ARTIFICIAL INTELLIGENCE AS A LEXICOGRAPHICAL TOOL: ANALYTICAL STUDY ON THE PERFORMANCE OF CHATGPT, COPILOT, AND GEMINI IN SPANISH LEXICAL UNITS

Abstract

The popularization of generative artificial intelligence (GAI) tools has created unprecedented research scenarios in metalexicography, adding to those already derived from new ways of accessing lexical information. In this context, the present study analyses the lexicographical entries generated by three generative artificial intelligence tools (ChatGPT, Copilot, and Gemini) for twenty-four Spanish terms belonging to six different categories. A descriptive approach is proposed based on three aspects: types of information, ways in which it is presented, and similarity to the information provided by two monolingual dictionaries (Diccionario de la lengua española and Diccionario del español actual). An analysis was carried out based on ten variables related to the formal components of the lexicographical entry and two additional variables related to the sources of information and the presence of errors. Results show dissimilar behaviour for the formal elements in each of the GAI tools, with irregular performances across all dimensions addressed. The overall assessment of their performance allows us to conclude that, nowadays, these tools do not provide steady information, which makes human intervention still essential for developing lexical resources. Nevertheless, they might be valuable after post-editing by lexicographers or as a basis for didactical applications to improve lexical proficiency in language teaching, for instance. Finally, it should be noted that this study may help to shape valid methodologies for emerging research in these new environments.