Trankit is a lightweight, transformer-based Python toolkit for multilingual natural language processing, developed by the NLP group at the University of Oregon. Hosted on GitHub with about 800 stars under the Apache-2.0 license, its accompanying paper won the Outstanding Demo Paper Award at EACL 2021.
Trankit provides pretrained pipelines covering the core NLP annotation stack: sentence segmentation, tokenization, part-of-speech tagging, morphological tagging, lemmatization, named entity recognition, and dependency parsing. Its 1.0 release shipped more than 90 pretrained transformer pipelines for 56 languages, trained with XLM-Roberta large over the Universal Dependencies corpus. Despite the broad coverage, the toolkit is designed to stay light-weight compared with loading separate large models per task and language.
Using Trankit centers on a single Pipeline object: you choose a language, the toolkit downloads the appropriate pretrained model, and documents can then be annotated in one pass. Models share a multilingual transformer backbone and use adapter-style task modules, which is how the project keeps dozens of language pipelines practical to distribute and run. Published benchmark tables compare its accuracy against other toolkits across the Universal Dependencies treebanks, and the toolkit is a common baseline in multilingual NLP research and teaching.
Trankit is free and open source under the Apache-2.0 license. Pipelines and pretrained models are downloadable at no cost, and you run everything on your own hardware.
Trankit suits NLP researchers and developers building multilingual text pipelines โ for example in linguistics, information extraction, or cross-lingual analysis โ who want one consistent toolkit instead of a different library per language.