Translation – Writing – ChatGPT: a linguistic analysis

Abstract

This study focuses on the translation and writing output of the ChatGPT (version 3.5) artificial intelligence language model in several languages. According to some research results, ChatGPT performs competitively with translation products (e.g. Google Translate) for high-resource European languages, but lags significantly for low-resource languages such as Hungarian. In the present study, the language products of the two AI tools (ChatGPT, Google Translate) were analysed by translating different types of texts. Both tools turned out to have advantages and disadvantages, with the quality of the translation depending on the text to be translated (e.g. type of text, style, topic), the languages to be translated, and the purpose of the translation. The study analyses the translation and writing performance of ChatGPT (version 3.5) through different text types, compared with the language products of Google Translate, taking into account the linguistic and stylistic characteristics of each text type: short introduction, cover letter, academic/technical text and literary text (poem). Different aspects of the linguistic data are analysed based on three languages (Hungarian, Italian, and English), with Hungarian as the reference language. The research aims to determine: (1) differences and similarities in the translation and writing of each type of text produced by the two AI tools; and (2) the characteristics that can be detected based on the analysis criteria, including spelling, lexical, grammatical, and pragmatic (language use) levels. The results of the analysis show that each type of text is produced in different linguistic qualities by the two AI tools (ChatGPT and Google Translate) created for different purposes.