从元话语视角看生成式人工智能与学术语篇写作  

A Study of Generative AI in Academic Writing from the Perspective of Metadiscourse

作　　者：杨林秀[1] 平嘉鹏 Yang Linxiu;Ping Jiapeng

机构地区：[1]山西大学外国语学院,太原030006

出　　处：《当代修辞学》2024年第6期11-23,共13页Contemporary Rhetoric

基　　金：教育部人文社科规划项目“基于语料库的英汉学术语篇中作者自我呈现方式对比研究”(项目编号:23YJA740048);山西省教育教学改革课题“基于企业微信知识管理平台的研究生‘云学习共同体’的建设与实施研究”(项目编号:2022YJJG020)的阶段性成果。

摘　　要：2022年ChatGPT横空出世,越来越多的学者开始将生成式AI用于学术写作和翻译,那么生成式AI在语篇生产方面的“效”如何呢?基于此问题,本文尝试以元话语为观测点,通过分析生成式AI和人工写作的英语学术论文摘要中元话语使用的异同,探究生成式AI在语篇生成方面的“效”。研究发现,从语篇交际式元话语的使用情况来看,生成式文本具有较强的文内互动性,然而部分过渡词的语义衔接不当与框架标记语使用丰富度的不足,容易造成语义不连贯和影响句子理解的深度与广度的问题,且言据标记语的缺省容易造成学术不端等问题。从人际互动式元话语的使用情况来看,生成式文本同样显现出一定的不足。在人工智能生成的学术论文摘要中,因其对加强词与态度标记语的忽视会使语义表达模糊,从而削弱学术论文的权威性和有效性,但在参与标记语的使用上生成式AI表现得更胜一筹,为作者与读者搭建起良好的互动交流空间,体现出独特的“人性化”特征。本研究可以为自然语言的生成和处理提供实证依据和参考,也为今后更好地从元话语视角进行人机互动提供借鉴。In 2022,the emergence of ChatGPT prompted an increasing number of scholars to harness generative AI for academic writing and translation.So how effective is generative Al in discourse production?Based on this question,the study focuses on the analysis of the divergences and convergences in the use of metadiscourse within English research article abstracts written by generative AI and writers,aiming to scrutinize the“effectiveness"of generative Al in discourse production.From the perspective of interactive metadiscourse,it is shown that generative AI demonstrates a relatively high degree of intra-textual interaction.However,the inadequate semantic cohesion resulting from improper usage of hedges and the insufficient richness of frame markers may lead to semantic incoherence and give some negative impacts to the depth and breadth of sentence comprehension,and the absence of evidentials tends to raise ethical concerns related to academic misconduct.In the aspect of interactional metadiscourse,certain deficiencies also come to light in discourse produced by generative Al.For instance,the neglect of boosters and attitude markers introduces semantic vagueness,thereby undermining the rhetorical effectiveness of research article authors in asserting their research authority and validity.Nevertheless,generative AI exhibits a distinctive facet in its utilization of engagement markers,fostering a conducive interactive communication space for both research article writers and readers and exhibiting distinct“humanizing characteristics.The findings provide empirical evidence and serve as a reference for the generation and processing of natural language,and also serve as a valuable resource for future endeavors in enhancing human-machine interaction from the perspective of metadiscourse analysis.

关 键 词：元话语 生成式AI 人工写作 学术语篇 论文摘要 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] H05[自动化与计算机技术—控制科学与工程]