AI and the Translation of Idioms – Challenges, Success, and a Corpus Perspective

Abstract

This study explores the capabilities of artificial intelligence in translating English idioms into Slovak, with and without contextual information. Using a dataset of 100 idioms and evaluating AI-generated translations against a validated bilingual dictionary of idioms, both qualitative and statistical analyses were employed. The results show an unexpectedly high accuracy in context-free translations, while context occasionally led to deterioration. McNemar’s test and a t-test confirmed a statistically significant shift in performance. The study highlights key advantages and limitations of AI, suggesting further research into reverse and cross-linguistic translation as well as employment of corpus-based methods.