为了更好地适应不断变化的情境，人类需要根据外部反馈监控和调节自身行为。在日常生活中，反馈并非是全或无的，而是概率性的，例如，下了一步棋，有80%的可能会赢，但也有20%可能会输。已有关于反馈概率对学习的影响的研究主要采用的是简单学习任务范式，即反馈是紧接着一个学习行为之后的，但是，有时反馈是在一系列学习行为之后才出现的，例如，学习下棋，需要经过多个步骤的学习后才会获得反馈，这个学习过程是一种“序列学习”的过程。人的生活中遇到的大部分的任务都是这种性质的。 序列学习（sequence learning）需要个体学习一个特定规则，并遵循这个规则去感知、表征并且执行一系列行为。反馈概率（feedback probability）对序列学习的影响存在很大的复杂性，这主要是因为反馈概率的影响既可以体现在几步选项后的最终反馈概率的差异上，又可以体现在各步骤的不同反馈概率上。以两步骤的序列学习为例，“最终反馈概率”是指个体需要区分各种不同的两步组合在最终反馈概率上的差异，即最优解（第一步和第二步都选对了）和较优解（两步中只有一步选对了）获得奖赏反馈概率的差异。“各步骤的反馈概率”是指，在最终反馈概率相同的情况下，各步骤中的正确选项对最终反馈的贡献大小。例如，“一着不慎满盘皆输”指的就是关键性的一步棋走得不当，整盘棋就输了，即某一步的反馈概率对最终反馈概率的贡献要大于其他步骤。已有关于反馈对序列学习的影响的研究，研究者们采用的最终反馈概率各不相同，研究结果也并不一致。这一结果的冲突是否与各研究中反馈概率的不同有关呢？然而，关于不同的最终反馈概率对序列学习的影响，目前还缺少研究。另外，神经生物学和计算科学领域对各步骤反馈概率进行了考察，并且取得了一定的进展，但是，人类的序列学习效果、策略和脑电活动特点是否受各步骤反馈概率影响，目前研究尚未涉及。此外，在不同的各步骤反馈概率条件下，各步骤之间增加线索是否会调节个体序列学习效果、策略和脑电活动，目前尚无定论。 基于此，本研究主要采用概率性序列选择任务（probabilistic sequence choice task），在该任务中需要被试学习两步序列规则来研究反馈概率对序列学习的效果、策略和反馈阶段的脑电活动的影响作用。论文通过三个研究5个实验，首先考察在各步骤反馈概率相同的条件下，最终反馈概率对个体的学习效果、策略和反馈阶段的脑电活动的影响作用；进而，揭示在最终反馈概率相同的条件下，不同的各步骤反馈概率对个体的学习效果、策略和反馈阶段的脑电活动特点的影响作用；最后，通过增加暗含第一步结果的线索，探究线索是否影响个体序列学习效果、策略和反馈阶段的脑电活动特点。 研究一，通过设置高、中、低三种最终反馈概率水平，考察了最终反馈概率对个体学习效果、策略和反馈阶段的脑电活动的影响，包括两个实验。实验1选取66名大学生为被试，从行为层面发现，最终反馈概率越高，即最优解与较优解的奖赏反馈概率差异越大，被试的正确率越高，反应时越短。被试在高概率和中概率条件下的策略可能都是先学习第一步再学习第二步（以目标为导向为主的策略）。实验2以21名大学生为被试，在实验1的基础上，通过ERP技术，以与反馈有关的负波（feedback-related negativity，FRN）为指标，探究在高中低三种水平的最终反馈概率下，被试对奖赏和无奖赏刺激加工的脑电活动特点，结果表明，不同的最终反馈概率会影响被试在反馈阶段的脑电加工活动，这一影响主要体现在奖赏反馈的脑电加工上。具体而言，相比高反馈概率条件和低反馈概率条件，中反馈概率条件下奖赏诱发的FRN波幅最大；低反馈概率条件下的奖赏诱发的FRN波幅要显著小于高反馈概率条件下的奖赏诱发的FRN波幅。另外，在高、低反馈概率条件下，奖赏反馈刺激诱发的FRN波幅可以预测被试的最终学习正确率，但是，在中反馈概率条件下则未发现该预测作用。此外，本实验选取了有明显学习进程的中反馈概率条件，进一步比较了学习前期和学习后期的反馈阶段的脑电活动差异。结果表明，学习前期反馈诱发的FRN波幅要显著大于学习后期的。 研究二，通过设置第一步最优选择对最终反馈贡献高和第二步最优选择对最终反馈贡献高两种条件，考察了不同的各步骤反馈概率对个体学习效果、策略和反馈阶段的脑电活动影响作用的差异，包括两个实验。实验3以31名大学生为被试，从行为层面发现，在第一步和第二步最优选择对最终反馈贡献高的两种条件下，总体正确率和反应时均无显著差异。但是，被试在策略上存在不同，即在反馈概率高的步骤上的正确率更高。实验4以22名大学生为被试，考察了不同的各步骤反馈概率对脑电活动的影响，结果发现，在第一步和第二步最优选择对最终反馈贡献高的两种条件下，与奖赏反馈刺激相比，无奖赏反馈刺激诱发的FRN波幅都更大，FRN潜伏期更短。但是，两种条件的反馈阶段的脑电活动并无显著差异。 研究三，在研究二的基础上，通过在两步之间，增加暗含第一步结果的线索，考察了在不同的各步骤反馈概率的条件下，线索对学习效果、脑电活动特点的调节作用，包括1个实验。实验5的结果表明，线索在各条件下诱发的FRN均无显著差异。进一步比较实验4和实验5的行为数据，结果表明，有无线索对不同学习下的效果和策略均无影响。 基于以上结果，本研究主要得到以下结论：1）最终反馈概率会影响序列学习效果、策略和反馈阶段的脑电加工特点。学习前期和学习后期的脑电活动存在差异。2）不同的各步骤反馈概率在序列学习策略上的作用存在差异，但在序列学习效果和反馈阶段的脑电活动上并无显著差异。3）增加暗含第一步结果的线索对两种条件下的学习效果、策略并无显著影响。在两种条件下，被试对奖赏反馈刺激和无奖赏反馈刺激的脑电加工存在差异。

Human beings need to monitor and regulate their behavior on the basis of external feedback information to adapt to the changing environment better. In our daily life, feedback is given by probabilistic. For instance, we make a move in a chess game, and the feedback probability of the move was 80% for winning but 20% for losing. Previous researches about the effect of feedback probability on learning primarily focused on the simple learning, that is, feedback is followed by an action. However, sometimes feedback depends on the consequences of a series of actions. Sequence learning needs an individual to learn a particular rule, and then he or she follow this rule to perceive, represent, and perform a series of actions.The effect of the feedback probability on the sequence learning is very complicated. Because the influence of the feedback probability can be not only caused by the difference of the final feedback probability after the actions, but also by different feedback probabilities of the two options of each step. Take two-step sequence learning for example, the final feedback probability means the differences between the final feedback probabilitiesof the optimal solution (two steps are chosen correctly) and the better solution (only one of the two steps is chosen correctly). The feedback probability of each step refers to the contribution of the correct option in each step to the final reward probability. The adage, "One careless move loses the whole game", means some actions can have significant for the finial outcome. In addition, no conclusion was gotten about whether adding clues between different steps would mediate the learning effect, strategies and EEG activities under different feedback probability conditions. Based on mentioned above, this study mainly used a probabilistic sequential choice task, which people learn to sequentially choose actions with probabilistic outcomes and receive feedback only after the whole action sequence is executed. We conducted three studies to investigate the effect of the finial feedback probability on the learning effect, strategies and the EEG activities of feedback processing under the condition of same feedback probabilities in each step. Moreover, the effect of different feedback probabilities of each step on the learning effect, strategies and the EEG activities of feedback processing under the condition that the final feedback probability was same. In addition, we tried to add the clues implying the result of the first step, and explored whether the clues affect individual sequence learning effects, strategies and EEG activities of feedback processing. Study 1 includes two experiments, which were to investigate the influence of the final feedback probability on the learning effects, strategies and the EEG activities. Results from experiment 1 showed that the higher the final feedback probability was, the higher the accuracy was and the shorter the reaction time was. The strategy which individual adopted was that participants learned the first step first and then learned the second step under the condition of high probability and middle probability. Based on the experiment 1, in experiment 2, we used the ERP technique to investiagte the EEG activities of reward and non-reward feedback processing under different final feedback probabilities. The main results are as following: First, the amplitude of FRN (feedback-related negativity) for non-reward feedback was larger than for reward feedback. Second, the final feedback probabilities affected the EEG activities of feedback processing, but only occurred in reward. Thirdly, under the condition of high and low feedback probability, but not middle feedback probability, the amplitude of FRN for reward can predict individual final accuracy of the task. Fourthly, the EEG activities between pre-learning and post-learning feedback processing was different under the condition of middle feedback probability. Sepecifically, the results showed that the amplitude of FRN for pre-learning feedback was significantly larger than for post-learning. Study 2 includes two experiments, which were to explore the effects of different steps of feedback probabilities on individual learning effects, strategies and the EEG activities of feedback processing. In experiment 3, there was no significant difference in the total accuracy and reaction time between the condition of high feedback probability of the first step and of the second step. However, the strategies participants used under two conditions was difference. Besides, the abilities of cognitive flexibility of the subjects correlated with the final learning accuracy of the subjects under the two conditions. The effects of different feedback probability of each step on EEG activities of feedback processing were investigated in experiment 4. The results showed that, compared with reward feedback, the amplitude of FRN for non-reward was larger and the latency of FRN was shorter in general. However, there was no significant difference in EEG activities of feedback processing between these two conditions. Study 3, which consisted of one experiment, investigated the effect of clue on learning effects and the EEG activities under the conditions of feedback probabilities of each step were different. Based on the study 2, we added a clue implying the results of the first step between two steps. The results of experiment 5 showed that there was no significant difference in FRN for cues under various conditions. The amplitude of FRN for non-reward was larger than for reward. Compared with the condition that the second step is high probobility, the latency of FRN for non-reward was shorter under the condition of high feedback probability in the first step. By comparing the behavioral data of experiment 4 with those of experiment 5, we found cues did not mediate the learning effects and strategies under different learning conditions. In summary, several conclusions were drawn: 1) The final feedback probability could affect the learning effects, strategies and the EEG activities of feedback processing. In addition, the EEG activities under pre-learning condition and under post-learning condition were different. 2) The feedback probabilities of each different steps has different effects on learning strategies, but not on the learning effect and the EEG activities of feedback processing. Moreover, there was a difference between the EEG acitivities for reward and for non-reward feedback in general. 3) By adding the cues implying the first step result, we found that cues did not mediate the learning effect and strategies. In addition, there was a difference between the EEG activities for reward and non-reward feedback in general.