FFE-Hallu:Hallucinations in Fixed Figurative Expressions:Benchmark of Idioms and Proverbs in the Persian Language

TL;DR

FFE-Hallu introduces the first benchmark focused on figurative hallucination in fixed figurative expressions for Persian, combining generation, detection, and cross-lingual translation tasks. The method tests whether LLMs can retrieve authentic FFEs from meanings, distinguish fabricated FFEs, and produce culturally appropriate Persian equivalents, using a carefully annotated 600-item dataset. Key findings show that GPT-4.1 generally offers the best performance across tasks, yet persistent hallucinations and translation gaps reveal substantial gaps in figurative and cultural grounding, especially among open-weight models. The work also evaluates automatic judging for figurative language and finds that judge quality strongly affects reliability, highlighting the need for stronger, culturally aware evaluators. Overall, FFE-Hallu provides a diagnostic framework and dataset to drive development of hallucination-resistant, culturally informed multilingual models.

Abstract

Figurative language, particularly fixed figurative expressions (FFEs) such as idioms and proverbs, poses persistent challenges for large language models (LLMs). Unlike literal phrases, FFEs are culturally grounded, largely non-compositional, and conventionally fixed, making them especially vulnerable to figurative hallucination. We define figurative hallucination as the generation or endorsement of expressions that sound idiomatic and plausible but do not exist as authentic figurative expressions in the target language. We introduce FFEHallu, the first comprehensive benchmark for evaluating figurative hallucination in LLMs, with a focus on Persian, a linguistically rich yet underrepresented language. FFEHallu consists of 600 carefully curated instances spanning three complementary tasks: (i) FFE generation from meaning, (ii) detection of fabricated FFEs across four controlled construction categories, and (iii) FFE to FFE translation from English to Persian. Evaluating six state of the art multilingual LLMs, we find systematic weaknesses in figurative competence and cultural grounding. While models such as GPT4.1 demonstrate relatively strong performance in rejecting fabricated FFEs and retrieving authentic ones, most models struggle to reliably distinguish real expressions from high quality fabrications and frequently hallucinate during cross lingual translation. These findings reveal substantial gaps in current LLMs handling of figurative language and underscore the need for targeted benchmarks to assess and mitigate figurative hallucination.

Figures (1)

• Figure 1: Fake FFE Detection accuracy by model and category for two prompt settings.