认知与发展心理学研究室知识库

人脑可以从连续变化的听觉信号中快速准确地切分语音单元、识别特定词汇并与前文语境进行整合以理解言语，这涉及到多个言语层级的加工和动态交互过程。预测可能是人脑在进行言语理解时所运用的其中一种内在机制，通过预先激活相关信息的方式来运行。基于产生的预测可能是进行主动预测时所使用的根本机制，该理论主张理解者运用言语产生系统对即将到来的词汇进行预测，预测可以在任何层级和阶段上发生，其中对高层级信息的预测早于对低层级信息的预测。虽然已经有研究开始关注言语词汇预测的具体内容，确定了多个层级上信息预激活效应的存在，但目前对相关实验范式的探索和对层级性预测内容之间关系的认识还不够深入。基于此，本文使用眼动视觉情境范式依次递进开展了三个实验，探索实验范式的有效性并厘清言语理解过程中高层级词汇语义预测和低层级词汇语音预测的先后和强弱关系。

实验一主要关注预视时长这一实验参数对言语词汇语义与语音预激活效应的影响，发现缩短预视时长至1000毫秒虽然影响预激活效应的起始时间，但依然能够重复出与前人使用2000毫秒预视时长类似的效应。这表明缩短预视时长并没有从根本上影响语义和语音预激活效应的出现，说明言语词汇预激活效应具有一定鲁棒性。实验二主要考察句子语境对词汇语义预激活效应的影响，发现当语义相关词更容易和句子语境发生整合时产生的语义预激活效应更强。此时的语义预激活效应本质上是语境整合与词汇预测的共同作用结果而不单代表预测效应，提示语境整合因素可能会对词汇语义预激活效应的解释造成干扰，在使用视觉情境范式探究言语词汇预测过程的实验中应着力排除语境的影响。实验三采用排除语境影响的实验材料，进一步揭示仅由词汇预测引发的语义预激活效应和语音预激活效应的时间进程和层级性关系，发现词汇语义预激活效应早于词汇语音预激活效应发生且效应更强。该结果符合基于产生的预测理论中对于言语词汇认知过程的假设，支持了其“高层级信息的预激活早于低层级信息”的阐述。

总言之，本文深入探索了言语理解过程中词汇预测的认知加工过程，在确定了实验范式有效性并排除了语境干扰的前提下，观测到了语义和语音预激活效应的存在并刻画出了它们在时间进程和效应强度上的差异，为基于产生的预测理论提供了实验证据上的支持。本文研究发现有助于深化言语理解和预测的认知机制，为脑启发语言智能的迭代提供潜在参考。

The human brain can quickly and accurately split speech units from continuously changing auditory signals, then recognize correct words and integrate them with the preceding context to understand speech, involving hierarchical and dynamic interaction processes. Prediction, operating by pre-activation, may be one of the intrinsic mechanisms in speech comprehension. Production-based prediction may be the underlying mechanism in making active prediction, and the theory claims that comprehenders use their production system to predict upcoming words, which can occur at any level and stage, with lower-level information following higher-level information. Although many studies have explored the contents of lexical prediction and verified the pre-activation effect at multiple levels, the relevant experimental paradigms and the nature of hierarchical speech prediction are not yet been fully understood. Therefore, the present study conducted three experiments to explore the validity of the visual world paradigm as well as to investigate the relationship between high-level semantic prediction and low-level phonological prediction, gradually.

Experiment 1 focused on the possible influence of the preview duration on the preactivation effect in a visual world paradigm experiment. Results showed that shortening the preview to 1000 ms affected the onset of the pre-activation effect but still reproduced the effect similar to the previous study which was conducted under 2000 ms preview, indicating that changing the preview did not fundamentally affect the emergence of the semantic and phonological pre-activation effect, which reveals that the lexical pre-activation effect is somewhat robust. Experiment 2 focused on the influence of sentence context on the semantic pre-activation effect, and found that the experimental effect was stronger when the semantically related words were more easily integrated with the sentence context, so the semantic pre一activation effect at this time was essentially a joint outcome of contextual integration and lexical prediction rather than the pre-activation effect alone, suggesting that the contextual integration factor may interfere with the interpretation of the semantic pre一activation effect when using visual world paradigm to explore lexical prediction. Using materials that could exclude the influence of sentence context, Experiment 3 focused on the time course of semantic and phonological pre-activation effect and the hierarchical relationship between them again. Results showed that the semantic pre-activation effect occurred earlier and was stronger than the phonological pre-activation effect, which is in line with the hypothesis of the prediction-by-production theory, supporting the statement that "the pre-activation at lower levels follows pre-activation at higher levels".

In conclusion, this study explored the cognitive process of lexical prediction during speech comprehension. After determining the validity of the experimental paradigm and excluding contextual interference, this study observed the existence of semantic and phonological pre-activation effects and characterized their differences in time course and effect strength, providing experimental evidence to support the prediction-by-production theory. This study helps convey a contribution to the future exploration of the cognitive mechanism of speech comprehension and prediction and provides a potential reference for fields of brain-inspired AI.