What makes learning, storage and retrieval more efficient?

Bjork, R. A., Dunlosky, J., & Kornell, N. (2013). Self-regulated learning: Beliefs, techniques, and illusions. Annual Review of Psychology, 64, 417-444.‏ http://cognitrn.psych.indiana.edu/rgoldsto/courses/cogscilearning/bjorkdunlosky.pdf

Here are insights from this interesting paper (with a few additions by me):

We do not store information in our long-term memories by making any kind of literal recording of that information, but, instead, we do so by relating new information to what we already know. We store new information in terms of its meaning to us, as defined by its relationships and semantic associations to information that already exists in our memories.  Storing information in human memory appears to create capacity—that is, opportunities for additional linkages and storage—rather than use it up.   I'll add, that when a piece of information is tied to a broader net of ideas and concepts, it is learned better.  This is why people with a broad knowledge base and good fluid abilities (that help them conceptualize links between new and old pieces of information), have an advantage in learning.

When we retrieve information from memory, we do not retrieve an exact "recording" of this information as it existed when it was learned.  We re-edit the memory.  When we recall an event, for example, we integrate features of the event with our assumptions, goals or experiences.  I'll add, that the manipulations we perform on the information we retrieve, like telling it or writing it down, also cause it to be re-edited and then later re-stored in memory as slightly different.  The context in which we retrieve the information also changes it and adds aspects that were not in the information at the time it was first stored.

The mere act of retrieval strengthens the information retrieved, so it becomes easier to retrieve it later.  Thus, when we study for a test, it's better to test ourselves and to retrieve the information from memory than to re-read it time and again.  The mere act of retrieval of specific information weakens our ability to retrieve similar competing information, which is tied to the same retrieval cues or semantic associations.  This phenomenon is called "retrieval induced forgetting".

We learn and retrieve information better when we introduce difficulties and challenges to the learning process.  Self testing (as opposed to passive rehearsals of the information) is an example of such a "desirable difficulty".  Self testing not only makes sure that we studied the information.  It is a part of the learning process itself.  When we answer a question on a self test, we reorganize the material and elaborate it.

Meta-cognitive monitoring and control are important parts of the leaning process.  During learning we evaluate the situation and make decisions, like what to learn next, how to learn it, whether we've learned enough for the information to be retrieved, whether we retrieved the information correctly, and so on.

It's recommended to space learning sessions across days, rather than to cram it in one session.  It's better to study two or three subjects simultaneously and to alternate between them ("interleaving"), than to study only one subject in depth.  When we study one subject in depth, processing is easier and more fluent than when we alternate between two subjects .  This is because when we massively study one subject, we hold the same concepts in working memory.  When we alternate between subjects, we need to re-retrieve the concepts of each subject that we alternate to from long term memory to short term memory.  The act of retrieval helps to learn them better.

In an experiment, participants were asked to answer trivia questions. The participants were told that they would be tested later, and the nature of the test was made very clear: They would be given a blank sheet of paper and, without being asked the questions again, they would be asked to free-recall the answers. During the question answering phase, after each correct answer, the participants were asked to predict the probability that they would be able to recall the answer again on the final test. The results were surprising: The more confident participants were that they would recall an answer, the less likely they were to recall it. This outcome occurred because participants predicted they would be able to recall the answer to questions that they answered quickly, but they were most likely to free-recall answers that they had thought about for a long time.  This means that efforts to process the questions and look for the answers in memory strengthened learning and increased the likelihood of retrieving the information later.  

In another study, researchers presented to participants paintings by 12 different artists. Some artists’ paintings were presented on consecutive trials while other artists’ paintings were presented interleaved with other paintings. As a test, participants were asked which artist painted each of a set of previously un-presented paintings. They were more accurate following interleaved (i.e., spaced) learning than following blocked (i.e., massed) learning. Blocking may have made it easier to notice similarities within a given artist’s paintings, whereas the value of interleaving appears to lie, at least in part, in highlighting differences between categories.  The comparison process causes the mental representation to be more abstract, with conceptualization of the categories of the differences and the relations between them. 

In the domain of motor skills there is substantial support for the idea that interleaving practice on separate skills to be learned, such as the several strokes in tennis, requires that motor programs corresponding to those skills be repeatedly reloaded, rather than executed over and over again, which has learning benefits.

The authors recommend designing textbooks that don’t mass one topic at a time but periodically return to prior topics in an effort to promote spaced learning.

What about mistakes and errors made during learning?  Anticipating, unsuccessfully, a to-be-learned response can enhance learning.  If participants are asked to predict what associate of a given cue word (e.g., Whale) is to be learned before they are shown the actual to-be-learned target (e.g., Mammal), their later cued recall (e.g., Whale: __?__) of the target word is enhanced, versus a pure study condition (e.g., Whale: Mammal), even when the predicted associate  differs from the target associate.

Making errors is often an essential component of efficient learning.  Manipulations that eliminate errors can often eliminate learning. Thus, for example, when retrieval of to-be-learned information is made so easy as to insure success, the benefits of such retrieval as a learning event tend to be mostly or entirely eliminated.  Introducing desirable difficulties into learning procedures, for example, such as variation or interleaving, tends to result in more errors being made during the acquisition process, but it also tends to enhance long-term retention and transfer.   Making  errors appears to create opportunities for learning and, surprisingly, that seems particularly true when errors are made with high confidence.  Feedback was especially effective when it followed errors made with high confidence versus errors made with low confidence.