We have become experts at remembering pieces of information that share no natural relationships. For example, your friend may have just started a new job, and it is easy for you to remember where she works because the company’s logo and her favorite color are both red. When thinking about remembering two pieces of information, it can be quite easy to focus on the quality of the association between them, but another aspect of associative memory that psychologists are interested in is the direction of these associations in memory.
To help illustrate this idea, imagine an association between two pieces of information, such as a friend’s name and her new workplace, as a bridge that connects the two, like a bridge that crosses over a river to connect different parts of town. Are associations in memory for paired items like a single two-way bridge? Or are associations like one-way bridges that are built separately from each other?
These two types of bridges illustrate the contrasting views of associative directionality. On the one hand, a common interpretation in cognitive psychology – referred to as the associative-symmetry hypothesis – is that studying items paired together creates a new holistic and bi-directional representation between the items themselves (like the two-way bridge). A second interpretation – referred to as the independent-association hypothesis – is that studying pairs of items results in encoding separate representations for forwards and backwards associations in memory (like separate one-way bridges).
So, which hypothesis (if either), the associative-symmetry hypothesis or independent-association hypothesis, is correct?
Vencislav Popov and his colleagues at Carnegie Mellon and the University of Pittsburgh (pictured below) investigated this question by conducting a series of experiments that were described in a recent paper in the Psychonomic Society’s journal Memory & Cognition. Their main idea is: When people encounter a pair of arbitrarily selected words, they create a single memory connection between them, and they use the same connection regardless of the order in which these words are tested later. In contrast, when faced with a pair of semantically-related words, people remember separate forwards and backwards connections between them, and they can strengthen memory for each direction separately and to a different degree. However, testing this idea is anything but easy.
In a 2002 paper published in Memory & Cognition, the author, Kahana suggested that a simple correlation in memory performance when testing forwards (A → B) and backwards (B → A) associations could help elucidate whether associations are symmetrical or not. If the representations for both types of associations were different (as the independent-association hypothesis holds), then a small or zero correlation would exist between memory for forwards- and backwards-associations. But, if the two types of associations were similar to each other (as the associative-symmetry hypothesis holds), then memory performance between the two types of associations would be highly correlated, which is what he found.
Correlations in these types of tasks can obfuscate true relationships in memory, however. For example, interference from successive studying and testing reduced correlations in memory, even when only testing participants on forwards-associations. And this pattern of results was replicated. Therefore, in addition to pitting the associative-symmetry and independent-association hypotheses against each other, a test of word pairs should also account for the complexities of memory performance that can affect comparisons between directional associations.
To get around this problem, instead of using correlations, Popov and colleagues compared performance for both types of associations to a typical learning curve (i.e., increases in memory for items that are studied repeatedly) over alternating blocks of testing memory for forwards and backwards word pairs. They hypothesized that memory for word pairs without pre-existing semantic relationships would show a smooth increase in accuracy successive learning blocks with related decreases in the amount of time it takes to recall a paired-associate, regardless of the direction of the association (in line with the associative-symmetric hypothesis).
In contrast, they hypothesized that memory accuracy for items that have pre-existing semantic relationships would exhibit a “stair-step” pattern, where memory for each of the two associations only increases every other block (in line with the independent-association hypothesis). If related word pairs show this pattern, then the two types of directional associations – forwards (A → B) and backwards (B → A) – are modifiable in memory only during their respective trials. If not, then there is a unitary association between two related words, such that increasing representational strength for one direction also increases strength for the other.
The procedure is shown in the figure below. In the initial learning task, participants studied pairs of related and unrelated words. They then alternated between test cycles, where one cycle would only test forwards-associates (e.g. “MACARONI-?”) and the other would only test backwards-associates (e.g. “CHEESE-?). In Experiment 1, participants studied 80-word pairs and engaged in alternating test cycles until they achieved perfect recall (up to 8 cycles), and in Experiment 2, participants studied 106-word pairs and were tested over 6 alternating test cycles.
The results of both experiments favored the independent association hypothesis. As shown in the panel on the right in the figure below, learning curves for unrelated word pairs were smooth across alternating trials, while learning curves for related words were jagged. Similarly, as shown in the panel on the left in the figure below, response latencies for unrelated items exhibited a smooth, decreasing pattern across alternating trials while latencies for related words exhibited the hypothesized stair-step pattern. Thus, the memory response data show that forwards and backwards associations are separate and modifiable, but only for word pairs that shared a relationship prior to learning and test.
Distinguishing between the two types of directional associations may seem like a theoretical exercise, but the implications of the results may be much more direct than you think. For instance, a German teacher who only rehearses vocabulary from German to English (e.g., “Tschüß” → “Goodbye”) may be instituting an unfair policy in her tests when she gives the English translation as a retrieval cue (e.g., “Goodbye” → “?”). Or perhaps we are unfair when we are amused by a toddler, who has had less time to establish normative word associations, asking for “cheese and macaroni”. Regardless, we will continue to investigate how directions of associations affect language and memory.
Psychonomics article featured in this post:
Popov, V., Zhang, Q., Koch, G. E., Calloway, R. C., & Coutanche, M. N. (2019). Semantic knowledge influences whether novel episodic associations are represented symmetrically or asymmetrically. Memory & Cognition, 47(8), 1567-1581. doi:10.3758/s13421-019-00950-4.
