Scientific conquest of the forgetful memory

SuperMemo: a finalist of Software for Europe 1992 Andrzej Horodenski, Computer World, Feb. 17, 1992 Translated and updated: January 1997

New article by Dr. Piotr Wozniak: The true history of spaced repetition

This is a translation of the first ever wide-circulation article written about the SuperMemo method. It was written by Andrzej Horodenski, Poland, and appeared in the Polish edition of Computer World on February 17, 1992. This article initiated commercial sales of SuperMemo by arousing the first cautious interest of the public. Translator's annotations are placed in square brackets

Laziness is the mother of progress. How to learn in order to get good results and not to overwork at the same time? This question pervaded the mind of Piotr Wozniak, from Adam Mickiewicz University of Poznan, who in 1982, as a student of molecular biology, commenced his research on mechanisms of memory and forgetting.

The outcome of his least of all lazy work, lasting three years, was a phenomenological model of human memory, which recovered statistical properties of the process of memory formation and forgetting. Conclusions coming from the model provided guidelines making possible to define optimum repetition scheduling that could be used to minimize learning time. Consequently, a set of learning principles was developed and named by their author as the SuperMemo method. By sticking to the principles of SuperMemo one can see an up to 50-fold increase in the speed of learning as compared with traditional methods.

Repetitio mater studiorum est - repetition is the basis of learning. True, says Piotr Wozniak joining the over 2000-year-old discussion on the learning methodology. It is not, however, unimportant in what intervals repetitions take place. The effect of memory formation is the greatest if the inter-repetition interval is as long as it is only possible. In SuperMemo, the repetition occurs at the moment when a defined proportion of the material is likely to be forgotten (10% by default)

The success in learning is determined by proper choice of intervals between repetitions. That is the basic thesis of SuperMemo.

Optimal intervals between repetitions are computed in order to satisfy two opposing criteria:

  • to maximize retention, i.e., the proportion of remembered knowledge
  • to minimize the learning time, i.e., the number of repetitions

In other words, intervals between repetitions should be long for the sake of minimizing the learning time, and short enough to make sure that the learned knowledge is not forgotten.

As it can be demonstrated, the optimized learning process may closely approach the maximum speed with which human brain can form memory traces. This can be done with the retention level determined arbitrarily by the student (usually between 90% and 99%). Thereby, using SuperMemo, the student may learn as fast as his brain makes it possible. Moreover, he or she can choose between fast and superficial as opposed to slow and in-depth learning.

To compute optimum intervals between repetitions, SuperMemo uses a sophisticated algorithm. Intervals are different for different students, for different repetition numbers, and for pieces of knowledge with different difficulty. In other words, intervals are adapted to individual learning capability of the student and the character of the mastered material.

The SuperMemo method may be used in all forms of learning from touch-typing, through playing musical instruments, through textbook learning, to performing well in intelligence tests. Knowledge or skills that are to be mastered should be formulated in the impulse-response form, i.e., usually as pairs of questions and answers. For example, when learning Esperanto:

  • Q: What suffix denotes a noun?
  • A: -O
  • Q: school
  • A: lernejo

or when learning biology:

  • Q: What does DNA stand for?
  • A: deoxyribonucleic acid

Pieces of knowledge, called items, should be well structured, i.e., they should comply with knowledge structuring principles formulated by the SuperMemo method (e.g., minimum information principle, univocality principle, etc.). Partition of knowledge into small pieces makes it possible to determine the difficulty of individual questions. Difficulty, expressed as the so-called A-factor, is determined on the basis of the history of repetitions of the item in question. A-factors are one of the basic elements used to determine the date of the next repetition.

An interesting and important characteristics of SuperMemo is the fact that the learning speed is practically independent of the size of the learned material. This implies, for instance, that the learning time for 100,000 items will be, more or less, 100 times longer than in the case of memorizing 1000 items. In the lifetime perspective, in to neglect the effects of aging, the student will not observe any substantial decrease of his learning power despite accumulation of repetitions from the previously learned material. In a well-structured SuperMemo database, time necessary to learn and retain a typical item is about one minute per lifetime. For retention of 95%, majority of items will be repeated no more than 10 times in the whole life.

Usually, the first repetition takes place in 1-10 days; however, as soon as after 5-6 repetitions, intervals reach the order of months and years. By means of repetition spacing, SuperMemo virtually eliminates the problem of forgetting. Although incidental lapses of memory form an inherent part of the method, all forgotten items are put back into the learning process, and the overall knowledge retention may be kept as high as 99%. It goes without saying that SuperMemo is not inherently a computer product. However, because of the bulk of computation that has to be done after each repetition, the idea of software implementation inevitably springs to mind.

First computer implementation of the method was developed in 1987. In the years 1988-1990, the optimization algorithms of SuperMemo were improved, making it possible to increase the speed of learning yet about twice.

In July 1991, SuperMemo World company was set up with the intention of making the SuperMemo method and software world-wide popular. The present owners of the company are: the author of the method, biologist and computer scientists, Piotr A. Wozniak, computer scientists Krzysztof Biedalak and Tomasz Kuehn, a physicist, Janusz Murakowski, and a mathematician, Marczello Georgiew. At present, SuperMemo World distributes SuperMemo version 2 as shareware, and versions 5 and 6 as commercial products [at the moment of the last update to this text, SuperMemo World had SuperMemo 6 for DOS, SuperMemo 7 for Windows, and a hypermedia authoring tool SuperMemo 98 for Windows all released as freeware].

The latest, sixth implementation of the SuperMemo program, SuperMemo 6 Professional has, for several months, been beta-tested in Poznan scientific circles. Its first success SuperMemo 6 had scored long before its first commercial copy was sold. In November 1991, the program was chosen a Polish finalist of the Software for Europe 1992 Award Competition.

In present implementations, knowledge is represented exclusively in the textual form; however, this is absolutely sufficient to obtain spectacular results in many learning subjects [SuperMemo 98 and later use hypermedia].

Practically, any information that can be represented in the question-answer form may be used in learning aided by SuperMemo. The learning speed is several times greater than when using any other traditional memorization technique.

Work with SuperMemo is straightforward. The student may use existing ready-made collections of items or he may develop his own collection. All operations related to determining dates and order of repetitions are done automatically so that the student may focus his attention on questions asked by the computer, and on assessing the quality of his responses in 0-5 grade point scale. Grades produced by the student are used in repetition spacing. The grade scale is clearly defined and, after all, indulgent self-assessment for the sake of self-satisfaction makes no sense at all.

Inquisitive individuals may use the program to trace statistical parameters of their progress. Statistical menu options used in supervising the learning process include among others:

  • forgetting curve graphs characteristic for a given student and given knowledge
  • distribution of repetitions in time (i.e., number of repetitions assigned to particular days, months and years)
  • distribution of items with respect to their difficulty [A-factor distribution]
  • approximation of optimum intervals which depend on the student, item difficulty, and repetition number
  • parameters of the learning process (total number of items in the database, number of memorized items, number of items scheduled for repetition on the present day, average difficulty of items, average interval between repetitions, average number of repetitions per item, average response time, average daily working time, average daily number of repetitions, total response time in the current learning session, proportion of remembered items, etc.)

Until now, the most popular and spectacular application of SuperMemo appears to be learning English vocabulary. SuperMemo World developed a collection containing 3000 basic words and phrases of contemporary English. The selection of material was based on the word frequency analysis of standard texts. This vocabulary is considered sufficient to use monolingual dictionaries of English. An average student, working 30 minutes a day, will need about 30 days to master the 3000-item database [currently Advanced English is available with 40,000 items].

An important difference between SuperMemo and other methods of speed-learning, e.g., based on mnemonic techniques, is that the once acquired knowledge is constantly maintained in the student's memory.

For a knowledge hungry student with some background in grammar, the 3000 item database may be mastered in just 3-6 days, thus allowing to develop a basic reading capability in less than a week!

SuperMemo World established the SuperMemo Database Bank, a collection of learning material created by users of the program. Most of the existing databases is distributed free (only costs of postage and packaging are charged). Majority of databases was created for the purpose of learning English. Other subjects include foreign languages, medicine, molecular biology, physics, mathematics, general knowledge, computer science, etc.

See also: