Uzbek NER Gold: A Gold-Standard Dataset for Uzbek Named Entity Recognition

Description

Uzbek NER Gold is a gold-standard token-level named entity recognition dataset for the Uzbek language. The dataset contains 59,569 annotated token rows from 4,176 sentences and is provided in UTF-8 TSV format. Each row includes a sentence identifier, token order, token text, BIO-formatted NER tag, and POS/category label. The annotation schema covers eight named entity types: Person (PER), Organization (ORG), Location (LOC), Miscellaneous (MISC), Money (MONEY), Number (NUMERIC), Date/Time (TEMPORAL), and Work (WORK). The released dataset includes 9,250 named entity spans and 17 BIO labels plus the outside label O. The text was normalized to Uzbek Latin script, and Uzbek apostrophe variants were standardized. This dataset is intended for Uzbek NLP research and development, including named entity recognition, token classification, information extraction, and evaluation of Uzbek language models.

Steps to reproduce

1. Download the file `Uzbek_NER_Gold.tsv`. 2. Load the TSV file using UTF-8 encoding and tab (`\t`) as the delimiter. 3. Each row represents one token. Use the `Sentence` column to group tokens into sentences and the `TokenOrder` column to restore token order inside each sentence. 4. Use the `Token` column as the input text and the `NER_Tag` column as the BIO-formatted named entity label. 5. For Hugging Face Datasets, the file can be loaded with: from datasets import load_dataset dataset = load_dataset( "csv", data_files="Uzbek_NER_Gold.tsv", delimiter="\t", split="train", ) 6. For sequence-level NER experiments, group rows by `Sentence`, sort by `TokenOrder`, then convert each sentence into a list of tokens and a corresponding list of BIO labels.

Institutions

• Alisher Navo'i Tashkent State University of Uzbek Language and Literature

  Tashkent, Tashkent
• National University of Uzbekistan

  Tashkent, Tashkent

Categories

Licence