关键词: 语法复杂性, 选择性注意, 人工语法学习, 双人工语法

Abstract: Eitam, Schul, & Hassin, 2009; Eitam et al., 2013; Tanaka, Kiyokawa, Yamada, et al., 2008 demonstrated selective attention is indispensable in artificial grammar learning. Nevertheless, Eitam et al. (2009) pointed out there existed two thoughtful questions. Firstly, due to the complexity of artificial grammar, the learning of selected grammar might exhaust cognitive resource with the result that ignored grammar had little resource to be learned. Secondly, the ignored grammar might have been learnt without being detected in the test phase. Therefore, we designed the complexity of artificial grammar (complex grammar A & B vs. simple grammar C & D) and the types of illegal sequences (illegal sequence a vs. illegal sequence b) to examine both possibilities above. Sequences of inner color and sequences of outer color with different grammars were presented simultaneously. Two groups of participants were instructed to only memorize the sequences of outer color with grammar A and grammar C, respectively; anothers, also respectively, the sequences of inner color with grammar B and grammar D. That means each group of participants need to pay attention to the inner (outer) color of each sequence, and then indicate the color of the inner (outer)-color block that immediately preceded the last stimulus using the response color matrix. All participants were random successively tested on the grammar underlying the selected and the neglected training sequences. They were asked to categorize each sequence which may come from 10 novel legal sequences , 10 illegal sequences a and 10 illegal sequences b as grammatical or not. Results showed that there was a significant Attention Direction×Complexity of Grammar interaction, F(1, 54) = 10.74, p < 0.05, η2 = 0.17. Participants who were tested on selected grammar showed significant difference on correct rate between simple grammar (0.64 ± 0.14) and complex grammar (0.56 ± 0.10), F(1,54) = 6.98, p < 0.05. While testing the ignored grammar, the scores gained from simple grammar (0.47 ± 0.11) showed no significant difference to the scores gained from complex grammar (0.52 ± 0.09), F(1, 54) = 3.84, p = 0.06. This indicated that the learning of the simple grammar rules chosen to be noted have a certain grammatical advantage over the learning of selected complex grammatical rules chosen to be noted. What′s more, no matter the illegal sequence was a or b, there was no reliable difference between probability level and the scores of the grammatical rules that had been neglected. It meant that even if the complexity of the grammar chosen to be noted was reduced, the ignored grammar could not be successfully learned. Participants who classified sequences based on selected grammar showed reliable learning as compared to chance, but participants who were tested on the ignored grammar were not. Moreover, a mixed-model analysis of variance revealed a significant Attention Direction×Ungrammatical Sequence interaction, F(1, 54) = 5.80, p < 0.05, η2 = 0.096. This analysis revealed that there was no significant difference for participants who classified sequences based on selected grammar between the condition of illegal sequences a and the condition of illegal sequences b, F(1, 54) = 3.51, p = 0.07, so did participants who classified sequences based on neglected grammar, F(1, 54) = 1.23, p = 0.27. The results above ruled out the two prossibilities which came from Eitam et al.( 2009), also showed the effect of selective attention on AGL.

Key words: grammatical complexity, selective attention, artificial grammar learning, double artificial grammars