Skip to content
Publications

Dataset · Figshare · November 11, 2026

Gold-set and Probe for LLM Interpretation of Contested Images in Classical Uzbek Poetry

Shaxlo Masutova, Farida Tolibova, Makhbuba Mukhammadova, Gulnoza Salimova, Shoira Axmedova

Abstract

This package accompanies the paper “Mystical-Aware but Not Always Accurate: Probing LLM Interpretation of Contested Images in Classical Uzbek Poetry (A Low-Resource Case Study)”. It contains a gold-set of 13 contested classical-Uzbek (Chagatai) poetic images, each with an example couplet from Alisher Navoi, a literal surface sense and an expert figurative/mystical reading drawn from published Uzbek commentary; the exact prompt and run settings used to probe DeepSeek (web interface, search and reasoning disabled, a fresh session per item, three runs); the raw model responses for all three runs; and the per-item verdicts behind the reported results (literal-collapse rate and error taxonomy). The couplets are public-domain classical poetry. See README.md for the file list and the scoring scheme.

Topic ModelingNatural Language Processing TechniquesDigital Humanities and Scholarshiplarge language modelsUzbek classical poetryAlisher Navoi

More by Gulnoza Salimova