Can AI simulate or emulate human stance? Using metadiscourse to compare GPT-generated and human-authored academic book reviews

IF 1.7 1区文学 0 LANGUAGE & LINGUISTICS

Journal of Pragmatics Pub Date : 2025-08-14 DOI:10.1016/j.pragma.2025.07.018

Guangyuan Yao , Zhaoxia Liu

{"title":"Can AI simulate or emulate human stance? Using metadiscourse to compare GPT-generated and human-authored academic book reviews","authors":"Guangyuan Yao , Zhaoxia Liu","doi":"10.1016/j.pragma.2025.07.018","DOIUrl":null,"url":null,"abstract":"<div><div>This study investigates whether generative AI (represented by ChatGPT) can simulate or even emulate the stance expressed by human authors in the specific genre of academic book reviews. Through a comparative analysis of ChatGPT-generated reviews and human-authored reviews, this study focuses on the use of interactional metadiscourse markers (e.g., hedges, boosters, attitude markers, and self-mention) to reveal current AI's capabilities and limitations in handling complex evaluative discourse and interpersonal interaction. The findings indicate that ChatGPT overall employs interactional metadiscourse markers more frequently than human authors, due to its significant overuse of attitude markers. However, it underuses hedges and self-mention significantly, suggesting a reliance on explicit evaluative language while lacking nuanced caution and authorial presence. These results highlight that current AI's simulation of human writing is genre-sensitive but incomplete, particularly in achieving the balance of caution, conviction, and authorial presence, which is typical of human reviewers. The distinct metadiscoursal patterns identified may serve as linguistic fingerprints for distinguishing AI-generated reviews from human-authored ones. The study also offers pedagogical implications, emphasizing the need for educators and students to recognize current AI's limitations in modeling nuanced stance and fostering authentic authorial voice in evaluative genres.</div></div>","PeriodicalId":16899,"journal":{"name":"Journal of Pragmatics","volume":"247 ","pages":"Pages 103-115"},"PeriodicalIF":1.7000,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Pragmatics","FirstCategoryId":"98","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0378216625001833","RegionNum":1,"RegionCategory":"文学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"0","JCRName":"LANGUAGE & LINGUISTICS","Score":null,"Total":0}

引用次数: 0

Abstract

This study investigates whether generative AI (represented by ChatGPT) can simulate or even emulate the stance expressed by human authors in the specific genre of academic book reviews. Through a comparative analysis of ChatGPT-generated reviews and human-authored reviews, this study focuses on the use of interactional metadiscourse markers (e.g., hedges, boosters, attitude markers, and self-mention) to reveal current AI's capabilities and limitations in handling complex evaluative discourse and interpersonal interaction. The findings indicate that ChatGPT overall employs interactional metadiscourse markers more frequently than human authors, due to its significant overuse of attitude markers. However, it underuses hedges and self-mention significantly, suggesting a reliance on explicit evaluative language while lacking nuanced caution and authorial presence. These results highlight that current AI's simulation of human writing is genre-sensitive but incomplete, particularly in achieving the balance of caution, conviction, and authorial presence, which is typical of human reviewers. The distinct metadiscoursal patterns identified may serve as linguistic fingerprints for distinguishing AI-generated reviews from human-authored ones. The study also offers pedagogical implications, emphasizing the need for educators and students to recognize current AI's limitations in modeling nuanced stance and fostering authentic authorial voice in evaluative genres.

查看原文本刊更多论文

人工智能可以模拟或模仿人类的姿势吗？使用元话语比较gpt生成的和人类撰写的学术书评

本研究探讨了生成式人工智能（以ChatGPT为代表）是否可以模拟甚至模仿人类作者在特定类型的学术书评中所表达的立场。通过对chatgpt生成的评论和人类撰写的评论的比较分析，本研究侧重于使用交互式元话语标记（如模糊限制语、助推器、态度标记和自我提及）来揭示当前人工智能在处理复杂评价话语和人际互动方面的能力和局限性。研究结果表明，由于过度使用态度标记，ChatGPT总体上比人类作者更频繁地使用互动元话语标记。然而，它明显没有充分利用模棱两可和自我提及，这表明它依赖于明确的评价语言，而缺乏细致入微的谨慎和作者的存在。这些结果突出表明，目前人工智能对人类写作的模拟是体裁敏感的，但不完整，特别是在实现谨慎、信念和作者存在的平衡方面，这是人类审稿人的典型特征。识别出的不同元话语模式可以作为区分人工智能生成的评论和人类撰写的评论的语言指纹。该研究还提供了教学意义，强调教育工作者和学生需要认识到当前人工智能在建模微妙立场和培养评估类型的真实作者声音方面的局限性。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Journal of Pragmatics Multiple-

CiteScore

3.90

自引率

18.80%

发文量

219

期刊介绍： Since 1977, the Journal of Pragmatics has provided a forum for bringing together a wide range of research in pragmatics, including cognitive pragmatics, corpus pragmatics, experimental pragmatics, historical pragmatics, interpersonal pragmatics, multimodal pragmatics, sociopragmatics, theoretical pragmatics and related fields. Our aim is to publish innovative pragmatic scholarship from all perspectives, which contributes to theories of how speakers produce and interpret language in different contexts drawing on attested data from a wide range of languages/cultures in different parts of the world. The Journal of Pragmatics also encourages work that uses attested language data to explore the relationship between pragmatics and neighbouring research areas such as semantics, discourse analysis, conversation analysis and ethnomethodology, interactional linguistics, sociolinguistics, linguistic anthropology, media studies, psychology, sociology, and the philosophy of language. Alongside full-length articles, discussion notes and book reviews, the journal welcomes proposals for high quality special issues in all areas of pragmatics which make a significant contribution to a topical or developing area at the cutting-edge of research.