Psychology improves learning: It’s still boring, but your test results will be higher

Let’s say that you are helping your niece with the French vocabulary that she has to learn for school. She has an important test in two days and has already learned quite a lot of words. You ask her the translation of “fromage” and she, without a bit of doubt, says “cheese”, then you ask her “maison” and after some thinking, she says that she doesn’t know the answer, which you provide to her. Clearly, your niece still needs to study the meaning of “maison”. Intuitively, you will probably wait for a longer period of time before asking her again about the translation of “fromage” and will ask her quickly for the translation of “maison”.

“The system, in all its simplicity, works remarkably well”

If one looks at online learning systems that help with learning factual materials, they typically use this type of data — correct or incorrect — to decide which word to present you on the next trial. If a word was answered correctly, it will be put at the back of the stack, only to return if you have seen all other words, but if a word was answered incorrectly, the system will show you the correct translation, and revisit the word after, a couple of other words have been presented. This system, in all its simplicity, works remarkably well. The reason why it works so well is that it adheres to the principles of the testing and spacing effects. The spacing effect is that the more time there is between repeated presentations of an item, the better that item will be stored in memory. The testing effect, on the other hand, reflects the fact that one of the best ways to learn factual material is to test yourself to see if you actually know that item. Related to this is that learning has shown to be optimal if, during testing, you still know the item. If we return to the “fromage”/”maison” example, it’s clear that you are helping your niece to learn using the testing method, and that, by waiting longer before revisiting “maison” than “fromage”, you are increasing the changes of her still knowing both items at the next presentation, but you’re also adhering to the spacing principle.

Although these computerized learning systems can help in learning, you are probably a better tutor for your niece than most computerized learning systems. That is, let’s assume that you now ask for the translation of “saucisse” and your niece takes a long time, but eventually comes up with the right answer. You immediately pick up on that, realize that “saucisse” is encoded, but not that well, and will repeat this item quite soon. Most computerized learning systems will encode this item as “correct”, and put “saucisse” on the end of the stack, just like “fromage”

“it is difficult to determine how to translate thinking time into how well the learner knows an item

The reason why most learning systems are only using correct and incorrect answers, is that it is difficult to determine how to translate “thinking time” into “how well the learner knows an item”, a comparison which needs to be made so that the system can determine how long it can wait before presenting the item again. A couple of years ago, together with Master students from Experimental Psychology and Artificial Intelligence, my group started exploring if we could use modern memory theories to help in this translation. The idea is pretty straightforward: any time someone learns something new, that item is stored in memory and initially has a very high “activity” (a word used to describe how easily available something is in memory), but after the very first learning trial, that activity will quickly decrease until the item cannot be retrieved anymore. For every subsequent presentation, there will be another temporary increase in activation, but each time the decrease will be slower. Although the activation itself cannot be measured, activation can be translated in how long it will take the simulated memory system to retrieve that item. Therefore, we can measure the time it takes your niece to answer an item, and figure out how “active” that item is. If we know how active an item is, and if we know how long ago we presented that item previously, we can calculate how quickly your niece forgets the item. By calculating this person-specific forgetting rate, our system, called “SlimStampen“, can present personalized learning to anyone aiming for optimal fact-learning. Obviously, the proof of the pudding is in the eating, and we have tested this system in the lab, in an experiment during which university students had to learn factual information that sometimes was (materials for a course), but sometimes wasn’t relevant to them (such as Swahili-English translations), but also in school settings. In all these settings, SlimStampen outperformed the correct/incorrect based systems, sometimes by improving the eventual grade by a whole grade point (i.e., an 8 instead of a 7, roughly corresponding to a 80% instead of 70% correct score).

Obviously, this is of interest to many different types of learners, and as a start, we have teamed up with one of the largest publishing houses in the Netherlands of secondary school educational materials, Noordhoff Publishers. Starting this school year, all students using the online learning systems of Noordhoff Publishers can use the SlimStampen method. Admittedly, this won’t magically make learning fun, but because of its effectiveness, your niece will have to learn for a shorter period of time, so that the two of you can do more interesting things instead.

Relevant Publications:

van Rijn, D.H., van Maanen, L., & van Woudenberg, M. (2009). Passing the test: Improving learning gains by balancing spacing and testing effects. In Howes, A.,  Peebles, D., & Cooper, R. (Eds.), Proceedings of the Ninth International Conference on Cognitive Modeling (ICCM). Manchester, UK.

 

NOTE: Image from the U.S. National Archives

Hedderik van Rijn is a professor associated with the departments of Experimental Psychology and Statistical Methods and Psychometrics. In one of his lines of work, he focuses on how we can apply psychological theories to improve the interaction between humans and computer systems, or make computer systems more useful for humans. For example, together with his students, Van Rijn has developed an individualized learning system which adapts itself to the learning characteristics of individual learners and which is now integrated in the online learning system of one of the largest Dutch publishers of educational materials. In addition to this more applied line of work, he is also interested in how humans perceive short time intervals, and how these time intervals influence cognitive performance. This work on temporal cognition is based on building computational models of time perception, but also on theoretical and empirical neuroscience studies, and has been published in high ranking journals. But what he loves most about his job, is that it allows him to work with bright students who can provide new, fresh views on decade old problems in cognitive science.


You may also like

One comment

  • Pia January 23, 2015  

    As I understand from your post, we should wait longer to test people for information that they already know, compared to information that they do not know well. You used the example of “fromage” and “maison”, with the girl knowing “fromage” quite well. Accordingly, would we not have to test for “maison” before asking for “fromage” in order to improve learning? I am a bit confused because you state the opposite above.

Leave a comment