CONLLU

CoNLL-U, often written as CoNLLU, is a plain-text data interchange format for annotated corpora used by the Universal Dependencies project. It provides a standardized representation of morpho-syntactic annotations and dependency structures, enabling cross-lingual comparison and easy machine processing. CoNLL-U was designed as the successor to earlier CoNLL formats and is maintained by the UD community.

Data in CoNLL-U is organized as lines of tokens grouped by sentences. Each non-comment line contains ten tab-separated fields: ID, FORM, LEMMA, UPOS, XPOS, FEATS, HEAD, DEPREL, DEPS, and MISC. Comments begin with a hash (#) and can include sentence metadata. The ID field identifies tokens; ranges such as 1-2 denote multiword tokens. The HEAD field points to the syntactic head, and DEPREL marks the dependency relation. FEATS encodes morphological features, and DEPS carries enhanced dependencies; MISC stores miscellaneous annotations such as spacing.

CoNLL-U is the de facto interchange format for UD treebanks and is widely used for training and evaluating dependency parsers. Numerous corpora are published in CoNLL-U, and a range of tools can read and write it, including UDPipe and various programming libraries. The standard supports universal POS tags (UPOS) and language-specific tags (XPOS), promoting cross-lingual research and multilingual NLP applications.

History and scope: CoNLL-U emerged with the Universal Dependencies project in the 2010s as a unified format for syntactic and morphosyntactic annotation. It builds on earlier CoNLL formats and aligns with UD guidelines to enable comparability of treebanks across languages and the development of multilingual models.