Using Testing to Enhance Learning: A Comparison of Two Hypotheses

Students learning facts such as foreign language vocabulary often rely on a self-testing procedure in which they cue themselves with the English word and try to recall the foreign language target, instead of simply memorizing cue-target pairs. The value of this strategy has been empirically verified by a long history of research, yet existing computational models of human learning do not address the enhancing-learning-through-testing phenomenon. Using a simple, well studied model—a feedforward neural net with no hidden units—we propose two different hypotheses for characterizing the phenomenon. Hypothesis 1 is that self-testing generates a target which is used for additional training. Hypothesis 2 is that self-testing produces a more reliable error signal for training than rote memorization. Through simulation studies, we find that hypothesis 2 readily explains the phenomenon whereas hypothesis 1 does not. Further, hypothesis 2 makes predictions worthy of further empirical study, and can be viewed as a natural consequence of temporal difference learning. When learning foreign language vocabulary and other facts, students often study using index cards that have an English vocabulary word (or cue) on one side and a foreign language vocabulary word (or target) on the other. The intuition is that by testing oneself, the associations are better learned and retained. This intuition has been supported by a long history of empirical demonstrations (e.g., Izawa, 1966; Young, 1971). For example, Bartlett and Tulving (1974) asked participants to learn a list of paired associates (the study phase), and later tested retention of the pairs using free recall or recognition (the final test). Before the final test, subjects were given a cued-recall test (a self test) of some of the paired associates. Retention was better on the final test for those items that received the self test. In this paradigm, it is unclear whether the benefit of the self test is attributable to attempting retrieval per se, or to the fact that successful retrieval of an associate also results in a representation of the pair—an additional training trial. An obvious strategy for examining the effect of retrieval is to conduct and experiment with, in addition to the initial study phase and the final test, an intervening phase in which participants are given either a self test or an experiment-provided representation of the paired associate (which we'll refer to as study only). In this paradigm, the outcome is ambiguous (Carrier & Pashler, 1992): self testing outperforms study in some …