Building knowledge requires bricks, not sand
Monday, July 13, 2015
Here is a neat finding, originally reported by Gillian Cohen in 1990. It is easier to remember that someone is a baker than that someone’s name is Baker. Although both memories seem to require making a connection between the person’s appearance and the very same word “baker,” the profession baker is meaningful, evoking rich imagery and other associations. The name “Baker” is (at least nowadays) completely arbitrary; a random fact disconnected from other knowledge.
The idea that meaningful material is more easily remembered than meaningless material, is far from new, of course. The way people can memorize thousands of digits of pi, or a full deck of playing cards in under half a minute, is by using methods (some of which were invented millennia ago) that transform arbitrary lists and facts into rich and meaningful stories.
And yet, when psychologists started to study memory in earnest, they tended to view the role of knowledge and meaningfulness as a nuisance factor. For example, in one of the first empirical investigations of memory, the psychologist Hermann Ebbinghaus used nonsense syllables (MEZ, VOK, NAD, GOF) that were novel to his subjects (though he mostly tested himself) so that he could study memory that was unaffected by prior knowledge.
As hard as Ebbinghaus tried to get away from people’s reliance on prior knowledge, even seemingly meaningless syllables like MEZ depend on our knowledge of letters (this is why we see MEZ as three letters rather than eleven line segments) and of the sounds they make (this is what lets us think of MEZ as a single syllable rather than as three letters). And so, when seventy years or so later, George Miller reported that people’s ability to remember arbitrary materials was in the range of 7 +/-2, that “magical number” was expressed in the number of chunks where a chunk corresponded to some basic unit of meaning—whatever that might be.
Two recently published studies in the Psychonomic Bulletin & Review offer a fresh examination of the role of knowledge and chunking in memory.
The first study, by Kaiser, Stein, and Peelen begins by noting that traditional memory research, perhaps under the lasting influence of Ebbinghaus, has focused on very simplistic stimuli which tend not to reflect the kinds of relationships that our memory systems may be relying on in more real-world situations. The authors showed participants pictures like these:
The pairs on the left show combinations that accord with real-world experience: an umbrella above a beach-chair; a mirror above a sink. The two pairs on the right reverse those relationships.
The task the participants performed (shown below) was simple enough: View two displays containing two (or three) pairs like the ones above, and determine whether any object(s) was replaced between the first and second display. In the example above, the mirror/sink combination is substituted by a slightly different mirror and sink between the first and second display, and so the answer to whether there was a change is “yes.”
The prediction was that people would be able to perform more accurately when the pairs reflected real-world arrangements of objects. This is precisely what the authors found. To rule out the possibility that the regular-object advantage was due to easier verbal labeling of the regular pair (with such labeling serving as a kind of mnemonic device), participants had to rehearse a series of numbers while performing the main task which discourages labeling the individual images. An image of a mirror above a sink is more meaningful—a better chunk—and so easier to hold in memory.
The second study, by Reder, Liu, Keinath, and Popov, examined the question of whether a “chunk is a chunk”, or if some chunks are better than others. Specifically, the authors ask whether seeing new visual symbols more frequently allowed the symbols to become stronger chunks such that—and here’s the most interesting part—they would consume fewer memory resources, thus enabling people to better remember novel combinations of the symbols and to associate them with English words.
Participants, who had no prior experience with Chinese characters, were familiarized with 64 Chinese characters over four three-hour long sessions that spanned several weeks. During each session, participants performed a simple visual search task shown in the top row of the figure below. This task served to visually familiarize people with all the characters. The critical manipulation was that half of the characters were randomly selected to be shown more often (20 times more often) than the other characters. Each participant therefore received much more exposure to some (“high-frequency”) characters than other (“low-frequency”) ones.
In each session, participants also completed a memory test requiring them to learn an association between a pair of characters and an arbitrary English word (see the second row in the figure above). Each pair of characters was completely new, and was composed either of the low-frequency or the high-frequency characters. The question posed by the authors was whether the high-frequency characters, being better established in memory, would make for better chunks for forming novel combinations.
The results showed that the characters that were seen more often became easier to find in the visual search task (that’s just practice making perfect). But also, novel combinations made from the more familiar characters were easier to learn to associate with new information than combinations made from the less familiar characters.
As a further test of the idea that the characters that became more familiar due to higher frequency were easier to represent in memory, the authors had participants complete a popular test of working-memory (N-back) with both the high- and low-frequency characters.
The results showed improved performance for the more familiar high-frequency characters, suggesting that these “better” chunks were less taxing to remember. The authors’ conclusion is well-captured by their article’s title: “Building knowledge requires bricks, not sand…” As elements become more familiar, they become stronger chunks, (“bricks”) that allow for successful construction. Trying to construct a structure from the weaker chunks, on the other hand, is like trying to build from sand.
These findings remind us of the dynamic nature of memory. The meaning-making machines that are our brains are constantly searching and forming new bits of meaning, new chunks.