vosmiura Wrote:nest0r, so expanding intervals are not the key to memorization - but rather expanding intervals are a side effect of memorization we can take advantage of to reduce review times and learn more stuff?
That's my impression, but I'm still working this stuff out, since I only discovered the studies questioning the expanded thing a few months ago (
Edit: Actually nevermind, I saw it last year but ironically dismissed it while agreeing with them: “As for expanding vs. equal, I think neither of those is as important as 'spaced', just as I try not to focus on the algorithms--I think it should ultimately be a spacing that's semi-automated in accordance with the user's self-feedback, and the main reason for expanding the spacing is simply efficiency, maintaining recall with the least amount of review over time.”).
All this time I thought it was just me (thinking that the algorithmic fine tuning of the schedule wasn't as important as the user feedback that let you push reviews away as they got easier [in conjunction with something that was simply algorithmic]).
Here's an interesting bit from Karpicke, et al.:
“The theory about why expanding retrieval should be su-
perior to equally spaced retrieval practice is twofold. First,
an early initial retrieval attempt increases the likelihood
of success before forgetting has a chance to occur. Sec-
ond, gradually increasing the interval between repeated
retrieval attempts increases retrieval difficulty on those at-
tempts and thereby improves later retention (Bjork, 1999).
However, at least under our conditions, it is not clear that
retrieval actually grows increasingly difficult across re-
peated tests regardless of whether the tests are expanding
or equally spaced. Assuming that recall latency is a viable
index of retrieval difficulty (Benjamin, Bjork, & Schwartz,
1998; Koriat & Ma’ayan, 2005), in prior research we found
that recall times grew faster across repeated tests, indicat-
ing that repeated retrieval grew easier, not more difficult
(Karpicke & Roediger, 2007a). If retrieval grew more dif-
ficult, we might also expect recall performances to decline
across tests, but the present data show that recall remained
relatively constant or even exhibited a modest degree of
hypermnesia (see too Logan & Balota, 2008). Therefore
it may be true that increasing the difficulty of retrieval at-
tempts does in fact promote long-term retention (Pyc &
Rawson, 2009), but the data imply that current procedures
for implementing expanding retrieval do not successfully
induce increasingly difficult retrieval across repeated tests.”
I never experienced cards getting harder with increasing intervals, only easier, so it does seem wrong if the logic is that increasing intervals makes them more difficult, because they ought to feel the same as the retrieval makes the memory stronger. And when you grade Hard, it should be to increase the interval rather than shorten it. ^_^ Instead it's more like you're telling Anki that it was easy and thus it's okay to spread it out further so you can focus on other stuff.
Oh and to add a bit, I think for Karpicke and Roediger, the desirable difficulty through retrieval practice is the priority, and the spacing comes into play by ensuring that when retrieval occurs, they're not recalling from immediate memory. It's enough for them or through their research results that you generically maintain ‘a delay’ to increase difficulty. And to apply that to SRS, the longer the period of time you're spacing, and the more items you're spacing, naturally results in lengthier intervals that generically need ‘an algorithm’ to manage in conjunction with grading.
Here's another bit:
“But what about the mechanism for spacing of retrieval? Our
data reviewed above suggest that the critical ingredient is encouraging fairly
difficult retrieval, especially on an initial test. Beyond that point, it probably does
not matter whether students test themselves using expanding or equal interval
conditions. What matters is repeated spaced retrieval (with feedback if an error is
made).”
Edit: They suggest two days there for initial test delay.
(Not just retrieval errors, though:
http://www.ncbi.nlm.nih.gov/pubmed/18605878)
Apparently that idea of modulating cues and even shifting the modality was tested to good effect in the 80s and called multiplexing. Interesting.
Another idea of delayed feedback for desirable difficulties is interesting; I've been tinkering with small delays in feedback both for subvocalization/listening, but also when handwriting new words without consulting the screen in tiny moments (seconds? milliseconds?) between initially writing+consulting, and listening/speaking/encoding meaning. Simple but effective. (Yes, now I'm just riffing.)
Edit: Oops, I didn't notice this the first time I read the paper, but they found in a later study that retrieval difficulty did increase with expanding intervals, but it didn't result in superior recall, and “If anything, patterns of increasing retrieval ease, not difficulty, tended to be associated with greater levels of final recall.” The latter for the equal/contracting schedules. So natural and ideal subjective preference is, perhaps, for easier and easier cards, not cards which always feel hard. ^_^ Initial difficulty allowed to develop through spaced learning into strongly internalized knowledge. In that sense, rather than ‘just before forgetting’, i.e. maximal distance between retrieval and subsequent retrieval, any point beyond a substantive delay is fine for increasing performance, and the expanded aspect arises from the increasingly strong memories that allows for efficiency as inter-session intervals increase with very long retention intervals. Maximal distance through highly calibrated mathematics thus not optimizing recall. If so, perhaps it's better for the user to determine maximum approximations based on post-retrieval evaluations of expected difficulty, and the SRS to ensure there's an algorithmically increasing baseline delay that's not dropped through.
)