中国科学院心理研究所机构知识库

工作记忆和背景知识是制约预期推理生成和维持的两个重要因素。对这些制约因素的研究将揭示阅读的动态加工过程和认知资源分配方式以及长时记忆中信息组织和提取的性质，并且为汉语篇章理解系统的建立提供科学依据。本研究从工作记忆能力的个体差异和背景知识结构的角度，探讨阅读描述日常生活事件的段落时，预期推理过程的性质和资源分配特点及其与阅读焦点转移的关系。为了测量被试的工作记忆能力首先采用句子一尾词测量方法区分具有不同工作记忆能力的被试。为探讨背景知识结构中语义联系类型和强度与预期推理的关系，又采用了开放式问卷实验方法，由被试生成日常生活中的事件以及组成这些事件的各种活动和活动的参加者。然后，从事件的熟悉性、活动的典型性和可区分性三个维度进行评价，形成事件以及每个事件活动序列，并据此编制实验材料。实验主要考察了不同工作记忆能力的被试高熟悉性的事件与活动之间进行双向的预期推理时，活动的典型性和可区分性的约束作用;考察了由活动、场景、低熟悉性事件和行为参与者的变化引起的阅读焦点的转移对于维持预期推理所产生的影响，以及不同工作记忆能力的被试进行预期推理时的加工模式。实验结果提示，工作记忆能力在预期推理过程中有如下性质:

工作记忆能力高的被试加工速度快，(2)工作记忆能力高的被试在由成员活动指向事件的预期推理过程中，能够激活较多项目，但是每个项目上的激活量小;工作记忆能力低的被试激活的项目较少，但是每个项目上的激活量较大。在由事件指向成员活动的预期推理过程中，没有这种现象。(3)工作记忆能力低的被试容易受到阅读焦点转移的影响，并且在出现阅读焦点转移时，可以对预期推理进行抑制。

通过实验结果可以看出背景知识在预期推理过程中有如下性质:

(1)行为参与者是影响预期推理维持的一个主要约束因素。(2)事件熟悉性也是影响预期推理维持的一个重要约束因素。但是其约束作用不如行为参与者强。(3)典型性是影响从事件指向活动的预期推理生成的一个主要约束因素，可区分性也有一定约束作用。(4)可区分性是影响从活动指向事件的预期推理生成的一个主要约束因素。

通过本研究还可以初步作出以下结论:

(1)在满足一定约束条件时，读者不仅可以进行预期推理，而且这种推理的结果可以加快下文词汇加工的速度.(2)在本研究中背景知识是影响预期推理生成和维持的最重要因素，而焦点转移不是影响预期推理维持的主要因素，局部意义连贯性也不是影响预期推理生成和维持的主要因素。(3)预期推理在适当的时候会被抑制，不会过度消耗有限的认知资源。(4)低工作记忆的被试在预期推理过程中似乎经历了一个更加复杂的动态分配过程。

The differences of working memory and the structure of background knowledge are two main factors in constraining predictive inferences in reading. In order to investigate the properties of predictive inferences and allocation of resources in paragraph reading, 5 experiments were conducted. Working memory capacities of the subjects were tested by a version of reading span method (Daneman & Carpenter, 1980). The experimental materials were generated through a subject-generation method. FAMILIARITY of a event, STANDARDNESS and DISTINCTIVENESS of actions in the event were three dimension. An online naming probe paradigm was used in the experiments. The first two experiments focused on the effects of STANDARDNESS and DISTINCTIVENESS in predictive inferences making. The other experiments examined the influence of focal statement changes caused by actions, specific setting, events with low degree of FAMILIARITY, agents of actions. In above experiments naming patterns of the subjects with high and low working memory capacities indicated individual differences of allocation of resources in generation and maintenance of predictive inferences.

The results indicates the following effects of working memory capacities: (1) working memory capacity may reflect naming speed of the subjects, the higher the capacity is, the faster the speed will be; (2) When an action causes the generation of prediction, the high capacity subjects may activate more items in their knowledge base, but spend less amount of resources on each items; When an event arise a prediction this effect will disappear; (3) The low capacity subjects seems more effective in depressing predictive inferences when focal statement changes.

The results also shows effects of background knowledge: (1) agents and FAMILIARITY of an event seems two main factors in predictive inferences maintaining, but the former is stronger; (2) STANDARDNESS mainly constrains the prediction directing from an events to actions and DISTINCTIVENESS influence the prediction directing from an action to events. The experiments reflects some theoretical issues in predictive inference generation and maintenance. (1) Prediction making is a highly constrained mechanism in reading. (2) Focal

is not a main factor. Local coherence is not an importance too. (3) Background knowledge is more importance than working memory capacities. (4) Predictive inferences will not consume much amount of resources. (5) Low capacity subjects seems to employ more complex mechanism of depression.