Constrained Semantic Decompression in LLMs through Persian Proverb-Conditioned Story Generation

In three linesStudy on story generation conditioned by Persian proverbs. Researchers introduce PAND (Proverb Aligned Narrative Dataset) and identify a 'decompression gap': LLMs produce fluent text but fail to preserve the moral and causal structure of proverbs. Explicit reasoning and iterative refinement partially mitigate these errors.

Summary generated by Claude — human-verified