Learning Principles
Overview
The Cerebware Vocab Tester employs five learning principles: spaced repetition, contextual embedding, mnemonic associations, memory-retrieval hints, and long-term learning goals. None of these learning principles is revolutionary - some have been in use for thousands of years. Nor is Cerebware the first learning software to bring these ideas into the computer age. Some other vocabulary learning tools already employ one or two of these techniques. For instance, "200 words a day" uses mnemonic associations, VTrain, Supermemo and 200WAD use spaced repetition, and VTrain also gives some feedback about wrong answers - and every human teacher uses some combination of all five. The final version of Cerebware will be unique in that it will offer all 5 learning techniques in an integrated learning package, at a price that is a tiny fraction of a regular commercial language program.
What follows is a look under the hood of the Cerebware learning engine.
#1 - Spaced Repetition
Imagine that you have just been exposed to a new fact: that the German phrase "die Gießkanne" means "the watering can". If you are very lucky, you might remember that fact forever - but, in normal circumstances, the chances of long-term recall after a single exposure are low. (The chances are somewhat higher because you are meeting the fact here, in an unsual context rather than as one random item in a long list).
If you revised that fact every day before breakfast for the next six months then - obviously - you would probably remember it for the rest of your life. How long would it take to do each repetition? About ten seconds. So, long term recall is very achievable, with a small investment per day, for a single item. But what if your goal is to remember 5000 German words or phrases by the end of the year? (And this is the goal Cerebware recommends for serious language learners). If you revise the words once a day, you will need to spend 50,000 seconds on revision per day by the end of the year. That's 833 minutes, or about 14 hours per day. It's physically possible to spend that much time on revision, but you would have to put aside all other interests, and it would take a dedication few of us possess. And we all know that such intensive revision is unecessary - many of the words would be known well before the year was up. Revising them daily would become a boring chore, providing no intellectual challenge and no interest.
Considering the other extreme, what would happen if the items were revised just once per month? Or once every two months? This approach would reduce the amount of boring, wasted time spent on reviewing items that were already known, but it would raise new problems. After a one-month delay, items that had been seen only once would seem almost new. The memory benefit that is usually obtained from previous exposures, that "uh-huh" recognition feeling, would almost be lost because the items had been left too long before they were revised. There would be little chance of separate revision sessions reinforcing each other, and the student using this approach would find every session effortful and unrewarding.
Between these two extremes, there must be an optimal revision strategy, and researchers have been trying to establish what that revision strategy is. It has been recognised for many decades that the best time to review an item is just before it is forgotten. If we had a brain sensor that could tell when "die Gießkanne" was just about to slip out of your accessible memory, then that would be the perfect time to ask you: "What does 'die Gießkanne' mean?" In recalling the fact, you would reinforce the knowledge and probably feel a tiny flush of pride or satisfaction. Because the memory had been reinforced, the forgetting process would be a little slower the next time around. If we asked you earlier, we would be giving you an easier task, and although this would still boost your recall of the item, it would be inefficient because any time spent reviewing one item is time taken away from another item.
What this means is that the optimal time to review an item changes as the item passes from short term memory, to intermediate stages of memory, to permanent recall. It is not possible to scan the brain and predict when an item is just about to be forgotten, but it is easy to monitor a student's progress and calculate the time at which 90% recall is likely. The chance of recalling an item falls over time, and eventually the item becomes unobtainable. Studies have shown that this is the optimal time for recall - just when an item is about 90% likely to be recalled - we call it the 90% rule. Test sooner, and a lot of time is wasted on easy, boring revision. Test much later, and the revision items seem foreign, difficult, and the benefits of repetition fade away.
Cerebware incorporates this idea in two ways. Firstly, it keeps track of each students success rate at each of the various confirmation levels, as an item passes from 'unknown', to 'known' to 'confirmed in intermediate memory' and up through various levels to 'mastered'. Each revision is scheduled at a specific interval: within one day of learning the item, within a week, within a month, and so on. The software adjusts these global revision intervals to achieve a target of 90% success. If, for instance, the default one-week revision interval was routinely met with a 95% success rate, this would indicate that the student could better divert that revision time to learning new items. The one-week interval would be stretched to 8 days, or longer if needed. If, on the other hand, the one-week interval was too long, and the student recalled only 80% of items, the interval could be shortened to 6 days.
If you've been following the discussion to this point, you probably have an objection on your lips: but, but, what about difficult items? Some items are intrinsically more difficult than others, and are more likely to be forgotten. The appropriate time to revise these difficult items would be earlier than for easier items.
Cerebware copes with this variability by keeping track of each user's success rate with each item. An item that has been recalled with 100% accuracy every time it is met gets the standard revision interval, but items that have been more difficult get a proportionately shorter interval. The appropriate shortening factor is monitored and continually adjusted (there is, in fact, an array of shortening factors, one for each confirmation level). Also, some attempt is made to calculate an item's intrinsic difficulty. This is somewhat subjective but it seems easier for most people to remember that "trinken" means "to drink" than "gießen" means "to pour". The former is a close cognate of the English verb; the latter seems arbitrary. By applying some simple language-transform rules, the Cerebware software calculates the intrisic difficulty of each item, and uses this to shorten or lengthen the initial revision intervals, before it gets enough data to know how difficult the item really is for the student.
There is one important caveat in applying this principle. It is possible that, following the 90% rule, the Cerebware software might decide that a hundred or more items are all overdue for revision when a student logs on. Before it subjects the student to a long revision session, the program factors in one more important piece of information. How well is the student going today? If the student sat down to revise a hundred old items, and saw no new items, the student's overall vocab score could only possibly shrink (because forgotten items would be recognised as such), and could not grow. This might be demoralising. There is no point in following the optimal program for each item individually if the overall learning experience becomes too hard, as a result. The software therefore monitors each learning session and creates a mix of revision items and new items, of easy items and difficult ones, expanding the "unkown" pool as needed. When it does need a difficult item, it chooses one that is also overdue for revision, but morale is made subordinate to the 90% rule. Students who prefer to get all their revision out of the way before learning new items can do a 'Confirmation session'that focuses on 'stale items' - items that are overdue for revision.
#2 Contextual Embedding
There is good evidence that facts are more easily recalled if they are encountered in some context, rather than as isolated pieces of trivia. That is, the more meaning a fact has, and the more ways in which it links to other areas of knowledge, the better it will be remembered. In deciding how to store a memory, the brain probably prioritises facts according to how likely they are to be needed again, and one measure of this is how many cross links there are between that fact and the rest of the knowledge base. This is directly analogous to the algorithm that Google might use in deciding where to rank a web page. Not only is the number of hits important (and that's where spaced repetition comes in), but also the number of other sites linking to a web page.
The type of linkages is also important. The brain has distinct mechanisms and structures for storing autobiographical, episodic memories, and those mechanisms are different to those used for storing impersonal facts. An item is more likely to be remembered if it is represented in both of these memory subsystems. Ideally, then, every vocabulary item should be tied to a story - preferably a true story involving the learner but, failing that, a fictional story that manages to stir up some sense of autobiographical involvement. Reading fiction and imagining yourself to be in the scene activates your autobiographical memory systems, to some extent, and most of us can remember novels and movies better than dry, dislocated facts.
Cerebware encourages language learners to link their vocabulary-learning efforts to reading a novel, because this gives each word a context. An added benefit is that users get to see how words usually combine, which is the first step in developing a feel for the language. That's why the Cerebware Vocab Tester can be set up to introduce words in groups, where every word belongs to a new sentence. The foreign sentence and its English translation are shown side by side, along with a list of the relevant new words and their definitions. The sentence is also spoken aloud - if you have obtained the relevant audiobook. For help in setting this up, contact Cerebware.
Here is an example of how a sentence might be introduced by the Vocab Tester:
#3 Mnemonics
Mnemonics are so important that they have been given their own help file - see 'Mnemonics'.
#4 Memory retrieval hints
Cerebware adopts the philosophy that it is better to recall an item of knowledge with the help of a hint rather than give up and be told the answer.
The range of hints offered by the program is discussed in a separate help file, 'Hints and feedback'.
#5 Long term learning goals
A lot of people like the idea of learning a language, and are prepared to put in the necessary work, but get distracted from their learning efforts after they have learned the first 200-500 words. It is all too easy for good intentions to be pushed aside during busy periods - for weeks and months to slip by with only only scant attention paid to the mammoth task of acquiring a foreign vocabulary. By the time such students get back to their vocab drills, they have forgotten a lot of material and they are not even sure which material needs to be revised most urgently to repair the damage. This is a common stage for students to give up, just when the rewards were within reach.
Cerebware can't force anyone to learn, but it can provide a powerful motivator in the form of an explicit learning target. Choose an achievable target of, say, 50 new words per week. It sounds simple, but achieving that target requires net forward progress during a complex evolution of vocabulary knowledge: some words will be met for the first time each week, some will pass from 'unknown' to 'known', some words will be consolidated, and some words will slip into the temporarily forgotten pile. Cerebware will keep track of whether you are sticking to this composite target, factor in how many words you are likely to have forgotten since the last session, which ones are most urgently in need of review, and (if necessary) tell you how much work you need to do each week to catch up. The program will even graph your overall progress visually, showing your increasing knowledge. For more details, see 'Setting targets' and 'The progress graph'.