Hi Hacker News! My name's Mike. I'm the creator of Clozemaster.
Clozemaster is language learning through mass exposure to vocabulary in context. The goal is to fill in the missing word in a given sentence. The missing word is the most difficult word in the sentence according a frequency list for that language, and the sentences are from the awesome dataset at Tatoeba.
I started the site just under a year ago to answer the question "what should I do after finishing Duolingo?". Since then it's grown to support over 50 languages, mobile apps, and thousands of users. It's useful for learners at almost any level, from beginner to advanced, and makes a great complement to Duolingo, textbooks, classrooms, etc. to practice vocab in context.
You can play the Fluency Fast Track which gives you a sentence for each unique missing word in order of increasing difficulty, jump in to sentences grouped by frequency from the 100 most common words to the 10,000 most common, or just play random sentences. There's also "cloze-listening" - hear the sentence first, then see it and fill in the missing word.
Thought Hacker New might find Clozemaster interesting and hopefully useful! It's still very much a work in progress - I have a bunch more features planned and I'm working to improve it all the time. I'm also open to any feedback and happy to answer any questions!
This was really hard for me to understand. When I skimmed this, my first assumption was that this was a direct replacement for Duolingo, when it's actually complementary.
It would probably be better to split up the long list of languages on to 2 different pages. 1. Language you want to learn, and 2. From which language. An alternative would be to have a separate section for each language to learn on the page with space in between.
I noticed something weird when doing Japanese. When you answer a question, it says the correct sentence out loud in one voice, but sometimes the voice says something different than what is written. Then when I click the play button, it says the sentence correctly but in a totally different voice.
At first I thought I was just crazy, but this has happened at least 3 times in 3 "rounds" so far.
Huh, after playing it a bit more, I've noticed that sometimes the hiragana it shows for the kanji in the sentence does not match the way either voice pronounces it.
This issue is most likely the result of the hiragana being automatically generated and some kanji having different pronunciations based on context. I tend to trust the voices, and look up conflicting words in jisho.org to be sure. Ideally in the future I'll be able to get some kind of human moderation in place for checking the hiragana/pronunciations.
Several years ago I started something similar. The idea was to create cloze test flashcards. That never really took off, but some teachers liked the general idea so I rewrote it to become a cloze quiz creator tool which is now http://www.learnclick.com (it now offers much more than cloze quizzes). Congratulations Mike for coming up with something for the individual learner. I think having premade cloze sentences was what was missing with my original solution.
1) Cantonese doesn't have a flag. Maybe just use the Hong Kong one, at the risk of upsetting the rest of Guangdong?
2) For hanzi-based languages, it would be super awesome if you could (optionally) display the romanizations as ruby annotations on top of the characters. I know romanizations are there, because they are displayed after getting an answer right, but it would be so helpful if they were available immediately.
The multiple choice mode is missing a button for "i don't know". I figured out you can hit enter to get the same, but i don't think everyone would.
E: Gave it a bit more of a try, and i won't be switching this one. The japanese set doesn't seem to particularly try to avoid throwing unknown words at me when giving me new ones and the voice is uncomfortably high-pitched.
I am a regular Duolingo learner but I have yet to see how this is better in a significant way (I am not trying to be a dick but just giving honest opinion - sorry :( ). Or maybe I just missed something.
I played around with your product and I've found that I need to enter words or select missing ones like in Duolingo. It seems a bit more difficult than Duolingo though.
The biggest problem to me with Duolingo that it doesn't explain certain language rules (I am learning German) and you have to brute-force until you "get a feeling" and memorize patterns.
Another problem, is that there is very little focus on developing your speaking skills which (to me at least) are the most important.
I would really love to see some emphasis on speaking skill development.
Even more, Duolingo sucks at text understanding development - you get to learn sentence by sentence, but you are never getting a "full picture". So when you get a proper, full text (some short article, for example) in another language, it just feels a bit overwhelming.
Basically, what I am trying to convey and what's missing (for me) in this kind of apps, is that learning language is not about filling in missing words in the sentences.
Btw, I really like that you charge $8 - IMO Duolingo here misses out a huge chunk of money. I would easily pay $5 and probably up to $10 but now I have just to "endure" one ad every 5min which doesn't bother me at all.
Language learning also isn't about finding the one app to rule them all. It's about using various different approaches in parallel that, together, expose you to the breadth.
This doesn't need to be better than Duolingo, it just needs to work a different part of the brain.
One problem with Duolingo is that its fill-in-the-blank always comes with a word bank which makes them all pretty trivial. You don't need much understanding to choose between a bank of "girl", "cloud", "yesterday", and a single verb conjugation.
Open-ended cloze tests, on the other hand, actually force me to think.
I think it's outside of Clozemaster's scope to care about speaking. Didn't Duolingo remove speaking tests from their app recently because everyone toggled them off?
True. Learning a language just from a single app is a lost cause (at least for now and, IMO, for foreseeable future).
>Open-ended cloze tests, on the other hand, actually force me to think.
In this regard, the Duolingo web-app fares much better, because all the answers you have to type in. After you get used to that, using mobile app feels like cheating :).
>Didn't Duolingo remove speaking tests from their app recently because everyone toggled them off?
They are still there. Web app has even a version of that on steroids - they give you a sentence in English and you have to vocally translate it.
> The biggest problem to me with Duolingo that it doesn't explain certain language rules (I am learning German) and you have to brute-force until you "get a feeling" and memorize patterns.
I'm not trying to be contrarian but, in my opinion that's a strong point, not a weakness.
I work with language learners a lot and many of them spend a long time looking at explanations of foreign languages in their own language. This basically never works. The people who really progress dive in and let their brain soak up the new rules and patterns. In a metaphor: reading grammar explanations is like try to learn to play the piano by reading about on how a piano works. You might learn a lot, and even work out how to play simple tunes but your fundamentally practicing the wrong thing.
I finished the Duolingo French tree. I now use Clozemaster for French ( and a number of other languages ) and find that it is a very efficient and entertaining way of studying French in context. I can now slowly and not too painfully, read some easy authentic French material, including parts of Le Monde and the French version of ZDNet. I have of course, used a bunch of other resources, including the excellent Assimil.
Thanks for the feedback! always_good's response is really on point. Clozemaster isn't intended to be better than Duolingo, rather it's meant to be a complement and another tool to add to your language learning toolbox. Speaking is also beyond the scope of Clozemaster at the moment. I'd totally recommend Clozemaster to help get more exposure, expand your vocab, and improve your comprehension, but for speaking - something like italki is probably your best option. Regarding learning sentence by sentence vs. getting a "full picture" - I would like to try to expand Clozemaster to include short texts at some point with a missing word per sentence. Hopefully down the line!
Your problem learning German probably stems from "Deutsche Methode", i.e. throwing you without any preparation into talking in a strange language without explaining anything, not even meaning of words very often. For some reason this is extremely popular and distinguished method in Germany, maybe stemming from German masochism formulated in "if it doesn't hurt, it doesn't count", which you see around here everywhere. More pragmatically it probably brings Goethe Institut much more money than making the language more accessible (which is certainly doable).
Well, there's also another thing: according to empirical research, "explaining things" about language doesn't make you better at that language. The part of your brain that handles language is simply incompatible with that kind of input data. (The input data that makes your brain better at language is comprehensible linguistic input embedded in communicative context. That is, language that is trying to communicate meaning to you, and is simple enough for you to get the rough understanding of that meaning.)
> The biggest problem to me with Duolingo that it doesn't explain certain language rules (I am learning German) and you have to brute-force until you "get a feeling" and memorize patterns.
I think gaining a feel for a language and memorizing patterns is more useful at the beginning of your journey. I am learning Spanish and I find it works better for me when I 'feel' something is right and then finally understand the reason why it is so.
It doesn't seem, however, to be sufficiently flexible in the various ways to fill in the blank (at least in Russian). For example, when given the sentence
Как вы на это ___
with the translation "how did you see that", I tried "увидел", and it told me the correct answer was "смотрите". That's clearly another answer, but I'm not sure it's any more than what I gave. Or perhaps my Russian isn't as good as I thought?
I'm Russian and "Как вы на это увидел" is not correct. In fact, I don't even know what that's supposed to mean, sorry. "Как вы на это смотрите" on the other hand is perfectly normal and basically means "What do you think of that".
Thank you. Your translation of "Как вы на это смотрите" makes sense and feels natural. However, the given translation "how did you see that" doesn't seem to match. Would "как вы видил ето" work for that?
"He was sneaking from behind. How did you see that?" translates to "Он подкрадывался сзади. Как вы это увидели?".
"Он предложил это. Как вы на это смотрели?" could be literally translated as "He proposed that. How did you see (or look at) that?", but the correct translation would be "He proposed that. What did you think about that?".
There could certainly be multiple answers that make sense like you mentioned, though at the moment only one answer is accepted by default. There is however the option to add your own alternative answers, http://imgur.com/a/Vl5o4, and I'd definitely like to come up with some way to support multiple default answers in the future.
This is really, very nice. I'm getting to the end of the flash cards I've managed to pull together for Chinese (just HSK stuff, really), and this works so well.
[EDIT: I'm curious, if there is by-default reviews, is there some way to shut off the reviews for sentences I got right on the first try? There are so many sentences that I don't imagine I'd get much benefit doubling back. That being said, I'm not sure there _are_ reviews, I just saw that review projections are in the pro thing, I think?]
Yep! There are by-default reviews included with each round. Pro users can change the number of reviews per round as well as manually set sentences to 'Mastered', or ignore them so they're out of your queue entirely. I'm also working to get a 'Set all sentences with this missing word to Mastered' added for Pro users within the next few weeks.
This is great. How did you do the frequency groupings for more complex languages like German or Icelandic? Did you use a word stemmer or did you find lists that include noun declensions?
Cool, that makes sense. Are those lists publicly available?
Do you also detect word types? I.e. "Der Mann" (nominative masculine article der) vs. "Das Haus der Frau" (genitive feminine article der) [German] or "að leita að" (infinitive marker að vs. prepositional að) [Icelandic]
Most of the lists come from https://en.wiktionary.org/wiki/Wiktionary:Frequency_lists with some additional processing to try to remove names, places, etc. I still need to make the exact lists Clozemaster is using available on the site somehow.
At the moment Pro users can play sentences grouped by missing word part-of-speech for German, Spanish, French, and Italian from English. I've tried a few different approaches to determining the word types / part-of-speech of words in a given sentence, but none have been super accurate so far. There's definitely a lot more fun/interesting stuff I'd like to do with Clozemaster if I ever do find or develop a good way to detect types.
Hey I know this might be late, but using it there's one major pain point with specifically the Japanese, in that what you have to input seems to be cut off at kind of arbitrary points, that make it annoying to input.
For example I just got the sentence 昨日本を買った where you have to fill in 買っ, but the Japanese input doesn't work for just inputting 買っ, I have to type 買った and then hit backspace to delete the extra character. This happens a lot, probably in more than half of the sentences I'm getting, it seems like it really loves to end words on っ.
It'd be cool to change it so it splits words differently, or if that's not possible make it accept extra input that matches the following characters, so inputting 買った would be accepted as correct.
Beyond that I'm loving the site, and it seems like you got a good business model, I'm still considering whether to try out pro.
Substitute a word requires certain language knowledge. This should be tested somehow, so that the application knows the user level. For example, having no prior Spanish knowledge I would not do substitute a word, and i can do listen exercises which show me some common words and already teach me something. This was enjoyable... for the first 3 minutes.
After just one run and 3 minutes into the registration it already is asking for money? I have no problem paying when I see what I'm paying for. Here I had too little time to see, and already am annoyed by this. So cloze-listening is already unreachable for me, and I would not play the other game without having any language knowledge. Fail right here.
Learning a language is a task for months, you'll (the author) be very safe asking for money after several days or even weeks of trial.
Thanks for the feedback! Adding some kind of testing / assessment of a user's level in a given language is on my to-do list. That said, users can also self-assess and choose the corresponding frequency grouping they'd like to practice for most languages (and I'm working on adding support for the rest). For example, I've been studying French for a while so I like to play the 20,000 Most Common Words grouping.
The core features are free, then there are some advanced features you can pay for. I'm trying make as much free as possible, at the same time I'm bootstrapping everything and need some way to support the site. The cloze-listening is a trial that resets daily.
Awesome work! I've built similar projects like this but nothing at this scale, or this polished. I have a bunch of questions:
How do you deal with bad sentences/mistakes in the source data? Even the best dictionaries I look at often have very odd example sentences (at least in Japanese/English dictionaries). Do you have any plans to vet for things like this?
You measure word difficulty by frequency, but do you do any heuristics for sentence difficulty?
Do you have any idea if you method works? I'm not attacking you, what your site does is very similar to what I did on my own to learn a second language, but having hard data would be great.
Again, I really love the site, keep up the good work!
- bad sentences/mistakes - users can report sentences with errors. I'm notified and those sentences are then removed from their queue. Pro users can also ignore sentences, I'm thinking I may make this a free feature in the future as well.
- Difficulty is just by word frequency at the moment, what kind of heuristics do you have in mind?
- I'd definitely like to be able to measure the effectiveness of Clozemaster somehow, but I'm not sure what kind of hard data I could come up with. Perhaps in the future I can come up with some kind of test/experiment to compare traditional single word flashcards vs. Clozemaster, or test reading comprehension somehow after playing Clozemaster for a certain period of time.
Good job! Having used Anki and the mass exposure technique to learn a bit of Spanish, I can see this becoming a useful tool in any language learner's arsenal.
Watch YouTube videos in whatever language you're learning and fill in missing words. No connection to it...Just discovered it a few weeks ago and it's super fun and addictive: you get to hear lots of new music and learn some serious comprehension.
Glad to hear! Yep I'm definitely considering a one-time lifetime payment for Pro, just trying to decide how best to price it - looking to make it available within the next few months.
Very cool! One suggestion: it might be good to remove words that are identical in both languages. I had multiple times a question where the missing word was (eg) the name of a sporter, which is evidently not translated.
I think this is awesome, but I wish the phrases were categorized in some way. Also, it would be great if the phrases were common idioms or expressions, as those are the hardest to learn for non-native speakers.
Learning English from Spanish the only words I see you learn here are hurry, train, began and brake, and train appears a lot of times. The 10000 words are a better learning experience to test this device.
Clozemaster is language learning through mass exposure to vocabulary in context. The goal is to fill in the missing word in a given sentence. The missing word is the most difficult word in the sentence according a frequency list for that language, and the sentences are from the awesome dataset at Tatoeba.
I started the site just under a year ago to answer the question "what should I do after finishing Duolingo?". Since then it's grown to support over 50 languages, mobile apps, and thousands of users. It's useful for learners at almost any level, from beginner to advanced, and makes a great complement to Duolingo, textbooks, classrooms, etc. to practice vocab in context.
You can play the Fluency Fast Track which gives you a sentence for each unique missing word in order of increasing difficulty, jump in to sentences grouped by frequency from the 100 most common words to the 10,000 most common, or just play random sentences. There's also "cloze-listening" - hear the sentence first, then see it and fill in the missing word.
Thought Hacker New might find Clozemaster interesting and hopefully useful! It's still very much a work in progress - I have a bunch more features planned and I'm working to improve it all the time. I'm also open to any feedback and happy to answer any questions!