it's the right time to imagine a world where we would add not crucial and proritary choices like having the accents. It would be a great idea, very intesting, but it's clearly not a priority, and will probably never be, regarding the lack of crucial audio in a lot of courses, and the difficulty they have to fix this problem.
I know some Portuguese speakers would disagree with that. But the bigger issue is "Who cares?" If I'm in Argentina and I have speaking tests to pass for class and we're learning British English (which they do), why should Duo, which after all is trying to get teachers to have their students use it, be so obstinately throwing a wrench in the works?
It would be nice if British English was available, but again, Duolingo doesn't have infinite resources, and Duolingo probably wants to teach more languages rather than dialects or variants. It also seems like almost all people are more eager to learn more different languages than any dialect or accent, from the course requests.
EDIT: I didn't realize that you were only suggesting an option for a course. That's definitely a viable idea.
No, people who are learning English and, in common with, I think, the majority of people learning English as a second language worldwide, at least historically, want to learn British English. I'm American, by the way, and am thrilled that Duolingo has brought my native dialect to the masses of language learners around the world, but I think people should have a choice.
NB: the confusion could be about the word "course"; I'm just calling for an option akin to doing the Russian or Ukrainian courses in transliteration: an option for how to do a single tree, not a separate tree
@Piguy As a non English native, I feel I can try to talk for my country, and maybe for the whole Europe too. People who don't speak English or have a poor English, doesn't care about learning American English or British English, they only want to learn English... any English... You are talking like someone who is in contact with English speakers, in countries where English is not the language or in the culture, people only wants to talk good English to get job positions or to be able to communicate, to travel, etc.. They don't care about grammar, spelling and even more about accents. Among people I know, some talk with a British accent (or rather: they try), some with an American accent (they try) and some with a mix of both. The fact you choose British or American depends on the nationality of the teacher you had at school, and the affinity you have for British or American English, or the affinity for the culture. A lot of people I know prefer British English because they are used to at school, and because they think it's easier for foreign ears to get accomodate to and to understand what they mean. Learning the difference between American and British English is a thing of very advanced users, everybody else don't care. Sorry English speaking people, but this is the reality.
Any insights I have on this matter come from my interactions with very-monolingual Argentines attempting to learn English. In school they learn British English (well, as you say, they try). If they came to Duolingo, they would come upon a different phonetic system entirely. I simply suppose this can't make things any easier for them. Indeed, I don't suppose they care a whole lot a priori about which variety of English they learn, and the dialect distinctions are an extremely advanced matter (find an American who actually knows the/a British sound system or vice versa: not likely!) That's why I suppose that forcing them to change the accent they're learning in (or having to learn two at once) is an unnecessary difficulty. Probably not a tip-top difficulty, but a difficulty and one that could be fairly easily ameliorated (I don't think any comparison to the difficulties of getting recorded speech are relevant; Duolingo clearly has to expend less resources for TTS than for recorded audio, and I am assuming for present purposes that there is a passable British TTS accessible out there).
In my opinion, "forcing" people to hear a variety of English accents is good, even as beginners. I've been an English learner once with 0 knowledge in English, and it's not so difficult to adapt between UK and American accent, it is even beneficial I think to a better understanding of oral English, and how syllabes are pronounced in English. We had this case at junior high school, switching as low level learners between British teachers and American teachers. A bit surprising in the beginning, but good for our ears.
They already have a lot of trouble to give us sounds for hard Asian languages like Vietnamese, so why they would provide different pronounciations? Starting from there, everybody would ask to record his own regional pronounciation, Canadian French, Belgian French, Haitian English or whatever. It's very risky, and is not so different enough to be a priority comparing to adding very needed audios.
I think some of the earlier recorded audio courses were done half heartedly or with unduly limited budgets or something. Hungarian and Guaraní both have recorded audio. It's excellent and missing for only the tiniest number of sentences. I'd hope they'd be looking into dealing with omissions in Vietnamese just as they eventually scrapped a dramatically sub-par Irish TTS in favor of recorded audio.
But English is far and away the most learned language on Duolingo. There'd be no expectation that what is done for English would be done for every other language (there's plenty of features I'm barely aware of: the Words list, matching exercises on the app for example b/c my primary language here, Russian, wasn't one of the first handful). However, it's not some completely unrealistic idea that the big, commonly learned pluricentric languages should have that pluricentricity reflected on Duolingo in some way. Globally, no one really learns Canadian French outside Canada that I'm aware of (but if there's a decent Canadian French TTS engine out there, far be it from to say they shouldn't just provide the option). But plenty of people learn European Spanish and Portuguese. And they would benefit from a relevant TTS option if it exists.
The whole inquiry was
When will the Spanish for English tree get updated? Especially when the English for Spanish speakers has more useful vocabulary than the Spanish for English speakers. If it does, can we get Castilian Spanish tree?
Luis's reply was "blunt." I wonder what he would have said had it merely been a request for different pronunciation options. I think it still would have been "no"; there's at least one Portuguese Portuguese former contributor out there who departed in something of a huff over unwillingness to accommodate his dialect even though Duo central swore up and down they valued his particular contribution, but I doubt Luis would have started his reply quite the same way.
I'm sure he is trying to improve Duolingo, I'm not sure it's the good way. When I saw clubs instead of immersion... Adding more language is fantastic, I was expecting a lot of them for a long time, but what is planned to make the user use the sentences and the word he learned in a creative way, i.e making his own sentences? And to add advanced levels to courses?
After reading this, I'm only more confused about the reasons for axing the Immersion feature.
At first, I thought it was because so many people were using it that it that the costs for maintaining it became too high, but now it seems like not enough people were using it. Considering that Immersion used to be DuoLingo prime selling point, having 0.05% of your users using it bring financial ruin to your company seems odd.
Then again, I'm just bitter because I miss this feature too much. Right now I just hope something cool will replace it so that we can still practice reading skills.
He justified it in part based on their porting the entire site do a different software architecture to very significantly increase access speeds. Having to do all that coding work for the Immersion system would have added substantially to the costs in doing so. If Immersion had had a bigger user base, undoubtedly it wouldn't have gotten the axe. User engagement is web companies' currency. If Immersion had taken off, they never would have stopped developing it.
Unfortunately, people didn't realize what an amazing tool it was: I think even better for the languages that didn't have it than those that did. Finishing a Romance language tree, heading to Readlang and reading authentic texts is at least somewhat possible. For a Russian, a Hungarian, or a Hebrew more handholding is definitely needed. But I doubt the number of people who manage to finish such trees is enough to make much of a market :( Really, the problem is that I don't think the number of people who even make it to the second checkpoint of any tree at all constitute much of a market. Hence the introduction of features that are thoroughly dubious from the perspective of the experienced language learners that haunt the forums but that might increase retention among new sign-ups by a few percentage points.
Justified or not, still a bad choice. To find another tool that let us learn deeper and with creativity all the subtilities of the language, I wish him good luck for that.
Except if they make a super reading comprehension tool (but the make-your-own-sentences part will be still missing).
I read they tested a lot of things on random users the last 2 years, and nothing was working, because they abandon everything except the clubs chatbots things, so I don't think they will arrive with a super tool that is not even created, if it happens, before 2 or 3 years.
I feel they don't tell us the truth. It's not about server ressources or the popularity of the tool, it's about the fact that immersion brough no money, and this site needs money to go on.
Obviously I, too, lament the departure of Immersion. I think one of Duo's big efforts is figuring out how to increase translations into target language while still maintaining users. I don't blame Duo for finding this a difficult balance. I blame the lackadaisical users who leave if they're challenged.
Audio is one of the big problems for small and endangered languages (scripts is the other one). Usually we cannot find a TTS engine for them (i.e. computer audio), so we have to manually record the voices, which requires quite a lot of effort.
We'll be working on 4-5 smaller languages this year, but there are over 6,000 languages in the world, so we won't have the resources to make a big dent any time soon.