Thoughts and impressions about SRS in general.

[mzatanoskas]

While my mnemosyne is well and truly out for the count, and I can't go
through my scheduled cards, I thought I'd share a few of my thoughts
and ideas on SRS in general and see what you guys think.

1. I seem to remember reading somewhere (on the supermemo site?) that
the algorithm was designed to present fcs when you had a 90% retention
rate. If so does that mean after every session of scheduled cards you
should have forgotten about 10%? And you should forget each card one
out of every ten times.



I imagine however different users will want/expect different amounts
of recall and change the way they grade the cards accordingly. I know
I have changed the way I grade the cards over time. Could this result
in an endless battle between algorithm and user, one trying to shift
the spacing to acheive a 90% retention rate and the other grading the
cards harder and harder to achieve a higher retention rate?!



2. Indeed, how do people deal with the necessity of different levels/
types of recall required for different flashcards. For example unless
you planning on winning quiz shows, immediate recall is not required
for most "factual" trivia type flashcards, ie the exact date of a
King's reign for example. You might want this knowledge so that when
you come across about some other historical event you can put in
temporal context or whatever. However some other set of flashcards
might deal with something more relevant to you
in your immediate life and you need faster recall. Is this simply a
matter of revising flashcards more often? To a certain extent I
imagine this would be true: the more you revise a flashcard the better
and the faster you can remember it. If this is the case, would it make
sense to be able to tag different flashcards/categories with the level
of retention required, ie. instant recall, slow recall etc. This would
change the frequency that the flashcards would be scheduled at, whilst
allowing the user to still grade all their flashcards by the same
criteria, well remembered, not well remembered etc.


3. I have to say I was a little skeptical about learning anything
other than factual or vocab style little knowledge units using
flashcards. Reading the supermemo site persuaded me that a much larger
range of knowledge could be divided up and effectively remembered
using SRS. However after 6 months of using mnemosyne I'm back to being
skeptical about the actual real life effective breadth of application
of SRS. Although I have become much better at paring down wordy
flashcards to the
bare minimum it seems to me that it is all too easy to remember
flashcards answers as words or information but not as meaning. That is
to say in response to a certain cue, ie a simple cloze deletion
phrase, I can remember the word that is associated with it, but that
does not mean that I am processing the meaning of the phrase, or the
word remembered and the information recalled is basically pointless.
Recently I've noticed with some old flashcards, that I can read the
cloze deletion phrase, think I've never seen it before but when it
comes to the missing word, I'll unerringly blurt out the exact answer
without any hesitation and complete confidence. This tends to be after
very long gaps, and I have obviously remembered the flashcard very
effectively, but it is a meaningless memory.

Also I find that I'm getting very good at recalling stuff in the quiet
of my room in front of my computer, when I'm in "mnemosyne" mode.
However I often can't remember the same stuff when I'm out and about
in the real world and feel like testing myself. I think the basic
problem is that it's easy to forget that SRS is a tool, you devolve
the responsibility of useful memory to the software. I try and combat
this by making sure I'm not focussed on just remembering the word, but
I'm remembering the concept, the meaning and everything behind it,
everytime I come across a certain flashcard. There is no doubt that
this actually slows the memory creation and makes a less precise
memory, often I will remember the wrong words in a cloze deletion but
with the same meaning, but it seems to be a more meaningful memory.

Another related impression, is the feeling that some of the memories
I've created using flashcards are like a factoid that I've written on
a piece of paper that I can have a look at whenever I want. However
the second I want to really think about the factoid, process it,
combine it with other knowledge etc (the very reason I learn most of
the stuff I do), it's gone, absent. I have to "drop" all other mental
processes, go back to my mental "piece of paper" and read the factoid
again. Only when I've repeated this many times can I "use" this
"memorised" fact in the real life. This is especially true I think for
facts remembered using devices such as peg words, mnemonics etc. The
end result is that the "knowledge" memorised using the flashcards is
very similar to just having a copy of wikipedia on your ipod. Anytime
you want to think about something, you still have to go to your
wikipedia, "search" for the term, read about it and process it all
over again.


4. Have there been other ways of thinking about how to get the user to
"grade" flashcards. There seems to be a lot of leeway for users to
choose between the different grades. After working on a lot of non-
standard flashcards, (ie a bunch of sentences I half translate/half
remember back into a foreign language) I thought of this method. It
involves a number of grading steps which makes it more complicated,
but at the same time more specific criteria and so less leeway. The
first grading occurs before you see the answer. It is a confidence
question. How confident are you that you are right? Very, Somewhat,
Not at all. This helps identify mistaken memory, where you've
remembered the wrong answer. It's also interesting to see that
sometimes you have no idea, and you are sure you are wrong, but
actually you've remembered the card perfectly (don't know how the
algorithm should deal with that situation though). Next one you've
seen the card you grade again, not on how difficult it was to
remember, or anything about the interval, but whether you were
correct, slightly wrong, or completely wrong. I imagine by removing
any user feedback on quantifying the interval would remove some of the
user manipulation of the interval. Finally the timer would see how
long it took you to give your answer and ascertain any final
information from that.
I guess this kind of grading would need a very different algorithm
and some kind of theory related to mistaken memory, and how to correct
it etc...



To sum up, over the last 6 months it seems to me that reducing
effective, useful, real life memory to flashcards and software is much
more tricky than it initially seems, however many times you might read
the Supermemo 20 rules of formatting knowledge! It is very much an
art, and far more time needs to be spent coming up with the right
flashcards than actually then learning and reviewing them. These days
I seem to be treating the flashcards as memory prompts, (to think
about the concepts involved) rather than the aim of the memory itself
(did that make sense?).

Anyway, sorry for the long post, what do people think?

[querido]

To mzatanoskas:
I'm very happy to see your post.
All of your thoughts are very insightful and I understand everything.
In the past, I've tried *as hard as I'm able* to understand and to
explain similar ideas.

I don't have much time now. Please just let me say I understand your
post, and repeat some things I've posted here before.

Briefly stated problem:
Doing the standard, "minimum information" flashcards is like
memorizing a dictionary. This is very helpful, in Chinese above all,
but I get the impression that the information is **not stored in the
brain in the place where real words lie ready to be automatically
linked into thoughts** (as you suggest above). Use this dictionary,
yes, but more work must be done before the words become useable
language. Some of this work gets done *incidentally* (as you mention
above) while doing the flashcards, but manyfold more must be done,
eventually. Looking at this additional work (real reading, writing,
speaking, listening, etc), I offer two observations: 1. It seems to me
that it is *much harder in nature*, necessarily, than answering
flashcards (as riding a bicycle is, initially, harder than reading a
book on how to ride a bicycle). 2. Although it is harder, it is
*wholesale* work; while doing those four tasks, for example, tens to
hundreds to thousands of items get fed into streams of real language.
*That* is the necessary practice, facilitated by the internalized
dictionary. So the problem is that what the flashcard program is doing
for me is somewhere far short of what I thought (as you also seem
surprised about above).

Some thoughts about a solution:
Reinterpret the flashcard program's function, acknowledging that it
is just a dictionary.
Suppose you've been patient and have done the additional work
described above and are "fluent" through lesson n. Now you need to
learn some new words to master lesson n+1. The flashcard program 1.
ensures that you don't forget the words you needed through lesson n,
and 2. minimizes the effort of adding the additional words to your
mental dictionary. Although (as I argued above) this is not in itself
the additional language of lesson n+1, it is a convenience to use
while you're *really* learning lesson n+1.
Now *that* sounds like no more and no less that what the flashcarding
theoreticians claim. So where are my grounds for complaint and where
did I go wrong? I grossly overestimated what the flashcard system was
doing for me and forged ahead, by at least 2500 words, past my
"fluency", and ended up with what has been for me a millstone around
my neck. I'm very happy to "know" a dictionary of 3200 Chinese words,
but now I must retain it while real "fluency" hopefully catches up.

I would urge beginners to slow down and somehow really, really master
lesson n (the real language of the lesson, not just the flashcards)
before adding too many more cards. How many more? I'm not sure.
Probably 10-100 rather than 100-1000.

Attempt at a general solution:
Now I'm going to mention something that will happen someday.
There is a continuum, not yet handled easily by flashcard programs,
and out of their scope some will say, between the minutae of minimum
information cards, and larger real-language structures like whole
stories, etc. I tried to describe elsewhere a system that would
recursively parse whole books down into chapters, paragraphs,
sentences, words (storing the intermediate chunks), make minimum
information flashcards of the smallest bits (characters, words), then
as those cards are mastered, would automatically back up the parsing
to the sentences constructible from those words and make flashcards of
those sentences (perhaps then suspending the word cards). (These
larger flashcards would probably call for smaller multipliers when
promoted.) Continuing this process, one would work toward the ultimate
goal of let's say having the whole book scheduled for a once a year
reading. Miss a word, can't understand a sentence, reactivate those
cards. (Do this with audio and dialogue too.) So the program would
escort one from the bits to the whole language.

Does this help you at all, mzatanoskas?

[Gwern Branwen]

On Mon, Jun 7, 2010 at 6:28 AM, mzatanoskas <mzata...@googlemail.com> wrote:

> I imagine however different users will want/expect different amounts
> of recall and change the way they grade the cards accordingly. I know
> I have changed the way I grade the cards over time. Could this result
> in an endless battle between algorithm and user, one trying to shift
> the spacing to acheive a 90% retention rate and the other grading the
> cards harder and harder to achieve a higher retention rate?!

I don't think so. If you lie to the algorithm and say you're only
remembering 89% of your cards, say, the algorithm will just shrug and
move N cards up X days. If there were any endless battle, then
wouldn't you see a gradual ramp up in card numbers or a bimodal
distribution of well-remembered and nigh-forgotten cards or
*something*? I know that not many people looked at the stats Peter
distributed in a torrent last year, but nobody pointed out anything
odd like this.

> 2. Indeed, how do people deal with the necessity of different levels/
> types of recall required for different flashcards. For example unless
> you planning on winning quiz shows, immediate recall is not required
> for most "factual" trivia type flashcards, ie the exact date of a
> King's reign for example. You might want this knowledge so that when
> you come across about some other historical event you can put in
> temporal context or whatever. However some other set of flashcards
> might deal with something more relevant to you
>  in your immediate life and you need faster recall. Is this simply a
> matter of revising flashcards more often? To a certain extent I
> imagine this would be true: the more you revise a flashcard the better
> and the faster you can remember it. If this is the case, would it make
> sense to be able to tag different flashcards/categories with the level
> of retention required, ie. instant recall, slow recall etc. This would
> change the frequency that the flashcards would be scheduled at, whilst
> allowing the user to still grade all their flashcards by the same
> criteria, well remembered, not well remembered etc.

I think this is resolved just by either loosely grading things or by
adding more cards. For example, I was studying the history of Chinese
technology for a while, and I graded myself much more harshly than I
do now. (For example, I used to grade a card 2 or 1 if the year wasn't
exactly right; but now, as long as I have the century and decade
right, I'll give it a 3 or 4.)

If you really want to memorize a date, then you can take measures like
permuting all possible Cloze deletions. (I posted a Haskell script for
this on this ML earlier if you want to see examples.)

> 3. I have to say I was a little skeptical about learning anything
> other than factual or vocab style little knowledge units using
> flashcards. Reading the supermemo site persuaded me that a much larger
> range of knowledge could be divided up and effectively remembered
> using SRS. However after 6 months of using mnemosyne I'm back to being
> skeptical about the actual real life effective breadth of application
> of SRS. Although I have become much better at paring down wordy
> flashcards to the
> bare minimum it seems to me that it is all too easy to remember
> flashcards answers as words or information but not as meaning. That is
> to say in response to a certain cue, ie a simple cloze deletion
> phrase, I can remember the word that is associated with it, but that
> does not mean that I am processing the meaning of the phrase, or the
> word remembered and the information recalled is basically pointless.
> Recently I've noticed with some old flashcards, that I can read the
> cloze deletion phrase, think I've never seen it before but when it
> comes to the missing word, I'll unerringly blurt out the exact answer
> without any hesitation and complete confidence. This tends to be after
> very long gaps, and I have obviously remembered the flashcard very
> effectively, but it is a meaningless memory.

If you are memorizing the letters in context, and not the meaning, you
need more flashcards, and you need them to be hard enough that you
can't memorize them.

Suppose you have a '___ is not bar' / 'Foo' flashcard. Eventually you
memorize the three letters 'Foo' and forget the real meaning. Oops.

The problem is that you only had 1 flashcard. You should have had '___
is not bar', 'Foo is not ___', '___ is quux', 'Foo is ____'. What's
easier to remember, 'Foo is not bar and is quux', or 'Foo, bar, foo,
quux', etc.?

Do you understand the principle? When I'm adding assembler or Scheme
flashcards, I try to add multiple variants on each code fragment,
which differ subtly enough that unless I actually understand and can
parse them, I am just guessing - the flashcard whose answer is 'false'
looks almost exactly like the flashcard whose answer is 'true'.

You just need to come at the skill or knowledge from multiple
directions. See my discussion of how one can learn 'skills' (like
multiplication) via flashcards:
http://www.reddit.com/r/cogsci/comments/9aufn/ever_wanted_to_analyze_860mb_of_spaced_repetition/c0reso2

--
gwern

[mzatanoskas]

Thanks for the replies querido and gwern!

I feel that I kind of agree with all the insights in your comments
above.

I know nothing of the theory of SRS and memory aquisition, but I feel
that what we're talking about is a tension between the origins (I
presume) of SRS as a simple memory retention aid, and how far we can
push it to give us a complete knowledge and skill training software
that can teach us anything with maximum effectiveness. It is
interesting to hear how different people are approaching this.

As far as language is concerned, I've done a number of approaches so
far. For the language I am learning from scratch, Russian, I started
using picture question cards from Rosetta stone and audio + written
answers. I quickly had to adapt that because pictures are not clear
enough and so added english translations to the questions.

I also found vocab lists with audio and did the classic question
english word = answer russian equivalent. I quickly stopped this
though when I realised that just learning these words in isolation was
boring, unsatisfying and frankly not very useful.

After a while I also felt the Rosetta course was really not my thing,
(I don't think the whole not a single grammatical explanation approach
is a time effective way of teaching languages well Russian anyway for
adults). I started using the Princeton Russian course available
online. This is a classic grammar heavy approach and I added all the
sentences for each lesson as the question in English and the
translation in Russian as the answer. I've been carrying on with this
for quite a few months now. The problem with this approach was 1), at
the pace of 1 lesson a day, I was adding 40 or so difficult flashcards
a day and although every morning my brain gets a might translation
work out, it felt like I was spending too much time in English, and
well translating/memorising sentences is still very different to the
feel of speaking a new language.

All the time I've had my worries about the "naturalness" and the
applicability of these flashcards to the actual act of speaking a new
language, as I never touched a flashcard before in my life to learn
the other languages that I speak. One step I did which helps a lot I
think was to make sure I wasn't treating flashcard retention as the
goal during a revising session. In other words when dealing with vocab
lists, I would make sure that if the flashcard for book came up, I
would answer it and make sure it was correct, but then take the
opportunity to look around my room, pretend I was Russian and spot all
the books in the shelf, and loudly call out, "ah, a book!" Even though
I could only say one word at a time, if you try really hard, and have
a good imagination, you "feel" more Russian immediately and you "own"
the word much more this way! (at least I did). Whenever I go to the
kitchen for a break or something, I make sure I name everything I see
in sight too.

For sentences it's the same thing, but I stick myself in an instant
role play situation. "The cup is on the table." I make sure I get the
answer right first, and then turn away from the computer and imagine a
beautiful Russian girl is asking me where her cup is, a angry Russian
hoodlum is shouting at me to get his cup, a worried old man is looking
for the cup his grandfather made him etc... all sorts of ridiculous
situations which I answer with the sentence with varying degrees of
emotion and accent! It's all very crazy when you're doing it on your
own in your room, but for me it's the essential link that makes these
sentences mine, and part of my new language ability rather than just
being a flashcard answer in my "memory bank" as such.

Recently because of the amount of time spent on these sentences in
English, I've been thinking of other ways to make flashcards. What I'm
doing now is cutting up the dialogues, and all the other audio in
Russian I have for each lesson in the Princeton course. And simply
putting these into both the question and the answer sections. If I
have the written transcripts in Russian I'll put these in the answers
too. I listen to the question, repeat it and the check I repeated
correctly with the answer, and then do my little role play rigmarole
too when I have time. This way I'm getting much more listening
practice of Russian, much less translation practice (which is fine as
I don't want to be a translator), it's easier on the brain and takes
less time too. The only problems so far with this method, 1) it seems
to sometime crash Mnemosyne !! (see my other thread!) 2) I'm not sure
how well it works to get that understanding into practice. Ie although
I can repeat the sentence effectively how well will this help me to
use the sentence/vocabulary actively later? Well I'll give a few
months and see.

Well if I'm not careful this is going to turn into another humdinger
of post which noone will want to read.

Just a few comments on the more "classic" flashcard formulation that
you suggested Gwern: basically as I as I understood your post, I've
come to a very similar conclusion to you. That for "one" piece of
information, (and I don't mean one sentence that contains multiple
bits of information), you actually need lots of flashcards, from all
sorts of angles to help prevent that flashcard from just becoming a
"word" remembered in context. But I'm still not sure quite how easy it
is to remove the context from the flashcard, I didn't quite understand
the foobar example. Also as it is it takes me ages to formulate
flashcards anyway, I'm wondering if it's not more time efficient to
make actually "wrong" flashcards, ie vague and not simple like
suggested by Supermemo, and then work your brain a bit harder to think
and understand at reviewing time. Re your multiplication idea, I
actually had exactly the same idea (honest!) when I started using
mnemosyne, I was in a self-improvement binge and was practicing mental
arithmetic on other sites too. However my doubt was that the algorithm
in mnemosyne was not designed to do this, and that the skill of mental
arithmetic is very different from memorisation. There may be some
memorisation at the basic times table level, but that should be dealt
with as normal flashcards, ie however well you remember the answer to
9*2=18, doesn't help you remember the answer 4*5=20. Rather than try
and make mnemosyne, the one stop shop for all my cerebral activity, it
was just easier and made more sense, for now at least, just to
dedicate 15 minutes a day to mental arithmetic and leave it at that.

Final problem for the day: the amount of time thinking and preparing
and wondering about learning stuff, and the amount of time actually
doing it!! Especially with the net, I find it oh too easy to spend all
the time i wanted to spend on a learning a new lesson, on
investigating new ways of learning, new resources, formatting new and
funky flashcards etc! I guess it's always a matter of balance.

[Dougie Nisbet]

On 07/06/2010 11:28, mzatanoskas wrote:
> 2. Indeed, how do people deal with the necessity of different levels/
> types of recall required for different flashcards. For example unless
> you planning on winning quiz shows, immediate recall is not required
> for most "factual" trivia type flashcards, ie the exact date of a
> King's reign for example.

One of the reasons I abandoned anki was the timer countdown in the
status bar. It got on my nerves and I couldn't figure out how to disable it.

A lot of my questions are images, and even once I think I know the
answer, I may spend some time deliberately looking at the image and
thinking about the answer or spelling or thinking of other ways of
trying to remember the question if it's leaching.

> Also I find that I'm getting very good at recalling stuff in the quiet
> of my room in front of my computer, when I'm in "mnemosyne" mode.
> However I often can't remember the same stuff when I'm out and about
> in the real world and feel like testing myself.

I find this too and it's frustrating. A lot of what I try to remember is
plant identification and botanical names and often when out in the field
I can't recall the botanical name for a plant even though I might get it
instantly in front of the computer.

> I think the basic
> problem is that it's easy to forget that SRS is a tool, you devolve
> the responsibility of useful memory to the software. I try and combat
> this by making sure I'm not focussed on just remembering the word, but
> I'm remembering the concept, the meaning and everything behind it,
> everytime I come across a certain flashcard.

That's what I now try and do for the same reason as yourself. I'm also a
bit more ruthless at deleting leach cards, and often edit 'live' cards
on the fly to make them more meaningful.

As gas also been stated elsewhere in the thread I also try and break
down the information into as small units as possible and, with things
like plant identification, may have loads of images, all slightly
different, for the same plant id.

Dougie

[Brainious]

Gwern:

>On Jun 7, 10:51 am, Gwern Branwen <gwe...@gmail.com> wrote:

> Suppose you have a '___ is not bar' / 'Foo' flashcard. Eventually you
> memorize the three letters 'Foo' and forget the real meaning. Oops.
>
> The problem is that you only had 1 flashcard. You should have had '___
> is not bar', 'Foo is not ___', '___ is quux', 'Foo is ____'. What's
> easier to remember, 'Foo is not bar and is quux', or 'Foo, bar, foo,
> quux', etc.?

Would you mind explaining your idesa on this example?

[Gwern Branwen]

I thought I was fairly clear.

I'll try again.

'Is Foo X?'
no

is a bad flashcard. You might just memorize 'no', nothing about Foo & X.

What you should do is have both flashcards:

'Is Foo X?'
no
'Is Foo not X?'
yes

Now there is no shortcut. What are you going to memorize, 'x = no, and
not x = yes'? Or the fact 'Foo is not X'? I know which one I would
find easier to memorize - the latter.

The principle: ask the question with enough different correct answers
that it is simpler to memorize what you want to know than to memorize
the answers themselves.

--
gwern

[le_sacre]

On Jun 10, 1:44 pm, Brainious <gers...@gmail.com> wrote:
> Would you mind explaining your idesa on this example?

I think I can provide another kind of example. I'm learning the
Chinese radicals (a finite set of graphic elements that combine to
constitute the practically infinite set of Chinese characters),
several of which are similar-looking and thus easy to confuse. So I
wanted to make cards to practice distinguishing similar radicals, such
as:

prompt: [力, 刀]
response: [lì (strength), dāo (knife)]

But with that card alone, I'll soon memorize just saying to myself,
"strength, knife," without even looking at which radical shape is
which. However, I added a second card, with the two radicals
reversed:

prompt: [刀, 力]
response: [dāo (knife), lì (strength)]

Now, whichever card I see, I am forced to pay attention to the
differences between the two radicals' shapes in order to come up with
the right answer. This can be extended to cards containing more than
two radicals to distinguish, by adding enough cards with different
permutations of the orders, e.g., [米, 禾, 木, 釆], [木, 釆, 禾, 米], and [釆,
禾, 米, 木]. You don't need a card for every possible permutation of
your information, just enough so that you force yourself to memorize
the information itself, rather than simply recognizing the card and
answering reflexively without thinking.

I also have cards testing each of the individual radicals alone. It
may seem like a lot more work to have multiple cards for each piece of
info, but beyond the fact that it solidifies my real-world recall when
away from the computer (because the info is not tied exclusively to a
single context), it also actually saves Mnemosyne-time because over
the long run it keeps me from making the same errors over and over
again.

[Brainious]

This is a great explanation, item formulation is very specific to the
kind of knowledge we are dealing with. In this context you examples
are truly efficient.

Germán Salízar Pareja
wanakumbuka

[Brainious]

Answering a question with many good and bad answers might increase the
repetition difficulty, which is a good idea to improve recall.
However, at the same time, having multiple answers for the same
question also builds up to interference, for long term memories
interference is recall's greater foe. Each time you remember
something, the single act of recalling this information make the
memory it self vulnerable to interference or modification (re-
consolidation). Hence if you make a mistake, you are also re-
consolidating this mistake,

For an efficiency point of view, having many answers to many
questions, might not be the best solution in all cases. It adds more
time building the questions, typing them, or augments time used by
simply clicking on the program used to memorize, and of course recall
time invested in each question.

A similar effect in this type of questions could be accomplished by
using a distraction on the question (this of course has to do more
with recognition more then active recall) but the purpose of this type
of questions is to reinforce the correct answer while being exposed to
a wrong answer.

If D has X characteristic, the following Q&A leads only to rote
memorization

[Question: D has X? / Answer: yes] (I)

But:
[Question: D has/does not have X?/ Answer: D has X] (II)

or;
[Question: D or E has X? / Answer: D has X] (III)

It all boils down to which is the cognitive relationship that you want
to remember. in (II) you reinforce the fact that D has certain
quality, in (III) you are looking to discard other probable elements
that do not have this peculiar characteristic.

I think that the minimum information principle (MIP), in the above
descriptions, are not being applied correctly in all cases. When you
apply the MIP on arbitrary information pairs, like vocabulary, its all
a different matter to learn the translation rather than actually
procedurally learning how to build fluency on a second language. Hence
by using MIP type questions that ask for declarative answers its no
wonder you don`t get the fluency on writing, reading, listening, even
less talking.

Germán Salízar Pareja
wanakumbuka

[mzatanoskas]

Thanks for elaborating your example gwern, I hadn't quite understood
it the first time either. I absolutely agree with you that the
multiple approach is much better, but what has surprised me over the
last 6 months of using mnemosyne is how efficient my brain is at
parsing out all extraneous information related to the flashcard in
order to remember the minimum required to correctly answer the card.
And even a multiple card approach is not as effective as I had
imagined. That is to say I have found that I'm quite capable of
remembering that "Is Foo X?" = no, from context alone, and "Is Foo not
X?" = yes also from context, quite seperately and without any
connection being made between the two. They are just two unrelated
flashcards for my brain. As I'm not learning the two flashcards in the
same instance, my brain does not make a choice to combine the two and
learn them as one unit of "Foo is not X" but just (very effectively I
might add) learns the answer to each flashcard seperately. I've
noticed this particularly with flashcards that are simple and that I
have memorised very easily and quickly and am now having long gaps
between repetitions. Basically I have found I still have to make sure
I make an actual effort everytime to contextualise the flashcard,
flesh it out as such in my head when reviewing otherwise I find it
just too easy to do it on autopilot (which when I first started
mnemosyne paradoxically seemed to be the goal). Finally when it comes
to accessing this information in the real world there is always this
strong feeling of a seperation between this information and other
information that I have learnt and know well via other means. I guess
this is not surprising as I have just "memorised" the info with
mnemosyne and "learnt" other info by other methods. I think part of
this gap is also due to underestimating the amount of small brief
"factoid" style sentence/questions that are required to fully flesh
out the meaning of even a simple flashcard such as "Is Foo X?". For
example all the understanding related to the nature of Foo, the nature
of X, and what it means for Foo to be X etc.

As far as Chinese characters go, I recently started revising the kanji
(Chinese ideographs) for Japanese in mnemosyne. I've started off by
the old-fashioned (and possibly unnecessary?) way of having the
question as the character and the answer as all the onyomi (Chinese
readings) of the characters, and a seperate card for all the kunyomi
(Japanese readings) of the characters. Something that really surprised
me here, is that when reading a text in Japanese, I noticed a word
that I didn't know containing a character that I recognised but had
forgotten the reading of. I copied pasted it into a dictionary and
when the reading came up I immediately realised it was a character
whose readings I knew very well, that I had reviewed many times in
mnemosyne, and had never forgotten it in that environment. What was
interesting is that the flashcard itself contained a minimum of
context, the character itself, nothing else from which I would have to
remember the readings. Yet I did not recognise that same character
when it was placed in a word in a sentence. It wasn't the fact that I
was relying on extra contextual clues to remember the flashcard in one
instance and so could not recognise it when those clues were gone, it
was the simple fact that the "decor" as such was different that threw
me off. I have no doubt that if someone had told me, hold on a second
you do know that character, you've learnt it as a flashcard, I would
have been able to visualise it in my mind and remember it after some
reflexion, but it was interesting for me to see, how a veil had been
placed between the mnemosyne memory and real life usage and I needed
to make an extra conscious effort to combat that.

I guess along with these more specific problems is the fact that you
recall memories best in the environment in which you learnt them. I
remember being told a long time ago to study for an exam in an
environment that most closely mimicked the environment in which you
would take the exam. I think this is fairly accepted now, I seem to
remember reading recently about a study which showed that people
remembered facts learned at the bottom of a swimming pool best when
they were tested at the bottom of the swimming pool! This would make
it particularly difficult for someone like Dougie who presumably
reviews his flashcards in a very un-nature-like office or room but
wants to make use of this knowledge in a completely different
environment, outdoors with the wind, the sounds, the smells of the
forest for example. Now once we hook up SRS with a bit of Virtual
Reality, replicating senses of touch and smell and everything, that
could be very effective! I wonder what will come first, that real life
simulation approach, or the directly write data bits to the brain
approach. I vote the former.

Well to sum up my thoughts so far, I think that to a certain extent,
while the action of whittling down, dividing up information, basically
the process of formulating flashcards together with the unchanging,
strict and defined memorisation process using SRS always in the same
environment allows for a great increase in efficiency in memorisation
it also results in a certain lack of flexibility, applicability
regarding that "knowledge".