Feedback on latest test version of decay algorithm
Thanks for working on this Tatou - thought a bit if feedback might be useful after a couple of weeks.
a few early skills now seem to crop up over and over again - I seem to see the basics lessons implausibly frequently in all languages given that I don't make many mistakes here
in French, Être/Avoir and Modal Verbs also seem to recur far too often, especially as this is really not an area of weakness for me.
I think there is something similar going on in German, though I can't remember which particular skills
Could the problem be the incompleteness of indexing/cross-referencing of words and skills? Looking at the vocabulary page, common verbs in their various forms seem particularly likely to show as completely decayed in one form in one skill but fully strengthened in the same form in the context iof another skill.
- in Italian the decay rate in the last third of my tree is now huge - around 10 skills a day (previously ?3-5) which is more than I can refresh without abandoning other languages, hence the sad state of this section. This extra practice is probably a very good thing for my Italian! Edit: this seems to have been mostly a one-off, decay is now back to a manageable rate.
Thanks for the feedback! When you say certain words crop up to often, do you mean you simply see them too often in practice, or they are the words actually highlighted in blue (a signal that we're trying to get you to practice that particular word/concept)? I'm trying to get a sense for this because most students with the new strength model seem to find it it better than the old one in terms of refreshing older/earlier material less.
We don't index/track skill strength directly, because we are always experimenting with adding/changing/moving material around in difference version of the skill tree. The skills strengths are computed as an average of all the words/concepts in that skill for your version of the tree.
Now, sometimes words are actually different senses even though they appear to be the same at first glace: "I have a pencil" vs. "I have penciled it in" vs. "I have to use a pencil." We model these senses of "have" separately at the moment (and teach them first in different skills), so depending on how "correct" you tend to be with a particular version, they might have different strengths.
I don't think I see words in blue - would that happen in the app? I very rarely use the website.
My impression is that the Basics and other lessons I mentioned crop up much more frequently than they used to, as though the vocabulary and skills that they cover are no longer cross-indexed with their occurrence in the subsequent lessons. For example, the present tense of 'to be' and 'to have' occur in so many sentences (eg in units that are mainly focused on vocabulary) that I cannot imagine that I have not practised them exhaustively in completing a tree. 'Adjectives 1' also seems to be cropping up a lot across all languages. I also notice others making comments to the same effect, especially about 'Basics' lessons. I think I make very few mistakes here - less than one careless error per lesson on average.
The most recently learned things are getting a lot more practice, which is great: but the middle of each tree seems to get less attention than the earliest parts of the trees, as described above.
A further problem is perhaps worth thinking about. If a given verb is indexed separately under each tense, I will be required to practice it in each form. But there can be a lot of overlap here. Take an Italian verb like 'mangiare'. I might be expected to practice it in present perfect, future perfect, past perfect, conditional perfect and subjunctive perfect. But in all of these the participle 'mangiato' is the same, and the same considerations about participle endings apply. Over time I am likely to perceive that 'mangiare' is getting too much attention at the expense of words I know much less well.
Similarly the present, imperfect and future forms of 'avere', 'essere' and other common auxiliary verbs occur in multiple tenses.
Perhaps the tracking system could recognise that when someone uses(say) the conditional perfect correctly, they are demonstrating correct use of the conditional perfect of the main verb AND the conditional present of the relevant auxiliary verb AND the past participle of the main verb. After all, if they get any of these wrong, they will get the sentence wrong.
My reason for pursuing this is of course the pursuit of learning efficiency: the better targeted our practice is, he higher the pay-off for our learning efforts.
Now this is odd
- Basics 2 has come up again in Italian this morning
- I looked on the website to see which of the words are weak
- although the unit is said to have 16 words, the 'show all words' link gives me a list of only seven words, and an endlessly twirling icon - the rest of the list just will not load. (I have tried repeatedly)
- all of the words displayed are full strength and practised within the last few days
I also just tried this with another Italian skill not yet showing as decayed - Basics 1. I have the same problem: all words show as 4 bars, and some of the list won't load. But when I look at the next skill (Verbs Future) showing as decayed I have no problem loading the full list, and it shows the mix of 1,2,3 and four bar words that I would expect.
I have investigated this a bit more. If I do 'practice weakest words' for Basics 2 in the website version, the words that are being shown in blue all seem to be present tense forms of common verbs (essere, avere, scrivere, leggere). I have looked each of these up in my full vocabulary list. All are also linked to a specific Verb Present Tense skill, and all of these verbs are showing as full strength in those present tense forms. So it really makes no sense that I am being required to practice these verbs in the context of Basics 1 and 2. I am certain that I do know them forwards, backwards and sideways! I didn't actually complete any of the questions, I just went in and out of the 'practice weakest words' screen a dozen times, so you can see the vocab list as I found it.
I also looked up in the vocabulary list two of the forms of 'essere' that were boxed in blue when I tried 'practice weakest words' for Basics 2 - 'siamo' and 'siete'. These particular forms are showing as three bars and full strength respectively, so it seems odd that they are being singled out for practice. (I am doing multiple posts because for some reason I cannot start a new paragraph within a comment today.)
Ok one last post. If I look at the decayed Italian Basics 2 skill in the app, it lists eight 'weakest words': leggo, scrivete, siete, sono, ho, scrivo, beviamo and siamo. I have followed each of these through into the full detail verb page. Six of the eight are showing as full strength and two as three bars. This feels right to me as they are all forms that I know extremely well. So it is puzzling that they are being treated as sufficiently decayed to justify decaying the whole Basics 2 skill.
Further info from one of the French skills I mentioned in my original post, Verbs: Modal which has decayed yet again today. The app shows the weakest words needing practice as puisse, pourraient, pourront, voudrais, pourrais, pouvez, but the verb page in the vocabulary section of the website shows ALL of these forms as being at full strength. (The only part of pouvoir that is NOT at full strength is the infinitive form). And once again the full vocabulary list for this skill will not load.