Hi All,

Subject says it all. I have some software that generates nice sets of cloze deleted sentences, and I tried reviewing a few... It feels like production testing. Typically I _have_ to fail the first attempt as I simply won't remember every word I study, but the second, third, etc are MUCH more difficult because, well, it's production.

How do I reconcile the intermittent praise for cloze deleted cards that come by here with the intermittent production bashing session? Thoughts on best cloze delete methods, prompts, etc? Success? Failure?

I'm not stuck in cloze-delete land, but I thought I'd get some feedback before I edit my card layouts.

K.

I think for spaced retrieval where you're learning vocabulary words, cloze delete is a bad idea, because you're making the cue stronger with the extra context, which reduces the effectiveness of the retrieval. So as you said, it's basically just a (less effective) production card. Especially if you add a mass of sentences for context per card. But for practicing your ability to read and inference where learning a vocabulary word is only tangential (e.g. it works best with words you already know at least partially, rather than clozing unknowns), and you're just getting used to reading with a certain amount of uncertainty, putting together the gist, that's where the sentence/passage cue is useful. It's sort of a controlled, enhanced slow-mapping process. Or use clozes for grammatical function words and such, or if you have a much smaller cue, like a collocation (extracted with Antconc) where you cloze only one portion of it (the unknown, or if both are unknown, you could try clozing both but one at a time on different cards; not that they need to be unknown since the goal is associating commonly co-occurring words, perhaps refined by the context you took them from, e.g. a light novel via Antconc).

So basically the MCD thing, as I see it, is a misguided and thus inferior version of what I came up with in this thread: http://forum.koohii.com/showthread.php?p...#pid121789

And as for just single sentences, these days I'm against doing these for learning unknown words, but instead for learning grammar/structure/prosody, etc., often as a reinforcement for words learned in an isolated, focused fashion on vocabulary-oriented cards. If you have multiple unknowns in a single sentence, similar to the aforementioned collocation ratio, I think that's okay, because you're offsetting the ratio of cue strength to retrieval target difficulty. But at this point I think it's better to just have multiple overlapping card types, especially easy now with MorphMan.

Speaking of which, with production cards, I think for either o+1 expressions (need to polish my notes on this before posting further) or single word production cards, MorphMan works great, because first off, words done as recognition are best for developing recognition, and production for production (re: transfer-appropriate processing, and here single words work great because you can modulte the strength of the cue across modalities/senses i.e. audio/image/text, in a transfer-appropriate fashion using the multiple routes that one wants to build with a lexical item), but with MorphMan you can reverse them when the recognition cards are mature, so once you've developed your receptive abilities with those words you can refine your active lexical knowledge in a timely fashion.

Another idea I'm playing with, for reading outside of Anki, is to do things in a sort of Mad Libs style, using clozes with parts-of-speech indicators (or perhaps even the definition) to replace unknowns (since outside Anki the cue strength isn't a factor).

Edit: There's also the problem I mentioned elsewhere, where if you're using a single sentence/passage to create multiple cloze cards to learn new words, that repetition reduces the effectiveness of the spacing by messing with the intersession intervals, as you're giving yourself the answers to other cards by having clozed cues unclozed on the cue side of other cards.

Better to just sample passages in your reading and have one passage on one card and leave it for rough spacing and loose grading, clozing multiple different knowns on it where you think it's be most effective at destabilizing it and giving you a sense of uncertainty of the meaning.