Nhat Ha Nguyen

The University of Danang - University of Science and Education, Vietnam

University of Leeds, UK

Xuechun Huang

Thi Ngoc Yen Dang

University of Leeds, UK

Abstract

With the capacity to produce a large number of texts within a short time, ChatGPT can be an effective tool to develop learning materials for extensive reading activities. However, no studies have tested its potential. To address this gap, this study examined the lexical profile of ChatGPT-generated texts targeting learners at six CEFR levels from A1 to C2. We created six corpora from the texts generated by ChatGPT targeting learners with each CEFR level and used RANGE (Heatley et al., 2002) and Nation’s (2012) BNC/COCA 25,000-word family lists to analyse the lexical profile of these corpora. The results showed that regardless of the target CEFR levels, high-frequency words consistently constitute the largest percentage of ChatGPT-generated texts, followed by mid-frequency words and then low-frequency words. Moreover, ChatGPT-generated texts for lower levels (A1, A2 and B1) are less lexically demanding than those for higher levels (B2, C1 and C2). However, ChatGPT-generated texts for the A1 and A2 levels require the same vocabulary sizes as those targeting the B1 level, whereas the figures for C1 are slightly smaller than expected. Together, these findings indicate the potential of ChatGPT as a tool to create learning materials for learners with different proficiency but also highlight the importance of further checking and adjustment of ChatGPT generated texts to better tailor to learners’ language abilities.

Keywords

Vocabulary, ChatGPT, extensive reading, lexical demand