基于语料库的人机交互中的元话语标记研究

Corpus-Based Research on Metadiscourse Markers in Human-Computer Interaction

摘要: 近年来，元话语研究备受关注，但关于具体人机交互语境中的应用分析的研究仍不足。本研究自建了两个语料库：一个是基于12名中国学生用英语回答问题的访谈，共11,148个形符；另一个是通过ChatGPT回答相同问题，重复12次采集不同回答，共10,867个形符。通过对比分析人类与ChatGPT在回答中的元话语使用差异，本研究旨在探索元话语在人机交互中的表现。结果表明，中国学生在口语问答中更频繁地使用互动性标记，而ChatGPT的回答则呈现出机械化、标准化的模式。具体而言，学生语料库中的自我提及和参与性标记较多，且自我提及的频率显著高于ChatGPT。此研究不仅丰富了元话语在口语英语教学和人机交互领域的研究，还为改进相关设计提供了参考。

Abstract: In recent years, metadiscourse research has garnered significant attention, yet its application in the specific context of human-computer interaction remains underexplored. This study constructs two corpora: one based on interviews with 12 Chinese students answering questions in English (11,148 tokens), and another comprising responses generated by ChatGPT to the same questions, collected through 12 repeated iterations (10,867 tokens). By conducting a comparative analysis of metadiscourse usage between humans and ChatGPT, the study investigates the manifestations of metadiscourse in human-computer interaction. The results indicate that Chinese students employ interactive markers more frequently in spoken Q&A, while ChatGPT’s responses exhibit a more mechanical and standardized pattern. Specifically, the student corpus features a higher prevalence of self-mentions and engagement markers, with self-mentions occurring significantly more often than in ChatGPT’s outputs. This research not only enriches the study of metadiscourse in spoken English teaching and human-computer interaction but also provides insights for improving related designs.