1. Forum
  2. >
  3. Topic: Duolingo
  4. >
  5. Are the words presented when …


Are the words presented when we 'strengthen' random or targeted at those most needing practice?

July 30, 2017



In my experience it's neither. The general strengthening seems to favor reviewing skills in the beginning of the course, before strengthening later skills that are weaker. I don't like this behavior, so I manually select and review some of the weakest skills as well.


I have found that in the beginning they are repetitive of some of the easiest content you have learned but after you have progressed they will throw in something totally new and it is almost always translating from english into french. At first it was disconcerting to see something new even if they only changed the subject of a sentence but now i find that i am easily able to do it.


I have finished my tree, but when I do a general strengthening exercise they still target the easiest skill that needs strengthening. I would love it to be more random, throwing in sentences from different skills.


it is almost always translating from english into french.

This is so important. There are so many complaints (obviously well-founded) about insufficient translation into target language, but so rarely does it get mentioned that it does get better as one keeps going along.


Install the user script "DuoLingo skill strength viewer" and strengthen per skill, not with the global button.


What makes this different from the strength bars?


I find that strengthening per skill does not choose the words that need strengthening according to the word list.

If I look at the word list, choose a word I have not strengthened for 6 months and is down to one bar, and strengthen that skill even 6x (as I did yesterday when I tried to strengthen a word in the skill 'food') the word I wanted to strengthen did not come up even in the 6 attempts. Then I looked for the particular lesson it appeared in and redid that lesson, and amazingly it didn't come up then either.

Thus going up and down the tree to strengthen doesn't seem to help, which is previous advice I have been given some months ago.

So I am wondering if it is random, and, being random eventually the words I want to strengthen must appear. It is possible to throw six on the dice six times in a row, even when random. I don't know how big the choice of words is in the strengthening sections and if I simply must persist and I will get to the words I know are weak eventually but until then DL is concentrating on giving me words I need even more.

Is it a case of 'DL knows best'? Or is it all random?


i'm pretty sure it is random and the lessons are different depending on if i'm on my phone , pc, or tablet. I do all my lessons on paper also and have repeated them at least 2 times. None of them are ever the same.


It looks like the "Words" list is the old Python API?!?
How do I know?

Because the API is still compatible with the Android program lekz has wrote for the old portal, even on the new portal: https://www.duolingo.com/comment/19154823/I-ve-made-an-Android-app-to-review-learned-words-same-as-in-Review-Flashcards

I have made my test recently and the DuoLingo system works as crappy as before on the old portal (where the "Review flashcard" button was still there for menu register "Words).

Crappy means: Not all important words are shown, not updated, and new words are displayed as flashcards which I even have not learned before and which some are definitely NOT on my words list (I would know after 264 days if I have ever heard that word on Memrise or DuoLingo at the current point of my tree)!

At the moment I must have my doubts that this service backend code ("managing" adding/updating user learned words database) has been even ported to new Scala code or has been improved (re-written, better concept, bugfixed, etc.) in any way!?!
Of course I do not know in detail, but this is my personal feeling...

Please don't tell me about all the bugs :-) I know they are in the Duo system!

There have been multiple thread about all those issues (even 2-3 years ago) with zero feedback from the DuoLingo support, some threads created by me, many comments added by me, etc.

And of course strengthening does not update the fields, e.g "last practiced" and strength bar accordingly, no matter if correct or not.

Other threads - especially German threads (we German somehow like to talk in a "direct way'" deal with problems transparently) I had come across are even talking about "real spaced repetition" does not exist on DuoLingo and it is just "random"???

Also I find there is a bigger gap between "skill strength viewer" and checking the percentages of some skills vs of tree skill gold vs non-gold and display of strength bar presence.
How can that be? Are words assigned that wrong to specific skills?

Why is there a high death words (0 strength) count???

Well, how shall the Duo system efficiently select all my weak / difficult (marked) words (haha - I wish!!!) and strength 1-3 bars when the user learned words database is NOT updated accordingly either with flashcards or by strengthening or learned words can not even be found in that list?
This is one of the many reason because I moved to Memrise from the very first beginning to efficiently train my Portuguese vocabulary.
Lots of existing user scripts which you can (actually have) to use to make reviews more efficient.
I have to admit that training grammar with DuoLingo sentences is still fun...sometimes.

And having many alternatives answers available to questions / sentence translations is even more fun.
Memrise can not do that (even with offical PT1-7 courses and 50languages I get many longer phrases/sentences presented, not only single words).

I would be very curious to read from core backend DuoLingo developers, if they have been doing anything on that code parts, if they have been re-engineered in Scala, if they are still old Python code (service backend module) and if they are even aware of all those bugs and problems (how can they not be aware???) and why noone actually cares about these facts.

However I do not care really much, not anymore....as I just do NEW skills/lessons on Duo, review just for fun like how I want it to do (may it be random)...and I heavily focus on vocabulary spaced repetition (SR) - including short-term intervals - not on DuoLingo itself.

I can simply not expect, that DuoLingo trains me using new/weak/difficult/review due words in a mixed context with different content style sentences, and real world questions/phrases, I could actually use to train me actively speaking and learn those words better, if I fail some questions. Instead of I often have to repeat the same questions and same (basic) vocabulary over and over again...

The Memrise crew on the other side has also very important things to do like introducing a frontend change "Ziggy" :-) :-) If it wouldn't be the users like Cooljingle and other script gods who provide me with all the nice (and required) script workarounds, I may also not be using Memrise for vocabulary reviews!
Let's see where this ship travels to...


Thanks for taking the time to write all this. I have read it a few times and it is a bit complicated for me to understand fully, however I think you seem a bit angry with DL?

The word list didn't use to update when I learn a new word or practice an old one, but it does now, so that's much better.

I am simply trying to find out how best to use what we have. I don't want different scripts; I would not be able to use them as I do not understand. I am on windows 10 or on an android tablet and both seem to work the same to me. I always go online for DL and never use an app.


the word I wanted to strengthen did not come up even in the 6 attempts. Then I looked for the particular lesson it appeared in and redid that lesson, and amazingly it didn't come up then either.

Sounds like you may have encountered one of the infamous "ghost words."

In general, I do think the system is random, but I think it's actually even a touch worse than one would assume saying "random." I think by that one would assume that the different sentences in the course come up with equal frequency, so that the words would then show up with a frequency corresponding simply to the number of sentences they're in. However, it's more complicated than that: Duolingo's algorithms (based on wholly non-transparent criteria) actually select how often to present the different sentences to users. One crazy example from the Russian course had two very similar sentences presented with a frequency that differed by a factor of 100. So if a word is in its minimum three sentences, and those sentences get chosen to be presented at low frequency, then you're just not going to see those words often. Working on my GuaranĂ­ tree, I redid lessons more often than I would have expected to just because there were plenty of words that I'd just never see otherwise, even with many runs through timed practice.


Do you know about the available spaced repetition (SR) algorithms wiki articles?

Memrise e.g has better defined intervals and you can even see all outstanding (blue, to be watered) words level-wise (in the web):

If a single DuoLingo skill has ~70 words (7 words per lesson x 10 lessons = ~70 words) our problem on DuoLingo is that quite often a skill will be non-golden and does not have 4-5 strength bars.

For the tested vocabulary from the full list of ~70 words (probably 10-20 words per strengthen excercise???) a single skill would have to show non-golden quite often, even if the words select/update algorithm would be working fine?!?

I do not like DuoLingo's complete intransparence what words from what skills have to be reviewed.

Learn a language in just 5 minutes a day. For free.