◎ 郭秀艳

　　内隐学习（implicit learning）是指有机体在与环境接触的过程中不知不觉地获得了一些经验并因之改变其事后行为的学习。内隐学习研究受到广泛的重视，成为当今认知心理学的一大热点和前沿，因为它有着深刻的理论意义和广泛的应用前景。
内隐学习研究和内隐记忆研究息息相关，同时又受益于Reber的人工语法范式。本论文分四章，从理论和实验两个侧面对内隐学习进行全面而系统的研究。
　　在第一章中，我们对内隐学习研究的崛起作了精辟的阐述，详细介绍了以Reber为代表的人工语法范式的方法学特点，指明了内隐学习的研究，一旦有了人工语法范式，犹如"鱼儿得水"，促使内隐学习研究由外到内，由简单到复杂，由分离到协同，和由理论到应用的飞快发展。
　　Reber（1967）创立的人工语法学习（artificial grammars learning）是内隐学习研究中最常用的范式，学习材料是人工语法范式的第一要素。内隐学习实验所用的学习材料必须新颖，有一定难度，且和被试已有的知识表征没有联系，以保证被试在学习过程中不可能利用有意识的、外显的策略轻易地获得学习材料中暗含的规则。因此人工语法就成为最佳选择。到目前为止被研究者接纳的人工语法有两类--限定状态语法和双条件语法。指导语是人工语法范式的第二个要素。在内隐学习的研究中，实验的指导语在要求被试去做什么及如何去做的意义背后，是要引导被试去"进行"内隐、外显的学习。实验中，引导内隐学习的指导语为记忆，引导外显学习的指导语为规则发现。实验程序是人工语法范式的第三个要素。人工语法范式的共同程序是：在学习阶段，先给予被试引发内隐或外显学习的指导语--记忆指导语或规则发现指导语，然后向被试呈现一系列符合某一人工语法的字母串；在测验阶段，要求被试判断新的字母串，是否符合该人工语法，或是否与记忆过的字母串相象。如果被试对字母串的正确判断显著高于随机水平，就可认为他们已经无意识地获得了一定的人工语法规则。
　　总之，在人工语法学习范式中，方法论上的每次进步都带来了理论上的重大突破；反过来，理论上的深入又促进了方法论的完善。
　　在第二章中，我们阐述了内隐学习的本质，阐述了内隐学习研究的相关领域，阐述了内隐学习研究的新进展，最后我们还全面高度地分析了内隐学习研究给我们的重要启示。
　　究竟什么是内隐学习呢？近年的研究，使得人们对内隐的概念有了一定程度的认识。很多理论家认为，内隐与外显的认知过程属于复杂认知任务的操作（Brook，1978；Evans，
1972；Pollio，1974；Seger，1994；Howard，1997）。外显学习中发现和控制任务变量的机制类似于有意识的问题解决，它包括一个试图形成任务的心理表象，搜寻同功能系统的知识的记忆，以及试图建立和检验任务操作的心理模型（Mathews，1988）。这整个思维过程是有意识的，主观努力是被清晰意识到的。内隐学习则是不同于以往的另一种学习模型，它比外显思维更有力，且是自动的、无意识进行的（Mathews，1988；Fletcher，1998）。换言之，内隐学习是在（某种意义上）不知道刺激之间或刺激与行为之间存在连接的情况下对它们的学习。
第三章是论文的重头戏。它从实验的角度对前述内隐学习的本质特征及其与外显学习的关系进行了逐层递进的探讨。研究共分四个实验。实验1中，我们对Reber语法的一个变式进行了内隐与外显的任务分离。此外，我们还增添了字母集变量以了解内隐学习的抽象性，及时间变量（单元）以了解内隐与外显学习的详细进程，从而可以间接推测内隐学习的理解性。可以说字母集变量的加入是目前国内从抽象性角度对内隐学习进行研究的第一次尝试，而从学习进程来比较和分析内隐、外显学习并进而推断理解性则为本实验的创新之处。
　　实验2是实验1的会聚操作。实验2中，我们取消了常规程序学习阶段中的学习字表环节，只是呈现指导语，随后让被试进行一个一个的区辨测验，来减弱被试的学习强度，以求从反面证明内隐学习的存在。
　　实验3和实验4是对前面两个实验的继续和深化。它们运用了内隐和外显的强分离手段--匹配和编辑。实验3与实验4的区别在于，前者让被试学习的材料是限定状态语法，后者让被试学习的是双条件语法。通过这两个实验的研究和对照，可以了解内隐学习与外显学习之间的关系，它们是否存在协同效应，以及协同效应产生的最佳条件。
　　本研究的实验部分获得了可喜的结果，达到了预期目的。如下四点启迪，尤为重要：
　　1 内隐学习的自动性
　　内隐学习的自动性特征在本研究中得到了有力证明，实验1至3中被试试图找出所用的限定状态语法规则的思路，似乎无助于获取语法知识。在实验1中，记忆组具有要学习的样本表，实验结果表明记忆组被试比规则发现组被试做得更好些，从而以不同方法和不同侧面取得了Reber及其同事们相近的实验结果（Reber，1970；Reber等，1980，1994，1998）。但是，在实验2中，省略了学习字表，以致于弱化了指导语的作用，但内隐学习依然存在，只是内隐与外显学习效果之间的差异不显著。
　　实验3提供了能获得有关限定状态语法知识的自动学习过程的有力证据。在这一实验中应用的匹配和编辑训练比以前研究中所用的分离手段（例如，指导语控制）更强，能够产生对例子的内隐、外显的极端控制。实验4表明，内隐学习过程对前面三个实验中使用的限定状态语法的学习是足够的，对双条件语法的学习却是另外一回事。实验3和实验4之间相反的结果更令人信服地证明了复杂认知任务的学习中不止牵涉到一种学习过程。
　　2 内隐学习的抽象性
　　对于这些实验中探索的内隐知识的抽象性问题，采用的是一种较间接的方法推测出的。通过对同一语法的不同字母系列之间学习迁移的测定，我们能够推断被试对只在抽象特征上相似的不同刺激所进行的概括总结程度。
　　通过对学习迁移的测定发现，被试所获得的语法知识中，部分是特殊知识（Brooks，1978；1988；Vokey和Brooks，1988；Dienes，Altmann，Kwan和Goode，1995），部分是更为一般和更为抽象的语法知识（Reber，1969；1976；1993）。相当多的证据表明被试能把他们所学的语法知识转移到一个不同的字母集。同样，所出现的抽象性似乎也并不需要更进一步的外显过程。
　　3 内隐学习的理解性
　　内隐学习的理解性是通过它与外显学习在任务操作中的进展情况而间接推测出的，如果两种学习方式并行上升，那么可知内隐学习所获得的知识能够为意识所接近。
　　实验1中，随着学习的推进，内隐组与外显组被试的成绩都是逐渐上升的，换言之，随着内隐知识的增长，个体外显知识也在悄然积累着，只是不及内隐知识突出而已。由此，我们可以间接地推测，内隐知识是能够被意识接近的。此外，在实验2、实验3和实验4中也都得到了类似的结果。
　　4 内隐学习和外显学习的交互作用
　　在实验3和实验4中，同时应用内隐与外显学习而产生的协同作用得到了清楚的证明。这两个实验中，都是交替组被试的成绩大大优于其它组。几乎所有混合条件的成绩曲线也高于任何一种单一任务条件（匹配或编辑）。这一发现表明内隐获得的知识与外显获得的知识发生了交互作用。
　　内隐学习与外显学习之间的协同效应给我们的重要启示是，学习复杂任务时先应具备一个内隐知识基础，然后再试图建立外显的任务模型。由此可知，最理想的学习需要两种过程--内隐和外显。
　　第四章是整个研究的总结，我们总结了本实验的四点重要启迪，指明了内隐学习研究方法的日新月异，还介绍了近年来内隐学习的丰硕研究成果及其新兴理论。最后，我们展望，内隐学习研究具有广泛的应用前景。在人类学习和现实生活中，内隐学习以及它和外显学习的协同效应，将起着不可估量的作用。

　

◎ Xiu-yan GUO

　　 Implicit learning is the course of acquiring implicit knowledge, which is what people acquire unconsciously during contacting with the environment around them, which can change their behaviors later. Researches on implicit learning are emphasized widely, and become a big hot point today, because they are also a drive for social cognition, school teaching, human and computer interaction, games and so on.
From the beginning, Implicit learning has been interacting with implicit memory. They always take after each other, however, benefit from Reber's artificial grammar. The present paper includes four parts, which completely and systematically discourses on theories and experiments on implicit learning.
At chapter 1, we introduced the growing up of implicit learning in details and its founder, Reber, a famous psychologist. We also explained artificial grammar model, which comes from a Reber's original experiment（1967）. It is artificial grammar that has been urging the implicit learning research so that implicit learning research developed very fast.
Artificial grammar model created by Reber (1967), is in common use today in implicit learning. Learning material is the first factor in artificial grammar model. Implicit learning material must be novel, difficult, and not related to subjects' original knowledge token, in order to insure subjects against using conscious explicit strategies to easily grasp those implied rules. So artificial grammar is the best choice. Now there are two kinds of artificial grammar, definite-state grammar and dual-conditioned grammar. Instruction is the second factor in artificial grammar model. In researches on implicit learning, instructions do not only tell subjects what and how to do, but also let different subjects go in for different learning, implicit learning or explicit learning. In experiments, instructions for implicit learning are some words about memory; those for explicit learning are some ones about rule. Procedure is the third factor in artificial grammar model. The common procedure of artificial grammar model is as follows. In learning phase, researchers first tell different subjects different instructions (about memory or about rule), and then present a set of letter strings which accord with some artificial grammar; in testing phase, researchers let subjects answer if they accord with artificial grammar, or if they resemble those old ones. If subjects' correct judgements are significantly higher than chance, then we can infer that they have unconsciously acquired some artificial grammar rules.
To sum up, in artificial grammar learning model, every progress in methodology can bring on great breakthrough in theory; but the reverse is true.
At chapter 2, we explored the nature of implicit learning, involved the relational fields，introduced the new development, and in the end analyzed some great implications of implicit learning for us.
What is implicit learning on earth? Recent researches make people understand it to some degree. Many theorists think, implicit and explicit cognition belongs to the operation on complicated cognitive tasks.（Brook，1978；Evans，1972；Pollio，1974；Seger，1994；Howard，1997）. Explicit learning is similar to conscious problem solving (Mathews, 1988); but implicit learning is different to any other learning, which is automatic, unconscious, and stronger than explicit one (Mathews，1988；Fletcher，1998). In other words, implicit learning often happens under subjects' not knowing the relations between stimuli or between stimulus and behavior.
Chapter 3 is a key part. From the view of experiment, it successively explored the preceding nature of implicit learning and relationships between implicit learning and explicit one. The whole research consists of four experiments.
In experiment 1, we applied instructions to differentiating implicit learning and explicit learning under an ordinary artificial grammar model. But we added a letter set variable, in order to investigate abstract characteristic of implicit learning, and a timing variable, for exploring many details through implicit learning and inferring the understandable characteristic of implicit learning. These kinds of betterment are the first attempts within our country.
Experiment 2 is a converging operation of experiment 1. We canceled learning letter list, which was presented in experiment 1; we only presented instructions. What we did in experiment 2 reduced subjects' learning intensity. If under this learning conditions the performance of remembering instruction group was not different from that of rule finding instruction one, we can say, implicit learning maybe not take place at all. Therefore, experiment 2 accorded with experiment one from the reverse side.
Experiments 3 and 4 continued and deepened the existing searches because of applying the stronger ways (matching and editing) to differentiate implicit and explicit processing. The difference between experiments 3 and 4 is that, the former let its' subjects learn finite-state grammar, but the latter let its' learn dual-conditioned grammar. The researchers found that finite-state grammar was fitter for implicit learning, while two-condition grammar was fitter for explicit learning. So, by comparing the two experiments, we can acquired some powerful results, such as their relationships, synergism and its' conditions.
The experimental results of this present research accorded with the expectations. The four points as follows are more important:
1. Automatic characteristic of implicit learning
The present research strongly proved automatic characteristic of implicit learning. From experiment 1 to 3, we found that rule finding instructions were helpless to acquire definite-state grammar. In experiment 1, memory group's performance is higher than rule group's one, which is similar to Reber and his colleagues' results (Reber, 1970; Reber etc., 1980, 1994, 1998). But, in experiment 2, even if learning letter list was cancelled so that the function of instructions was weakened, implicit learning still existed.
Experiment 3 provided more sufficient evidence for automatic learning of definite-state grammar, because it adopted a kind of powerful differentiating means (matching and editing). Experiments 3 and 4 showed, definite-state grammar was enough sufficient to bring on implicit learning; but dual-conditioned grammar was not. The reverse results of experiment 3 and 4 proved that there was not only one kind of learning procedure during complicated cognitive task.
2. Abstract characteristic of implicit learning
In order to explore abstract characteristic of implicit learning, the present researcher adopted an indirect method for inferring.
By measuring learning transfer between different letter sets, we can infer, in subjects' acquired grammar knowledge, part of it is special knowledge (Brooks, 1978; 1988; Vokey & Brooks, 1988; Dienes, Altmann, Kwan & Goode, 1995), part of it is general and abstract knowledge(Reber, 1969; 1976; 1993). Good-sized evidence showed that subjects could transfer their acquired grammar knowledge into a different letter set. And this kind of abstract characteristic seems not to need further explicit procedure.
3. Understandable characteristic of implicit learning
Understandable characteristic of implicit learning was inferred by comparing the progress course of implicit learning with one of explicit learning. If the two progress courses are parallel and ascending, then, we can say, the acquired knowledge by implicit learning can approach consciousness.
In experiment 1, with the progress of learning, subjects' performances in implicit and explicit groups were both gradually ascending. In other words, with the growing up of implicit knowledge, individual's explicit knowledge is also accumulating silently, but is not standing out like implicit one. Therefore, we can indirectly infer, implicit knowledge can approach consciousness. In addition, the results of experiments 2, 3 and 4 were all similar.
4. Interrelations between implicit and explicit learning
In experiments 3 and 4, the present researcher applied implicit and explicit learning in the meantime and got synergism between them. In the two experiments, the performances of alternatively processing group between implicit and explicit learning were all much higher than those of other groups. And the performances of almost all mixed groups between implicit and explicit learning were higher than any group's with a single task. So this finding showed that interaction between implicit and explicit knowledge happened.
The important implication of the interaction between implicit and explicit learning is, people should first have an implicit knowledge base before learning complicated task, then be trying to establish explicit task model. In one word, the most ideal learning model needs two kinds of procedures (implicit and explicit ones).
Chapter 4 is a summary. We summarized four important implications; indicated implicit learning methods, results, and theories; in the end, we expressed our confidence, because we believe that implicit researches must be abundant and broad. In the fact, implicit learning and its' cooperation with explicit learning, have gone deep into people's everyday life.　

　　作者简介：郭秀艳，1970年生，女，华东师范大学2001级博士研究生，师从杨治良教授，主要从事实验心理学研究，现在华东师范大学心理学系工作。

（责任编辑：王 珉）