Word strength system does not work for me
Right now, all words in my vocabulary have full strength. This is terribly misleading because I'm sure I'm not that good.
I've just tried to practice "Adjectives 1" skill in Spanish. Why, I don't remember half of the words! In Vocabulary section, they are listed as practiced 3-5 days ago. Apparently I've forgotten them since then. How come words that have been learned not so long ago and not practiced for a few days still have 4 golden bars?
As I progress down the tree, it gets harder and harder to maintain real word strength. When I try "Lesson practice", I get easy stuff like "Yo como una manzana" again and again while new adjectives and verbs are left without practice. I have to look through the tree and vocabulary and ask myself if I remember this or that section.
Disclaimer: I'm not asking for workarounds like flashcards, memrise and other sites, but I want Duolingo to be good at making learners remember what's taught here.
It would be nice, though, to get some input from the others: Do you feel you can rely on the current word strength algorithm? What is your practice routine if you don't wait until your skills decay?
I reported this problem before, and I can only agree. When I encounter a new word/ do a new lesson I repeat it quite a lot before I move on and duo seems to think this guaratees that the words I have learned are stuck in my mind forever, while I just have started to memorize half of them.
I would want to be able to manually decrease the word strengths or to select words to repeat them more often. Just now I did some practice and got a word which I didn't remember. First, it came in a multiple choice so I could guess it's meaning. Then it came in a spanish-to-english sentence and I knew it from the question before. Finally it came up in a type-in-spanish question. It never came in a english-to-spanish question and I am sure I will have forgotten it by tomorrow again, yet I got it correct three times in the practice so duo will think I know it perfectly. I would love to be able to mark that word to be sorted into the "repeat soon" heap.
Additionally, I would want to be able to mark complete sentences as "I know them perfectly". I can mark single words as that, but I do not want to mark verbs as "know them perfectly" since there will be many new tenses coming that I still have to learn. But just because I don't know the future perfect of comer yet, I don't want to get sentences like "the girl eats an apple" over and over.
So yes, there is a lot to work on for the word strength system.
There's definitely something off with the algorithms. I have had some well known (!) words thrown at me 40+ times while others appear only rarely and I absolutely do not have a firm grasp of their meaning. Just got asked to translate "I like the red dresses" for the umpteenth time while I have all but forgotten half the words that could really need some practice.
This is not just a problem with adapting to different people and their speed, I consider it a bug, plain and simple.
Definitely, I cannot rely on the current word strength algorithm.
I recall having a bunch of words frequently showing up in the lesson practice, this is an example of such word: http://screenshoot.me/2p41IZ Last time practiced? 6 days ago, and 2 bars are already gone. Needless to say I already know it pretty well.
Some other words for comparsion, practiced long time ago: http://screenshoot.me/4losqD According to the algorithm behaviour (long periods between repetitions, slow decay) I should have already mastered it. In reality, though, I have no idea what some of these words mean. Example: http://screenshoot.me/rNUQGE Only 6 times seen, not much as compared to 30 in "espejo".
I always blamed the fact that It's just me who broke the algorithm by having a very long hiatus.
Although I can't say that the lesson practice is THAT easy. Sometimes I get all answers right and more than 30 seconds of time left, and sometimes I'm stuck on the first few questions, because I was unlucky to get all the long and tricky sentences with words I don't remember.
It's obviously not a solution, just my thoughts. I use Anki for SRS, and it works like a charm. Even though the intervals are evaluated automatically, they can be slighly adjusted by the user, so that the difficult words show up more frequently and it takes a lot of time and repetitions until they reach long revison times. I think the same principles were used here, but instead of prompting the user how confident he is, Duolingo evaluates it by itself, based on measurable factors, like response time or peeking. Or not?
Either way, it might be not working as expected, but due to lack of any kind of statistics I cannot tell why.
When you say "cannot rely" on the word strength algorithm, do you mean (1) cannot rely on it to give you words in practice that you need to practice, or (2) cannot trust the 4-bar ratings as being accurate?
I ask because you might disagree with the ratings, but still find that what it picks for practice sessions are still appropriate and challenging. For example, say your whole vocab is four bars, but the "weakest" word (at 3.51 bars, rounded up for display) is really something you need to practice more that the others. This is more important to us that getting the number of bars correct, and while we try to optimize both, we put more emphasis on better rankings.
An explanation for "espejo": according to our model of your vocab, you've seen this word 33 times and used it correctly 19 of those times (58% success rate), so that combined with other factors is why the system wants you to practice it sooner. On the other hand, you've never misused "barrio," so it's backing off and giving you more time between practices there, even though you've seen it fewer times total.
I find that the algorithm does not distribute word choice evenly across the words I already know. I have conjugated "manger" about a thousand times, but every once in a while I screw it up because: I'm out of practice OR I don't understand some other part of the sentence OR I get in too much of a hurry in timed practice OR the audio is garbled (I'm not criticizing the audio, it's generally pretty good, but nothing is perfect). So even though I've learned dozens of other verbs on Duolingo, they never show up, and as a consequence, I never get them wrong.
It seems like there is not enough weight put on the number of times that I've seen a word
Actually, a bit of both.
The four bar system might be misleading as it doesn't represent the easiness factor, in other words, having 2 words with 4.0 bars obviously doesn't mean they are both equally mastered, since they decay at different rates. But I'm well aware of that it represents something else, and that a graphical representation is just a siplification of what's actually happening behind it.
I guess that if I failed to translate correctly 42% of sentences with espejo, it that was either long time ago, or I made a mistake somewhere else, e.g. by translating "Tengo un espejo plano" to "I have the plain mirror". Otherwise I cannot explain why I know this word very well, and yet I apparently make mistakes with it. Hard to tell, but it isn't surprising that with a 58% success rate it shows up very often, that's just a very poor result.
Overall, the practice is challenging for me, but I remember seeing the same, easy sentences, more often than I'd expect it, while others show up very rarely. That's why I said that I cannot rely on the world strength algorithm. Maybe it's just my subjective feeling.
Tatou, I apologize for bothering you with something that might affect only me but if you find the time can you please check if there is some kind of bug here. I keep seeing these two words in my practice even though [in my opinion :) ] I know them perfectly :
I went to my vocabulary, found the words I was look for and then I clicked the link in the first column.
Within a few discussions Tatou( a staff member), hinted as to why some of this happens. It seems that aside from words you get wrong, the system also decays/shows words other users often get wrong. However, one thing they probably lack, is the user's data, in terms of what their native language, and perhaps age.
For example, words that are similar in two languages you are a bit more likely to remember, especially if they are spelt in the same way. But I think the biggest issue, is that practice is optional.
In my opinion, the fact that it is optional presents a problem, since the algorithm can't choose the best time to practice, nor can it evaluate the benefits for practicing a certain number of times in a day, or which words to show you often, since you forget them often.
I think that one way to improve the word strength reliability, would be for individuals to stop peeking, never skipping a question, and practicing the basics every day, or the skill you feel you often get wrong.
Same happens to me I guess. But I don't understand you when you say you "be good at retaining what's taught here in the learner's memory"?
Maybe it's a problem with my English :-)
I meant this: Duolingo teaches us some stuff. If we want more than that - extra grammar or audio lessons or whatever - it is fine to suggest other sites. But things that are taught here should be taught really well.
I agree, I think the system should involve the users more. Something like after a lesson a menu showing, a Question like:
Select which words:
- Are particularly hard;
- You're likely to forget; and
- You've already forgotten.
Most of the time, I've already forgotten the word by the time the lesson is over. Especially German words!
There should be some features preventing users from rushing through new lessons one by one. They don't have to be compulsory, but a recommended practice session (or two) after each new lessons would be great. I guess many users (like myself) go through the tree too fast and then realize they forgot quite a lot, so they have to 1) practice too many words to bring them back to memory; and 2) change their routine so that they would retain new words better.
My current idea of a good schedule is:
Refresh the skill(s) learned the day before.
Go to Vocabulary, sort it by "Last practiced" so that the less recent words are on the top. Practice weakest words until there are no words practiced long ago.
Take a new lesson.
Repeat it a few times or do a few practice sessions of the current skill.
(Highly advisable) Refresh the newly learned words later the same day.
I'm afraid I'll still have to adjust this schedule according to my results.
Indeed, I totally agree, that's what happened to me in French, and now I don't even know what I don't know!
I sometimes think duolingo is using us a test subjects to determine which approach the most successful users take to complete the language with the fewest mistakes (at least I certainly would). I recall tatou or Luis mentioning something to this effect, indicating that the most consistent users practice one or two times day, and do one or two lessons.
I like your idea of a schedule, but I lack your stern discipline, I just wish I could setup some schedule routine in duolingo to do whenever I login. My routine was a little bit similar to yours:
I followed these steps:
- Do several lessons;
- Do specific lesson's timed practice until you can beat the timer consistently (e.g. 3 - 5 times in a row), and with more than 15 points (75%).
- Do the main practice (homepage) till you can beat it consistently (3-5 times, 15 points)
- Once you're truly confident then do go to step 1.
Since you've already completed two languages(assuming you had no prior knowledge), I think you've already found out what works for you, maybe you should stick with it.
I've noticed that doing timed practice until I can achieve some result does not guarantee anything. The thing is, you get different stuff when you do timed practice. If you have, say, 100 words that need refreshing and you get about 10 words practiced in every session, you'll have to do about 10 sessions regardless of your results. Today I did a real "practice marathon" and practiced until there were no words in my vocabulary that were last practiced earlier than yesterday. If I stopped after 3-5 sessions with 15+ points, most of my weak words would not have a chance to get practiced.
My discipline is not perfect, that is why I'm so eager that Duolingo be more helpful in suggesting what's right for me. As for my two completed language trees, the algorithms have changed quite a few times since then, so even doing my third tree, I have to constantly adjust my learning plan.
Repetition is at the core of language learning. And although the word strength system seems unpredictable, there is nothing to prevent one from systematically revisiting the tree.. Duolingo is continually tinkering with the sentences used so repeat visits to skills already "mastered" not only reinforce what you have already learned, but add to it.
Exactly, that is what I did once for French, I surmised that I'm probably forgetting most of the things, so I decided to revisit each lesson, starting from basics.
Each day, I would do timed practice on it and until I beat it with ease, and then move to the next skill, and work my way back down to where the last lesson in the tree. Meanwhile, I advanced slowly in the current lesson, to make sure I wouldn't get bored.
I agree I can't rely totally on word strength system, because at the moment I am focusing on German and I very rarely see my skills deteriorate.
Therefore I try to refresh my skills by doing lesson practice as often as I do lessons themselves.
Also I must point out that current system is good for majority of users because they are not probably active in their learning as we are, so it focuses more on their engagement by not being to "harsh" on them.
- System could have different schemes of operating in regard of users engagement. But this is very tricky as well as manually adjusting the word strength of a single word.
- If possible the system should analyze our previous learning patterns and take that in account. (If we have learned another language here on Duo)
- Track how long does it take a user to complete an exercise and the lesson.
Yes, I have read that some people find that their vocables detoriate too fast. Maybe duo could introduce three levels of difficulty, and everyone can choose what fits their learning style best? Surely there are people who learn faster/ need less repetition that others.
Agreed, a tailored experience would be best since there is no silver bullet to learning, some people learn better(and faster) with illustrations, others with sounds, and others with a mix.
I think the system should assume we'll never click the darned practice button, and just randomly throw us practice lessons every once in a while.
Otherwise, if for example, all users suddenly decide not to use the practice, then what good would the word strength be ?
When I thought about users who are not very active, I remembered step strategies. Word decay should not depend on time but rather on the other actions of the user. The more active is the user, the faster is the skill decay. What do you think?
Isn't that counterintuitive?If you're more active it means you see the words often, so you're more likely to remember them then someone who hasn't. Memory relies on time, what you recall vividly today, 5 years from now you may not do so, unless it was either traumatic or an extremely good experience.
No, that's not counterintuitive at all. When you are too active and learn a lot of new lessons every day, you have to practice more, too. Otherwise everything just mixes up in your head and you're stuck.
If you learn one lesson a day or even not every day, you're fine with a couple of practice sessions a day. You don't have too much to forget.
Ha, I see, I partly agree with that reasoning, although being active also means you are seeing a lot of old words. You can't form a sentence without verbs, nouns, and other words, that's how duolingo expects you to practice.
The trouble is, it doesn't throw enough all old words from the vocabulary when learning new ones, so we struggle and have to go to normal practice or immersion, to refresh or re-learn them.
Yes, that's the problem: some of the old words are not repeated anywhere else, so you forget them if you don't practice them specifically.
@olimo: I think I understand why, the trouble is that sentence pool is limited, and they "design" or "write" each sentence one by one manually. So it is a bit hard to ensure every skill contains a word previously learnt . Otherwise they could just pull in a sentence that includes a weak word.
Doing it automatically may work in some instances: I like books( j'aime des livres); I like apples - j'aime des pommes
But it may fail miserably in others, resulting in what Luis called "junk" sentences: the apple with salt is good ( Der apfel mit salz ist gut), the orange is good - ([Der] orange ist gut. (- wrong));
I think what could help is for Duo to provide us every single day a small snipet of information, be it news, a small audio clip, a picture with captions, something that makes us unconsciously learn, like we would in a natural environment.
What do you mean by step strategies ?
I am not sure if your second sentences is referring to my last suggestion, but I will try to clarify it a bit more.
The last suggestion is something that could be just another small element of the word strength algorithm. With keeping track of the time need for completing an exercise or a lesson system could see if a user is struggling with a certain word/sentence/concept more than usually.
Regarding your suggestion I agree with Dessamator. Also with that kind of system I think a lot of people would lose motivation very soon.
Step strategies are games played step-by-step. You make some action, then computer does. You don't have to act in real time but can think as much as you need. So, the faster you act, the faster the computer responds.
exactly my issue - I've had "eat" (351 times in Spanish/272 in Italian, and that's in 25-26 days, so more than 10 times per day), "drink" (244/236), "apple", "water", "milk" (in both Spanish and Italian) way too many times, while some others I never see enough of. so yes, with you on this one 100% :)
Sometimes practice will choose sentences that use comer when trying to get you to practice other words like pollo, huevo, or fresa. By looking at your vocab, this is my guess as to what's happening. Also, you're still fairly early in the trees and all your words are all pretty gold, so practice also doesn't have a lot to choose from. :)
That's what I was thinking too. But it just seems that every day I have someone reading newspapers/books, or drinking milk/water :) (I usually practice the whole vocab by swiping to the right in iOS app and choosing "strengthen skills"). What I meant was that I do get some words several times in one day and other just once in several days. It would be nice if, for example, instead of just saying "I drink milk" it could say "in winter the tiger drinks milk", or "on Thursdays the cow doesn't want milk", or "the pink elephant doesn't hear the man". Or something like that, you know :) there are other words I don't remember we'll, yet I don't see them often I don't know, maybe it's just an issue with the iOS app and it's a bit different in the web version, it's just that I use the app most of the time.
after we've talked about I've been paying more attention to what words I get and to my vocal list and I don't know, maybe it really is, like I said before, the problem with the iOS app, but I keep getting the same words every single day in about 50-75 percent of my "strengthen skills" sentences - read, eat, drink, write, water, milk, chicken, apple, while some other words I haven't seen in a week and I don't remember what they mean (like "fresa" I've seen a week ago, "pollo" I see every day). is there anything I can do other than removing those words from the list altogether?
if you click on a word in the vocabulary, you often can click on a button that is labeled "remove from practice". Unfortunately for the ones you named, it is impossible for some reason. This may be because they are basic vocabulary and are used in many sentences that explain later concepts and words. As long as the sentences you get contain "newer" words or grammar as well it may be useful that there are those words you don't have to think about much.
I also read somewhere that duo tends to bring words more often that many users have problems with. So maybe those basic words are deemed problematic bacause often new users will start a language but give up after the first few lessons, and duo sorts those words into the heap at which point many users fail. Don't know if that's true, but if so, they certainly should change that.
yes, I don't want to remove them completely, because there will be more tenses in the future (I ate, I will eat, etc.), but it's just very tiring. there are some words that I really have trouble with, because I haven't studied them enough, and yet I never see them.
@tatou: For me that would mean to proceed beyond the point where I feel comfortable. If I proceed faster, I have even more words I do not feel confident in. As they pile up, I fear I might lose my step and the amount of words/concepts I do not understand might become overwhelming. We should be able to proceed as slowly as we want, even if it means to repeat one and the same lesson daily for a week before we move to the next, without messing with the decay algorithm.
Hmmm... I'd recommend that you start spending more time doing lessons than practice for the next few days... both because of (1) how the word strength model will currently behave when your whole vocab is gold and (2) the fact that spacing out practice is better for language-learning.
It might be that there's a difference in how words are picked in the iOS apps, so I'll check into that, but I don't think so...
There are many variations of the word eat(e.g. eating,ate). Especially in languages with more than one gender for words. That's probably why that happens to you.
I know what you mean, but I haven't gotten to all those complicated ones yes, currently it's only present simple (yo como, tu comes, el come, etc)
It does work for me, but naturally you need to be persistent and it is rather slow. I got good memory so that helps, but still get words mixed up. I'm Spanish learning Italian now, and I think it does work, again, very very slowly. Don't think this system is guaranteed to take you to fluency but does give you the basics, and a decent vocabulary from where you can use other methods and master the language. There is no magic method that will teach you in three months, a new language takes at least one year to normal people. Never mind super smart people and savants, most of us are not.