RDF-Based Structured Quality Assessment Representation of Multilingual LLM Evaluations

TL;DR

This paper tackles the challenge of reliably evaluating multilingual LLM outputs under conflicting or incomplete context. It introduces an RDF-based evaluation framework and vocabulary designed for FAIR compliance to capture multilingual assessments across four context conditions. The study, conducted in the fire safety domain with GPT-4o-mini and Gemini-2.0-Flash in German and English, reveals systematic patterns of context prioritization and language-specific behavior. By enabling standardized, queryable analyses of knowledge leakage and cross-lingual consistency, the work supports reproducible evaluation across languages and domains.

Abstract

Large Language Models (LLMs) increasingly serve as knowledge interfaces, yet systematically assessing their reliability with conflicting information remains difficult. We propose an RDF-based framework to assess multilingual LLM quality, focusing on knowledge conflicts. Our approach captures model responses across four distinct context conditions (complete, incomplete, conflicting, and no-context information) in German and English. This structured representation enables the comprehensive analysis of knowledge leakage-where models favor training data over provided context-error detection, and multilingual consistency. We demonstrate the framework through a fire safety domain experiment, revealing critical patterns in context prioritization and language-specific performance, and demonstrating that our vocabulary was sufficient to express every assessment facet encountered in the 28-question study.

Figures (1)

• Figure 1: Simplified visual representation of the RDF vocabulary for LLM evaluations.