Please use this identifier to cite or link to this item:
https://scholarhub.balamand.edu.lb/handle/uob/7791
Title: | A Linguistic Comparison between Human-and AI-Generated Texts in Lebanon | Authors: | Frangieh, Christelle Abdallah, Manal |
Affiliations: | Department of English Language and Literature | Keywords: | ChatGPT Human-generated texts Linguistic analysis LIWC Students |
Issue Date: | 2024 | Part of: | Centre for Applied Linguistics Research Journal | Volume: | 2024 | Issue: | 15 | Abstract: | On November 30, 2022, OpenAI launched ChatGPT-a leading chatbot that generates human-like output by using Natural Language Processing. While this novel technology has gained popularity worldwide, its integration in the educational field specifically has stirred controversies and mixed reactions. The major problem that many instructors have been recently facing is academic dishonesty in exams and written assignments, which led many institutions and professionals to redefine their policies and change their assessment methods. Considering the paucity of research exploring the language of ChatGPT, especially in Lebanon, the aim of this study is to provide an analysis of the linguistic features of AI-generated texts versus human-generated ones and highlight the divergence between the two. It is therefore essential to provide an insight into the capacities and limitations of AI in order to help instructors understand the platform the students use and spot, or at least question, any possible academic dishonesty. To derive the data corpus for this study, a total of 50 essays were collected from students in a private university in Lebanon and generated using ChatGPT 3.5 given the same instructions. Subsequently, a comparison between the AI-and human-generated writings was done through LIWC-22 (Linguistic Inquiry and Word Count)-a software for analysing word use-in order to investigate and interpret different variables such as pronouns, articles, positive and negative emotions, social words and other categories. Moreover, a comparison was made between different versions of ChatGPT-generated texts using the LMS (Language Matching Style) tool in LIWC to investigate the language of AI. The results have shown that, in terms of word frequencies, ChatGPT employed much more sophisticated words than students, who used simple and basic terminologies. Additionally, the detailed LIWC analysis ran for both has shown that AI tends to be more formal, objective and direct, while students’ writing remains more emotional and subjective. Last but not least, the LMS and LIWC analysis of AI-texts have shown very close percentages, denoting the similarities between the AI-generated texts. |
URI: | https://scholarhub.balamand.edu.lb/handle/uob/7791 | ISSN: | 20731175 | DOI: | 10.60149/TMSY9941 | Open URL: | Link to full text | Type: | Journal Article |
Appears in Collections: | Department of English Language and Literature |
Show full item record
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.