Using ChatGPT for Teaching Second-Language Writing Skills: An Experimental Approach (original) (raw)

2024, Tejiendo palabras: explorando la lengua, la lingüística y el proceso de traducción en la era de la inteligencia artificial y la innovación docente

ChatGPT, a Language Model (LM) built by the US-based organization Open AI, has stirred notable attention since its release in late 2022. Trained on large sets of data, ChatGPT interacts with users in a conversational manner and can perform numerous tasks, but it is mostly known for its text-generating capabilities. It is in this respect that ChatGPT can be of interest for language teaching and linguistic research alike, as texts produced by the model at the instigation of students and/or instructors can be used for a variety of purposes. In that vein, this paper presents an experimental, ChatGPT-based approach to the teaching of writing skills devised by the authors for their English for Specific Purposes (ESP) courses. The approach draws on prior extensive experience gained in the fields of Corpus Linguistics (O’Keeffe et al., 2007), Data-driven Learning (DDL) (Boulton, 2010), Academic and Professional English (Alcaráz, 2000), and Second Language Learning (SLL) in general, which altogether may contribute to ground a theoretical framework for a new paradigm in the teaching and learning of languages as a result of the onset and establishment of ChatGPT. The experiment (still in progress) is set up within a Technical English course for Telecommunication Engineering undergraduates, where a control group and an experimental one have been established. The aim of the course is introducing students to the writing of summaries of scientific texts, and both groups have received instruction on the main characteristics of the genre. Handbooks on academic writing and other traditional, print materials have been used with the control group (48 informants), whereas a combination of those same materials and ChatGPT-generated model texts was used with the experimental group (39 informants). This introduces a variation in the DDL approach, as the language model of study is not retrieved from a linguistic corpus, but from the output obtained by giving specific prompts to ChatGPT. These prompts are suggested to students by the instructors, who guide the former through the entire process and assist them with observing, analysing and identifying patterns and characteristics of summaries as an academic writing genre. Ultimately, the aim of the research is to ascertain the impact of ChatGPT-derived instruction on students written performance. To that end, a comparable corpus has been compiled gathering a series of written outcomes produced by three different sets of informants such as students in the experimental group, students in the control group, and the chatbot. The language samples are systematized according to Corpus Linguistics principles (Eagles, 1996; Rea, 2010) which facilitate reliable quantitative and qualitative analysis through Corpus Linguistics tools. In addition, the availability of this corpus will allow for the contrastive analysis of human intelligence versus artificial intelligence language production. As stated before, the experiment is still in progress. However, preliminary observations may advance some tentative insights into different issues that are relevant for the study. A case in point might be a relatively salient trend for ChatGPT to use more emotionally charged language than that observed by the authors in their students’ text output.