0100487714

open access

Version of Record

A Study on Differences between GPT-generated and ENS-written Essays on Doing a Part-time Job and Smoking in a Restaurant: Utilizing Corpus Linguistics Methodologies

English (英語)

Learner Corpus Studies in Asia and the World

6

39-58

神戸大学国際コミュニケーションセンター

2024-03-20

The present study focused on the differences between GPT-generated essays and essays written by English native speakers, seeking some implications for English education. In the analysis utilizing the corpus linguistics methodologies, the 100 GPT-generated essays and the 400 essays written by English native speakers (contained in the ICNALE) were compared in terms of the use of some function words such as conjunctions, personal pronouns, prepositions, and modal verbs. The themes of the essays were “the pros and cons of college students doing a part-time job” and “the pros and cons of smoking in restaurants”, which are the topics used in the ICNALE. The results showed some points, namely, ChatGPT tends to generate the same phrases repeatedly, using the words such as can, and, by, their, with, and, for many times. In contrast, it generates some negative keywords such as conjunctions (if, that, or, when, and as), personal pronouns (I, they, my, it, and us), prepositions (to, as, before, about, and since), modal verbs (will, would, might, and could), and adverb particles (up, out, and off). The present study analyzed the function words above in detail, yielding some implications for utilizing ChatGPT in language education in EFL classrooms.

Academic Writing

紀要論文