Can AI Understand Croatian Idioms? Assessing Large Language Models in Lexicographic Tasks
Can AI Understand Croatian Idioms? Assessing Large Language Models in Lexicographic Tasks
Author(s): Ivana Filipović Petrović, Slobodan BeligaSubject(s): Lexis, Semantics, Computational linguistics, South Slavic Languages, ICT Information and Communications Technologies
Published by: Inštitut za novejšo zgodovino
Keywords: artificial intelligence; large language models; lexicography; idioms; conceptual organisation;
Summary/Abstract: This paper explores the potential of ChatGPT in automating two lexicographic tasks within the Online Dictionary of Croatian Idioms (ODCI): (1) identifying semantic equivalents among idioms and (2) generating semantic fields for idiomatic expressions. The study evaluates how effectively AI can automate the process of distinguishing and grouping idioms by meaning, with the aim of reducing manual lexicographic work. As contemporary methodologies for employing language technologies continue to develop alongside technological progress, this research enhances understanding of AI capabilities in linguistic analysis. The findings suggest that ChatGPT shows considerable potential for conceptual organisation in lexicography. Nevertheless, challenges persist, especially regarding the unpredictable nature of AI-generated responses and the necessity for human post-editing
Journal: Prispevki za novejšo zgodovino (before 1960: Prispevki za zgodovino delavskega gibanja)
- Issue Year: 65/2025
- Issue No: 3
- Page Range: 218-242
- Page Count: 25
- Language: English
