UzNER-Style

The corpus contains 75,000 manually annotated sentences collected from five functional styles of Uzbek: official, journalistic, scientific, literary, and conversational. Each style includes 15,000 sentences, making the dataset suitable for both in-domain NER evaluation and cross-style robustness analysis.

The dataset is annotated using the BIOES tagging scheme and covers a broad set of named entity categories, including persons, organizations, geopolitical entities, dates, money expressions, quantities, durations, percentages, locations, laws, documents, positions, events, products, languages, and other domain-relevant entity types. The inclusion of different Uzbek styles allows researchers to evaluate how NER models perform across diverse lexical, syntactic, and contextual conditions.