AI Learns to Decode Arabic Metaphor with Cultural Context

Researchers developed CAMMAR, a new artificial intelligence framework that better understands Arabic metaphors by separating lexical, cultural, and figurative meanings. The team argues existing language models blend these concepts, leading to misinterpretations. CAMMAR organizes meaning into nested layers, mirroring a classical Arabic theory of rhetoric, and uses this structure to measure how metaphorical a phrase is.

The AI was tested on a newly created dataset of Arabic metaphors paired with their literal counterparts. Results indicate CAMMAR accurately identifies metaphors in 82.6% of cases where the same word appears in both forms, achieving an area under the curve of 0.84 with paired supervision.

The researchers found grounding the AI’s base layer in the morphological roots of words slightly improved performance. They plan to publicly release the datasets and code used in the study.