Universal NER v2: Towards a Massively Multilingual Named Entity Recognition Benchmark

Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)

Abstract

We present Universal NER (UNER) v2, a significant extension of the initial version released in 2024. UNER is a collaborative dataset for multilingual named-entity annotations, built to support research on NER methods in a cross-linguistic setting. UNER v2 adds 11 new datasets in 10 typologically varied languages to the resource, including multiple parallel evaluation benchmarks aligned with each other and other datasets in UNER v1, while maintaining the same annotation guidelines and high standards for inter-annotator agreement. We report detailed statistics for the dataset and benchmark UNER v2 using both encoder-based model architectures and LLMs.