A Study On The Transfer Effect In Artificial Grammar Learning

The transfer effect in artificial grammar learning refers to such a phenomenon that individual ' s learning in one situation influences his learning and behavior in new situations and contexts. The systematic laboratory research could be traced to Reber' s (1967,1969). After that, the researchers have put forward many hypotheses to account for the phenomenon. However, as far as the mechanism and some basic theories of the transfer effect in artificial grammar learning are concerned, there are still many controversy, in which the chunks information transfer hypothesis (Elman, 1990; Dienes, 1992 ;Cleereman, 1993; Redington&Chater, 1996; Dienes&Altmann&Gao, 1999) and the abstract analogy transfer hypothesis (Brooks &Vokey, 1991, 1992; Whittlesea, 1993; vokey & Higham, 2000)are the most important. At present, researchers attempt to systematically analyze these theories, in which Tunny and Altmann (1995, 1999) and Johnston and Shanks (1999) have concluded that the sequential dependence affects transfer in artificial grammar learning. Yet, facing with so many controversial experiment foundings (Brooks, 1978; Druhan&Mathews, 1989; Perruchet&Pacteau, 1990; Elman, 1990; Brooks&Vokey, 1991, 1992; Dienes&Broadbent&Berry, 1991; Dienes, 1992; Cleeremans, 1993;Whittlesea&Dorken, 1993; Roussel & Mathews , 1994; Reber & Gome & Schvaneveldt, 1994; Reber&Manaz, 1994, 1997; Dienes&Altmann&Goode, 1995; Redington&Chater, 1996; Vokey&Higham, 1999; Dienes&Gao, 1999; Gomez, 2000 ;Tunney&Altmann, 1999,2001), these theories need be further developed in order to explain these results more exactly. The present study is dedicated to improve experimental design under the guidance of the chunks information transfer hypothesis and the abstract analogy hypothesis, through which they hope to provide more evidence and stimulate further research.The present study includes three experiments, using symbols and syllables as materials and graduates as subjects. The material in Experiment 1 was displayed on paper, and the materials in Experiment 2 and 3 were displayed on the computer screen in & standardized form. The hit rate was selected as the dependent variable in all experiments and subjects were asked to indicate how confident they were for each decision. The purposes of the research is to explore the following questions:1. When the surface structures are different between learning and test sequences, whether subjects can learn and transfer sequential dependencies between nonrepeating elements.2. When the surface structures are different between learning and test sequences, how about the effects of the distributional properties in the transfer of dependencies between nonrepeating elements.3. When the surface structures are different between learning and test sequences, what is the basis of the transfer between repeating elements.The experiment results indicated:1. When the surface structures are different between learning and test sequences, subjects can learn and transfer sequential dependencies between nonrepeating elements.2. When the surface structures are different between learning and test sequences, the transfer of dependencies between nonrepeating elements is indeed subject to a match between the distributional properties of learning sequences and grammatical sequences.3. When the surface structures are different between learning and test sequences, the transfer between repeating elements can also be based on the abstract analogy.The results are not completely consistent with previous studies , deeply and systematically , we analyzed possible reasons of the difference, the implicitness of transfer and the implicitness of the representation of transfer knowledge in artificial grammar learning. In conjunction with previous research results, we draw a tentative conclusion that both kinds of mechanisms, the chunks information transfer hypothesis and the abstract analogy hypothesis, are in operation. And they can be dissociated under certain conditions.