Lexical density in texts generated by ChatGPT: implications of artificial intelligence for writing in additional languages

Abstract

Technological advancement has had a significant impact on written production, especially in Additional Languages (ALs). Although technology has brought new opportunities for AL teaching, it also poses challenges, including concerns about the complexity of writing and the authenticity of students’ work. One such tool is ChatGPT, an artificial intelligence (AI) platform that has been the subject of debate since its popularization in 2022. This study analyses a corpus consisting of six tasks produced by ChatGPT in five languages (German, Spanish, French, Italian, and Portuguese), considering the proficiency levels proposed by the Common European Framework of Reference for Languages (CEFR), totalling 2991 texts and 706,401 words. The data were generated by students in a computer lab at a British university from 100 different profiles on the ChatGPT platform, following the researchers’ instructions. Data analysis employs Systemic Functional Linguistics (SFL) and the concept of lexical density ( Halliday, 1985, 1987, 1993; Halliday; Matthiessen, 2014) to investigate the complexity of the produced texts, as lexical complexity is related to proficiency in writing, where more advanced texts proportionally use more “content words” (nouns, verbs, adjectives, and some adverbs of manner). The results reveal that ChatGPT does not adhere to task instructions regarding the requested word count, thereby impacting the calculation of lexical density, nor does it produce texts that show significant differences in lexical density among additional languages and proficiency levels.

Keywords:
Additional languages; ChatGPT; Artificial Intelligence; Systemic Functional Linguistics; Lexical density