---

license: cc-by-nc-4.0

SKR1 - Benchmark for Testing Knowledge about Slovak Realia for Large Language Models

Overview

SKR1 is a specialized benchmark designed to evaluate Large Language Models' knowledge of Slovak cultural and factual context. Developed by Marek Dobeš at ČZ o.z., this benchmark addresses the significant gap in culturally-specific evaluations for underrepresented languages like Slovak.

Key Features

• 35 carefully crafted questions covering four domains of Slovak realia:

  • Culture (15 items): Literature, music, film, and visual arts
  • Geography (10 items): Landmarks, regions, and natural features
  • History (5 items): Political figures, events, and milestones
  • Language (5 items): Dialectal nuances, idioms, and wordplay

• Human-generated content ensuring cultural authenticity

• Unambiguous answers suitable for automated evaluation

• Comprehensive coverage of Slovak-specific knowledge

Benchmark Results

Evaluation of three major LLMs revealed significant gaps in cultural knowledge:

Model	Overall Accuracy	Culture	Geography	History	Language
DeepSeek V3	51%	33%	70%	60%	60%
GPT-4o	49%	47%	60%	40%	40%
Llama 3.3 70B	37%	33%	50%	40%	20%

Significance

This benchmark demonstrates that even state-of-the-art LLMs struggle with culturally specific knowledge, correctly answering only about half of the questions. The results highlight the need for specialized, nationally tailored language models and better cultural inclusivity in AI systems.