PolyFact: Parallel Multilingual Factual Multiple-Choice QA Dataset

A dataset of 100,000 facts aligned across 12 languages, grounded in Wikidata. Created by jvonrad and last updated on 2026-05-19. It is structured for both single-language evaluation and cross-lingual training, with facts fully aligned by a common fact_id.

Use Cases

• Training multilingual question-answering models based on aligned factual data.
• Evaluating cross-lingual model performance on factual reasoning tasks.
• Benchmarking knowledge alignment across languages using the parallel fact structure.
• Fine-tuning models for specific languages using the single-language splits.

Strengths

• Contains 100,000 distinct facts, providing substantial scale for training.
• Fully aligned across 12 languages, enabling direct cross-lingual comparisons.
• Grounded in Wikidata, providing a structured knowledge source.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-05-19).

Provenance

Source

Wikidata

Collection Method

Constructed as parallel multilingual multiple-choice QA from Wikidata facts.

Freshness

Last updated 2026-05-19 17:45:23.