Abstract: Implicit learning hypothesis holds that the symptoms of Autism Spectrum Disorder (ASD) may be due to implicit learning. However, empirical studies have not yet reached a consistent conclusion on this hypothesis. In the literature of language acquisition, two studies investigated implicit learning of autistic children by Artificial Grammar Learning (AGL) paradigm (Brown et al., 2010; Klinger et al., 2007). Nevertheless, they did not balance the chunks of the test materials, possibly resulting in participants acquiring only surface features like chunks rather than the underlying grammar. Moreover, they only examined high-functioning ASD children and the results could not be extended to autistic children with more severe symptoms. Therefore, the current study aims to investigate the implicit learning of artificial grammar of autistic children by controlling the test strings' chunk strength and choosing children with a wider range of autism as participants.
Twenty-six children with ASD, 26 chronological age-matched typically developing (TD) children and 20 chronological age-and IQ-matched children with Intelligence Disorder (ID) participated in the experiment. Twelve grammatical Chinese character strings were used for training. Another 12 grammatical strings and 12 ungrammatical strings constituted 12 pairs of strings for test. Furthermore, the global associative chunk strength (GACS) was counterbalanced between grammatical and ungrammatical test strings. During the training phase, grammatical strings were randomly presented one at a time. On each trial, participant was asked to read the string 3 times and to recall it when it disappeared. During the test phase, pairs of test strings were randomly presented by one pair at a time. Participant was asked to read the pair one time, then judged which one he preferred. After the test phase, participant was asked to report his reason for liking judgment.
For the participants whose verbal report indicated unconscious knowledge of the grammar, the accuracy of liking judgment were .53 ± .12, .57 ± .11 and .56 ± .11 for ASD, TD and ID group, respectively. For ASD group, the performance wasn't significantly different with the chance level .50 ( t(20) = 1.272, n.s.); while for TD and ID groups, the performances were both significantly higher than the chance (TD: t(19) = 3.109, p < .01, d = .64; ID: t(17) = 2.221, p < .05, d = .55). It was also showed that, the performance of ASD group was not significantly different from TD and ID groups (with TD: t(39) = -1.122, n.s.; with ID: t(37) = -.589, n.s.). Taken together, these results showed that children with ASD retain part of the implicit learning ability, but this ability is not enough to enable them to reach the level of significantly exceeding the chance level as the TD and ID children do.
To conclude, the findings of this study tentatively indicated that the implicit learning of grammatical rules of autistic children has both defects and reservations, and provided supporting evidence for the implicit learning hypothesis for autistic children.

Key words: autism, children, implicit learning, artificial grammar learning paradigm

摘要： 由于自闭症的症状和个体的内隐学习都主要涉及语言领域,研究者提出了内隐学习假说,认为自闭症症状可能部分来自于内隐学习缺陷。然而,目前的研究尚未得出一致的结论。该研究采用人工语法学习范式和无意义的汉字串材料,并严格控制材料的组块特征,考察了一般的自闭症儿童、年龄匹配的普通儿童以及年龄和智商匹配的智障儿童的内隐学习。结果发现：（1）普通儿童和智障儿童的内隐学习成绩都显著高于随机水平,表明他们能够内隐习得语法规则;（2）自闭症儿童的内隐学习成绩与随机水平没有显著差异,同时与两组对照组儿童的成绩也没有显著差异。这可能意味着,自闭症儿童语法规则的内隐学习既存在缺陷、又有所保留。该研究为自闭症儿童的内隐学习假说提供了新的来自语言领域的证据。

关键词: 自闭症, 儿童, 内隐学习, 人工语法学习范式