Visual strength as the constraint condition in artificial grammar learning

Visual strength as the constraint condition in artificial grammar learning Daisuke Tanaka (tanaka@rstu.jp) Faculty of Regional Sciences, Tottori University Koyama-cho-minami 4-101, Tottori City, Tottori 680-8551, Japan Sachiko Kiyokawa (kiyo@isc.chubu.ac.jp) Department of Psychology, Chubu University, Matsumoto-cho 1200 Kasugai City, Aichi 487-8501, Japan Ayumi Yamada (ayumi.yamada@gmail.com) Human Innovation Research Center, Aoyama Gakuin University, Shibuya 4-4-25, Shibuya-ku, Tokyo 150-8366 Japan Zoltan Dienes (dienes@sussex.ac.uk) Department of Psychology, School of Life Sciences, University of Sussex, Falmer, Brighton BN1 9QH,UK Kazuo Shigemasu (kshige@main.teikyo-u.ac.jp) Department of Psychology, Faculty of Letters, Teikyo University, Ootsuka 359, Hachioji City, Tokyo 192-0395, Japan Abstract Previous research using the chain of compound letters as letter strings in artificial grammar learning (AGL) suggested that visual saliency known as global precedence influenced the extent of learning. In this study the luminance of letter strings in the learning phase was manipulated to investigate the effect of visual input on AGL regardless of top down attention control. As a result, participants assigned to the low luminance condition were not able to learn any grammar even though they could percept letter strings in the learning phase. This finding suggested that AGL is influenced by the visual saliency from outer environment independently of top down attention control. The results implied that AGL mechanism as adaptive system is affected both by the top down selective attention to acquire covariance sensitively and by the bottom up visual saliency from the complex environment rather than automatic processing system. Keywords: Implicit learning; artificial grammar learning; selective attention Background Implicit learning is a generic term used to refer to the phenomena that observers can implicitly identify the covariation between some variables when exposed to large amounts of information in order to adapt to their environment (Reber, 1989). Artificial grammar learning (AGL) is known as the one of the most popular experimental procedures in the realm of implicit learning research. The typical procedure of the AGL experiment comprises two phases. In the initial phase of the AGL procedure, i.e., the learning phase, participants are exposed to a series of letter strings that follow complex rules, typically a finite-state Markovian rule system (Figure 1). Figure 1: Artificial grammars used in this study. The illustration on the top of the panel represents Grammar 1, and Grammar 2 on the bottom. These can generate “grammatical” letter strings connecting letters from state 1(S1) to outputs through some states (for example, grammatical NVJTVT was generated from Grammar 1, S1- S2-S3-S1-S2-S5). Grammar 1 was the same as that used in Knowlton and Squire (1996), in terms of the abstract structure. The two grammars did not have any common chunks even in the abstract level.