Phonological Memory and Rule Learning

Our research reflects the current trend to relate individual differences in second language learning to underlying cognitive processes (e.g., Robinson, 2002). We believe that such investigations, apart from being of practical importance, can also shed light on the cognitive mechanisms underlying the language learning process. Here we focus on the influence of memory, which has long been recognized as an important component of language learning aptitude (J. B. Carroll, 1962; Skehan, 1998). We build on work that has been done in psychology on the relationship between phonological short‐term memory (PSTM) and vocabulary learning and ask whether a similar relationship can be traced at the level of grammar learning. Even within grammar, however, we believe that it is necessary to systematically target different domains, because each will potentially draw on different learning mechanisms and be subject to different variables. In this study we focus on an area of grammar that depends upon distributional analysis, since this has been stressed as an important component of “data‐driven”(Robinson, 1995) or connectionist (Ellis, 1998) approaches to language learning. In these approaches memory for associations between input elements constitutes the database from which generalizations emerge. This approach predicts that PSTM should be related to initial learning of word forms (vocabulary learning), to memory for combinations of familiar forms in the input (memory for input), and to eventual rule learning based on those combinations. In two relatively small‐scale laboratory‐based experiments using semiartificial microlanguages, we tested these predicted relationships. The domain was grammatical gender, which in our microlanguages could only be inferred from the distribution of determiners accompanying nouns. Participants first performed a PSTM test, then learned the experimental vocabulary as isolated words (vocabulary learning), and were then exposed to a selection of the possible determiner‐noun combinations from the microlanguage as part of a simple rote memory task (which in the early stages of training provides a measure of input memory uncontaminated by rule learning). Rule learning was assessed by testing the participants’ ability to generalize to determiner‐noun combinations that had not been presented during training. The predicted pattern of correlations was obtained, confirming a role for PSTM in distributional learning. However, there were also statistically independent effects of the participants’ prior knowledge of languages that encode grammatical gender. This indicates that there was a “conceptually driven” element to the learning process and that both memory and nonmemory factors were related to learning outcomes. With regard to methodology, we hope that this study demonstrates that it is possible to investigate individual differences within small‐scale, but highly controlled, learning environments and that individual differences can be used to explore learning processes, provided that it is clear what kind of learning problem is posed by the targeted structures.