认知与发展心理学研究室知识库

语言经验对人类语义表征产生影响，并进而影响个体的行为表现。然而，尽管心理学界对此达成共识，阅读如何影响语义表征的认知机制尚不清楚。迄今，很少有研究直接检验个体阅读经验与语义空间变化之间的因果关系；此外为探究阅读经验的作用，使用长文本将提升生态效度，但以往的文本表示技术阻碍了对长语篇概念结构的表征，难以探究阅读经验如何随之改变。本研究旨在探索阅读文本如何改变词语间的语义距离，研究一将探索阅读书籍是否能改变词语之间的语义距离；研究二将检验词语在句子结构中与其他词语的共现关系是否能提升语义相似度，以及阅读行为的特征与语义相似度变化的关系。

研究一采用线上实验，使用万字级别的小说文本作为实验材料。通过行为测验检验被试在阅读前后的语义空间变化。结果显示，不论是立即测验还是一周后的测验都显示，被试阅读后，原本无关但在书中相关的词语之间的语义距离变短了，说明了书籍阅读对词语之间的语义距离有影响，且影响持续时间较长。另外，我们还尝试使用分布式语义表征模型表征了书籍的语义空间，但结果显示读者的语义空间与所选模型的语义空间重合度并未提高。

研究一中，阅读经验影响语义距离可能有多种机制，研究二在前人研究基础上特别关注句子结构相关的共现信息对语义距离的影响(Jones & Love, 2007)。先前研究表明，熟悉词在句子结构中的相关信息可能影响个体对其的相似性编码。我们借此实验范式，结合眼动追踪技术探究何种阅读行为与词语之间语义距离的改变有关。采用前人实验设计，将施动者-动词-受动者的不同组合方式嵌入句子框架中，检验两个词语处于相同动词关系内、处于相同的角色、或在一个句子中直接共现等三种关系是否能提高词语间的相似度。行为结果显示，当词语在同一个句子中共同出现时，其语义相似度提高；然而未能发现相同关系、相同角色效应的可靠证据。将眼动和行为数据联合分析结果表明，对动词和施动者的重读越多，直接共现效应就越大。

综上，研究一发现阅读万字的书籍能够改变词语之间的语义距离，使得被试语义空间趋于书籍的知识结构，并提出了一个有效的实验范式来检验阅读经验的因果作用。研究二的结果表明，没有可靠证据表明重复接触两个词语出现在共同关系或共同角色上会提升感知相似性；相反，只有直接共现规律能被习得。此外眼动分析结果表明，对词语及其上下文的反复阅读可能是阅读中习得和调整词语语义表征的共同眼动控制机制。

Linguistic experience influences human semantic representations, thereby affecting their behavioral performance. However, despite this consensus in the field of psychology, the cognitive mechanisms through which reading affects semantic representations remain unclear. To date, few studies have directly examined the causal relationship between individual reading experience and changes in semantic space. Additionally, to explore the effects of reading experience, the use of lengthy texts would enhance ecological validity, but previous text representation technologies have hindered the representation of conceptual structures in long passages. This study aims to explore how reading texts changes the semantic distance between words.

Study one investigates whether reading books can change the semantic distance between words, while study two examines whether the co-occurrence relationships of words in sentence structures can enhance semantic similarity, and the relationship between reading behavior characteristics and changes in semantic similarity. Study one employed an online experiment using novel texts consisting of 2.9k words as experimental materials. Behavioral tests were used to examine changes in participants' semantic space before and after reading. The results showed that the semantic distance between words that were initially unrelated but related in the book decreased after reading, regardless of whether the test was conducted immediately or a week later. The results indicate that book reading affects the semantic distance between words, and this effect persists for a considerable duration. Furthermore, we utilized distributed semantic representation models to represent the semantic space of books. However, contrary to expectations, the results showed no increase in the overlap between readers' semantic space and the selected models' semantic space.

In study one, the influence of reading experience on semantic distance may involve multiple mechanisms. Study two, building on previous research, particularly focused on the influence of co-occurrence information related to sentence structure on semantic distance. Previous research has indicated that the related information of words in sentence structures may affect individuals' encoding of their similarity (Jones & Love, 2007). Study two adopted a design based on previous experiments, embedding different combinations of agent-verb-patient relationships into sentence frames to examine whether three types of relationships - being in the same verb relation, being in the same role, or directly co-occurring in a sentence - could enhance the similarity between words. Behavioral results showed that when words co-occurred in the same sentence, their semantic similarity increased; however, no reliable evidence was found for the same relation or same role effects. The contingent analysis of eye-tracking and behavioral data suggested that the more frequently the verb and agent were reread, the greater the same scenario effect.

In conclusion, study one found that reading novels consisting of tens of thousands of words can change the semantic distance between words, leading participants' semantic space to converge towards the knowledge structure of books, thereby proposing an effective experimental paradigm to examine the causal effect of reading experience. The results of study two indicate that there is no reliable evidence to suggest that repeated exposure to two words appearing in the same relation or role enhances perceived similarity; instead, only direct co-occurrence rules can be implicitly learned. Furthermore, eye-tracking analysis results suggest that repeated reading of words and their context may be a common eye-movement control mechanism for acquiring and fine-tuning word semantic representations during reading.