UDcorpora

UDcorpora is the collection of annotated corpora produced under the Universal Dependencies (UD) project. It provides multilingual corpora annotated with universal part-of-speech tags, lemmas, and dependency relations, all following the UD annotation guidelines. The aim is to enable cross-linguistic comparison and support research in natural language processing, corpus linguistics, and language typology.

Content and format: UDcorpora encompasses a wide range of treebanks across many languages and dialects, including English, Spanish, Chinese, Russian, Arabic, and others. The data are distributed in the CoNLL-U format, with metadata detailing language, dataset source, version, and annotation guidelines. Updates add new languages, reforms, and corrections to maintain consistency and coverage.

Access and licensing: UDcorpora is released under an open license and freely accessible through the UD website and related repositories. The data are maintained by the UD Consortium and contributing institutions, with versioned releases to ensure reproducibility and traceability of experiments.

Usage and significance: The corpora are used to train and evaluate dependency parsers, perform cross-linguistic analyses, and benchmark parsing systems. UDcorpora serves as a standard resource for comparing parsing performance across languages under a common annotation scheme, supporting both research and teaching activities.

Relation to UD resources: UDcorpora is integral to the UD project, designed to be compatible with UD guidelines and linked to other UD resources such as UD Core and language-specific treebanks. See also Universal Dependencies, CoNLL-U, and language treebanks.