Figurative language—idioms, metaphors, and the like—has long been a stumbling block for machine translation systems, because the intended meaning of such expressions rarely matches a literal word-for-word rendering. A new paper posted on arXiv takes up this problem for Moroccan Darija, a colloquial Arabic dialect that remains under-resourced in natural language processing.
The study, titled "Towards Figurative Language Understanding in Moroccan Darija: An Evaluation of Large Language Models," focuses on how recent advances in LLMs handle these non-literal expressions. The abstract notes that while LLMs have improved many translation tasks, figurative language still requires going beyond literal translation, and the paper evaluates models specifically on this challenge.
Because the abstract is brief, the full findings are not yet available from the announcement. The paper's existence signals growing attention to dialectal Arabic and the specific difficulties that figurative speech creates for AI systems.