I'm of the opinion that the random disconnected order that words appear in is both Skritter's biggest downside, and spaced repetition's biggest too. I'm going to suggest that some grouping of words by list is necessary for good recall in both Skritter and real life, and will lead to more efficient practise. Bear with me!

A problem with spaced repetition is that while sometimes words do get grouped together by virtue of being added together, for the most part they show randomly and you're left wracking your brain trying to figure out exactly what needs to be written. In terms of speed, this is slow, and I think not optimal for recalling either.

I tend to find that language is learnt best by making associations. For example, one day studying the word for "wardrobe" 衣柜, and a few weeks later the word for bedside cabinet 床头柜 is not as efficient as studying them together, being reminded of the links between them, and the differences. Especially when you come across a word which is similar, but not quite, and leaves you confused.

Now, the Skritter guys have done some great work recently to help us out here, including custom definitions, seeing list names, mnemonics and the star system (to highlight words you don't know). They're also wanting to include example sentences. This is all great, but words will still be thrown at you randomly.

I believe that if you have the words for wardrobe, and the word for bedside cabinet coming up together, this will reinforce recall. In and outside Skritter. Other than having taught language in the past, I've no great academic qualifications here, but I don't think it's that contentious.

My proposal is that while spaced repetition should of course remain, it needs to be supplemented by a more natural grouping of words. I hope this can be done without too much heartache.

I could see it work this way: Skritter should not only take into account grouping words by reading, writing, or spaced repetition, ***but also by being in a common list***.

Beside cabinet and wardrobe are in the same list. If on Thursday one should show up with "2 weeks since last seen" and on Sunday the other shows up, also with "2 weeks last seen" then they should appear both together, one after the other on Thursday. For words in the same list with a 12 hour difference, instead of 1pm and 3pm, put them immediately together at 1pm. Words separated by big time differences remained separated, but the closer ones in the same list get bunched together.

If this is done on a large scale, you will find that you will be able to get through reviews quicker, not constantly wracking your brain, and by association they will help your recall, and by seeing them side to side you will get less confused by similar but different words.

I hope I've made some way into convincing you here. I've been having a lot of trouble with the tremendous amount of vocab I've added, and I think this is the best way to deal with things.

In the long term, if this is implemented, I'd also like to be able to go through all the words I've added, and stick them retrospectively into lists, while allowing you to prune words you no longer want (instead of using the 'delete all' button which is far too extreme), but that would be more of an undertaking.

And er, thanks for reading!

(Correction: group by section of list).

I do understand your argument, but beg to disagree. While it is certainly true that grouping words improves recall, the problem is that grouping words improves recall. To be honest, I find that even indicating which list a word came from is sometimes more hint than I want. It often helps me remember the word _in Skritter_. The problem is that if I remember bedside table because of its proximity to wardrobe (or because I remember that it's the one from list X which kind of looks like...), then when I encounter bedside table in the real world I don't remember it because I haven't really learned to recognize bedside table; I've just learned that the word that comes after wardrobe is bedside table. Furthermore, when practicing I can't tell if I got the word right because I really know it or if I got it right because I remembered that it came after some other word.

To the Skritter team: if you decide to implement this, please, please, please make it an option that can be disabled.

dorritg: my argument was that associations help you in real life as well. I have some English theory books in the attic. Might dig them out.

But otherwise, how many words do you have added? I'm on something like 3300. It really becomes quite a chore.

Although I can see some definite advantages (and possible disadvantages) of grouping by list section, I can't get very far into thinking about it without panicking about how difficult it would be to do. I've already made the code for grouping by reading vs. writing prompts into a huge mess that doesn't always work.

I can see how tagging words would help with deletion later, but let me also ask: do you ever use the "delete" button inside the magnifying glass on the practice page when you see a word you don't want any more?

dorritg, if I did find some way to do this, the order wouldn't be the same each time, so it wouldn't always be one after the other. Sometimes it would be the other after one, or both with some other one in between. It would still be somewhat of a hint, yeah. I tend to find, though, that similar to Byzanti, seeing related words closer together really helps shake out confusion for me.

Hi,

I understand the argument but also disagree.

I think learning my association can give you an inertia so that you know words in context but this is, from my point of view, a weaker learning that the strong pure memory.

It's hard to go by little more than memory but it finally gets faster and, from my point of view, is the only way of finally having a command.

Recognizing characters is like a technique. I don't know if any of you studied music, but you must play and repeat scales hours and more hours before you can actually play naturally an instrument like if that instrument was part of your body.

This is similar, and, in order to get a full domain, in my opinion, you better get a really high level of characters out of context.

If you want to give context (which is also necessary) you can use textbooks. I, in fact, feed skritter with my own text books and read context outside, but, in skritter, just push technique going up.

Of course, all this changes from one to another and what it works for me, maybe does not work good for other person.

Luis

I'm with dorritg on this - providing an artificial context would likely decrease the effectiveness of the SRS algorithm. At best, it would create artificial mnemonics linking word pairs or groups together so that you could recall one by thinking about another. Regarding your example specifically, I find that, if I'm having even a bit of trouble on a character, say 柜 here, it's counterproductive to see it in two contexts so close to one another - I might fail and grade myself with a 1 or a 2 the first time, but seeing it again 30 seconds later makes it an easy short-term recall without really affecting my long-term recollection, so it's probably counterproductive to the SRS system unless I grade myself down to correspond with the previous instance of that character.

Aside from that, as Nick said, I could see this being a nightmare to implement. Not only would the team have to find an effective way to implement this and code it, but they would then have to find a way to respond to all of the people who rather have the SRS algorithm tweaked in a different way - 'I want radicals grouped!' 'I want words with the same character grouped!' 'personal names!' 'similar pinyin!' 'similar phonetic elements!' 'words with pinyin beginning with x!'

My personal take on the bottom line is this: if you have trouble recalling a word or a character, it's probably because you don't know it yet. Sure, it messes up your retention rate a bit (if you're setting personal goals), and it slows down the pace of your study a bit because you have to keep working on words that you don't quite have locked in, but presumably you're learning Chinese for practical application sometime, and when that time comes, you'll be glad that you don't have to keep the dictionary quite so close.

I wouldn't want this either.

I think the data in this report says don't group by sets: http://www.babylonia-ti.ch/BABY207/PDF/mondria.pdf even though it is often done in textbooks, vocabulary books and to do so may intuitively seem like a good idea because it feeds our need for ordering things. (Even Heisig did it by grouping by "primitives". I have also read some other research that confirms this anti-grouping hypothesis but I am not sure where I saw it. (If anyone is really interested, I will search around my vast and completely un-indexed and non grouped library to see if I can find more.) I think grouping words is bad, I've tried it and it was not effective for me. But then, neither is reviewing my SRS cards while watching TV! Hmmm... Anyway, I have a lot of dictionaries and my avatar even wrote a very famous one :), but I think they are more for interest and fun (BTW, my wife thinks I'm nuts) than for learning vast quantities of vocab or kanji/hanzi. They are great for finding stuff in an alphabetic way or some other list and interesting in that they are full of information that can be savored as knowledge, but for rapid fire fluency, give me random any day for the most effective learning aid. Grouping is good for making sure you know all the names of some category of something but I believe it tends to confuse when trying to maximize your reviewing effectiveness. The attached report confirms my experience! :)

The studies cited in that report actually suggest learning new words without (semantic) grouping at first, and then doing the grouping later. So it's arguing for learning from frequency lists or random words that you run into as opposed to "Clothing Vocabulary" or "Pizza Words" or such lists. But eventually you would want a function that would pull these semantically related words in together so you can disambiguate them.

Anyway, I don't have access to the original Waring or Tinkham papers he's referencing, but without reading their methodology, I am skeptical. It seems like this would be a tough question to fully answer with a controlled experiment.

Note, too, that they've only measured semantic grouping, not phonetic grouping or component grouping or other strategies. I would be extremely surprised to learn that Heisig-style grouping by primitives would be less efficient than not doing the grouping (on the same big set of single characters).

Anecdotally I seem to remember words better when they are not in groupings. I wish skritter would make the words appear with less association to other words in the list they came from (:

sorry but -1 for this feature (: