AI-generated vs human-authored texts: A multidimensional comparison

Applied Corpus Linguistics Pub Date : 2023-12-20 DOI:10.1016/j.acorp.2023.100083

Tony Berber Sardinha

{"title":"AI-generated vs human-authored texts: A multidimensional comparison","authors":"Tony Berber Sardinha","doi":"10.1016/j.acorp.2023.100083","DOIUrl":null,"url":null,"abstract":"<div><p>The goal of this study is to assess the degree of resemblance between texts generated by artificial intelligence (GPT) and (written and spoken) texts produced by human individuals in real-world settings. A comparative analysis was conducted along the five main dimensions of variation that Biber (1988) identified. The findings revealed significant disparities between AI-generated and human-authored texts, with the AI-generated texts generally failing to exhibit resemblance to their human counterparts. Furthermore, a linear discriminant analysis, performed to measure the predictive potential of dimension scores for identifying the authorship of texts, demonstrated that AI-generated texts could be identified with relative ease based on their multidimensional profile. Collectively, the results underscore the current limitations of AI text generation in emulating natural human communication. This finding counters popular fears that AI will replace humans in textual communication. Rather, our findings suggest that, at present, AI's ability to capture the intricate patterns of natural language remains limited.</p></div>","PeriodicalId":72254,"journal":{"name":"Applied Corpus Linguistics","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2023-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.sciencedirect.com/science/article/pii/S2666799123000436/pdfft?md5=eec63f0662cd28b0d80ac041ac33eae7&pid=1-s2.0-S2666799123000436-main.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Applied Corpus Linguistics","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2666799123000436","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}

引用次数: 0

Abstract

The goal of this study is to assess the degree of resemblance between texts generated by artificial intelligence (GPT) and (written and spoken) texts produced by human individuals in real-world settings. A comparative analysis was conducted along the five main dimensions of variation that Biber (1988) identified. The findings revealed significant disparities between AI-generated and human-authored texts, with the AI-generated texts generally failing to exhibit resemblance to their human counterparts. Furthermore, a linear discriminant analysis, performed to measure the predictive potential of dimension scores for identifying the authorship of texts, demonstrated that AI-generated texts could be identified with relative ease based on their multidimensional profile. Collectively, the results underscore the current limitations of AI text generation in emulating natural human communication. This finding counters popular fears that AI will replace humans in textual communication. Rather, our findings suggest that, at present, AI's ability to capture the intricate patterns of natural language remains limited.

查看原文

微信好友朋友圈 QQ好友复制链接

本刊更多论文

人工智能生成的文本与人类撰写的文本：多维比较

本研究的目的是评估人工智能生成的文本（GPT）与人类在现实世界中生成的（书面和口语）文本之间的相似程度。我们按照 Biber（1988 年）确定的五个主要变化维度进行了比较分析。研究结果表明，人工智能生成的文本与人类撰写的文本之间存在明显差异，人工智能生成的文本通常无法表现出与人类文本的相似性。此外，为测量维度分数在识别文本作者身份方面的预测潜力而进行的线性判别分析表明，人工智能生成的文本可以根据其多维特征相对容易地识别出来。总之，这些结果强调了目前人工智能文本生成在模拟人类自然交流方面的局限性。这一发现反驳了人们对人工智能将在文本交流中取代人类的担忧。相反，我们的研究结果表明，目前人工智能捕捉自然语言复杂模式的能力仍然有限。

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分钟内获得全文去求助

来源期刊