The Kanji Sketch Pad - Part Six, Learning Targets

The graph on the left shows the author's own modest kanji knowledge after about 3 weeks of using the program. Note that the 'live' version of the graph has a few extra features. It can be resized, or zoomed, for instance. Drag to the left of the y-axis to focus in on a range of item numbers, or below the x-axis to focus in on a range of days. Drag diagonally across a portion of the graph to zoom in on that area. Double click outside the graph to zoom out. Hover over the 'Target', 'Known', 'Secure' or 'Stale' labels to get an update on the actual values.

But what does it all mean?

The 'Known' count is a measure of how many items have been answered correctly often enough that the kanji coach accepted that you knew them reasonably well when they were last seen. It includes any items you have 'claimed' by pressing the Claim button (the blue tick), unless you have subsequently failed to get them right when they came up for revision. It also includes some items that were known when they were last seen, but which might have been forgotten since. (Once an item becomes 'known', it keeps that status until you get it wrong.)

The 'Target' parameter is set by pressing the Target symbol in the Mnemonic window. Enter a starting value and a climbing rate in items-per-week, and this value will climb relentlessly forever more, challenging you to keep up. As shown in the graph, I had a target of zero for the first two weeks of using the Kanii Sketch Pad, and then set the target to 50 with a climbing/learning rate of 10 words per week (not very ambitious but hey, I was programing). The learning rate has since been adjusted to 20 items per week but this is not yet relfected in the graph.

The difference between the Known and Target values gives a progress tally, which should be kept above zero if possible. In the graph shown on the left, the score is +22, which is fine. (Known 94 - Target 73 = Score 22). Negative scores appear in red and should be remedied as soon as possible.

The most subjective parameters on the graph are the 'Secure' and 'Stale' counts, which have a clear reciprocal relationship: the secure count falls as the stale count climbs. An item is considered stale when it is overdue for revision - a complicated concept discussed in more detail below. At the moment it turns stale, an item has an approximate 90% chance of being remembered. With increasing delays, the chance of it being remembered falls considerably. The Secure count is an estimate of how many 'Known' items are likely to be secure in long term memory at anyone time, factoring in that at least 10% of stale items will have ben forgotten.

Finally, the Total kanji count represents the size of your vocabulary pool, and the Seen count is the number of items you have seen. When you have eliminated all of your stale items, you are exposed to a mix of new items and seen-but-not-yet-known items. This mix is adjusted dynamically, and as you learn each item more new kanji become available to be seen. This process is largely invisible and automatic, but future versions of the Kanji Sketch Pad will allow you to adjust the parameters to suit your own learning style.


Spaced Repetition Systems and the Cerebware Learning Model

Cerebware uses a type of spaced repetition system (SRS) that is broadly similar to Supermemo, Mnemosyne and Anki. The idea is that recently learned items need to be reviewed soon, whereas items well-established in long-term memory can be reviewed at progressively longer intervals. The main difference between Cerebware and the other systems is that Cerebware programs such as the Kanji Sketch Pad assess a user's knowledge whereas most other SRS programs require a user to self-score. In Anki, for instance, you think of the answer and then flip the digital flashcard. You then have to decide if the answer you came up with was right and you have to judge how hard you had to think to recall it. You use this self-inspection to rate your current knowledge of the item. The program factors in your self-rating, the length of time since you last saw the item, and then schedules a future review.

The Kanji Sketch Pad does not need this information because it can deduce it easily from your performance: it watches your errors, your use of the hint function, and even the time taken between starting the first stroke and finishing the kanji. This means you don't need to waste time on the task of thinking about your own thinking: you can just think about the kanji themselves. The Kanji Sketch Pad does allow one form of self-rating, however: you can claim knowledge of an item before you have accumulated enough real evidence for the program itself to realise you know it. This bypasses some of the earlier revision stages, and is particularly useful for items that have been learned outside the program.

The appropriate revision time cannot be known with certainty but the estimate is based on a dynamic memory model that keeps track of how well you have performed on this particular item in the past, and how quickly you tend to forget items of similar difficulty. The model tries to present each item for revision when it has a 90% chance of being remembered, because this represents a nearly optimal compromise between the efficiency gains of moving onto new items (instead of wasting time revising items already well known), versus the efficiency losses of waiting just a little too long (leading to wasted time as the item is relearned). The process is self correcting because if the review intervals get longer than the optimal, your recall will fall below 90% and the program will modify the model. The memory model that emerges from this process is different for every user, and each kanji ends up with its own difficulty rating and review schedule.

Fans of other spaced repetition systems that rely on self-scoring sometimes point out that they can review 100 kanji in a few minutes, just by clicking on one button to show the question side of the card, another to flip it, and another to declare that they know the item very well, well , a little bit or not at all (the third step often triggers the next card, so only two clicks are needed). Although this is indeed a quick way to get through 100 reviews, especially compared to the Sketch Pad that requires each kanji to be drawn, it silently communicates to your subconscious memory centres that each kanji is only worth a few seconds of attention. The fact is, we cannot possibly remember everything, so the brain has a triage system for memory, rating everything according to how important is to remember. The requirement to actually draw the kanji before moving on to the next item makes the scoring process more rigorous and more interactive, and sends all the right cues to the triage system. You cannot simply give up and give yourself a low score, you will keep going until you finish the kanji. You may even find that you suddenly remember the kanji partway through, possibly with the help of hints, and the memory retrieval system can then take that nearly operational recall pathway and tweak it. This is potentially much more effective than the give-up-and-score route.

Other Differences in Cerebware's SRS Philosophy

Unlike most SRS programs, Cerebware deliberately uses short-revision intervals, at the 2-10 minute scale, for items that are very new. In other words, it recognises that brute repetitive exposure is an important factor in the initial acquisition of knowledge, and that the value of SRS really kicks in later. But note that the repetition is much more active than what used to count as 'rote learning' ; what you are repeating is the retrieval process, not the storage process. Filing something in memory five times in a row is fairly pointless. Practising its retrieval five times in a row strengthens the synapses that will ultimtely be needed to recall the item.

Note the gradual progression as an item moves from new to known, even before the SRS kicks in:
On first exposure, see the full package: a mnemonic, the kanji in its printed font form, the English keyword and the animated hand-drawn form.
On second exposure, see the kanji, its keyword and the animation but not the mnemonic. Try to recall the mnemonic even if it seems superfluous at this stage because the shape is still fresh in your mind; it will be needed more later as the recency effects fade.
On subsequent exposures, see the keyword and try to recall everything else, asking for hints if necessary (the starting point for each stroke will appear as a blue dot). The mnemonic will reappear when you finish the kanji if you had trouble recalling it. Take the chance to edit the mnemonic if it failed you.

The philosophy expressed here has not been formally tested in competition with other systems, but it would make a worthwhile research project. Teachers interested in assessing these techniques quantitatively should contact me via the Cerebware website.

In the meantime, enjoy the beauty of this ancient writing system.

Index
Part One - Overview
Part Two - The Main Buttons
Part Three - The Mnemonic Buttons
Part Four - The Text Editor
Part Five - The Pad's Hot Zones
Part Six - Learning Targets
Part Seven - Usage and Readings