/serbian-nlp-resources

Parsing, tagging and lemmatisation resources for Serbian

serbian-nlp-resources

Parsing, tagging and lemmatisation resources for Serbian

This repository contains the NLP resources for Serbian developed as part of the following work:

Miletic, Aleksandra (2018). Un treebank pour le serbe : constitution et exploitations. Thèse de doctorat. Université de Toulouse - Jean Jaurès, France.

These resources include:

  • ParCoTrain-Synt: a 101K token treebank, manually annotated with POS tags, detailed morphosyntactic tags, atomic morphosyntactic traits, lemmas, and syntactic dependencies.

  • ParCoLex: a morphosyntactic lexicon with 6M entries.

  • Transition-based parsing model: a parsing model created using the Talismane parser.

  • Graph-based parsing model: a parsing model created using the MST parser.

  • Morphosyntactic-tagging model: a POS tagging model using detailed morphosyntactic tags developed with the HunPOS tagger.

  • Lemmatization model: a lemmatization model developed with the CSTLemma lemmatizer.

The parsing and tagging models were trained on the ParCoTrain-Synt corpus, whereas the lemmatization model was trained on a training lexicon adapted from the ParCoLex lexicon.

The corpus and the models trained on it are distributed under the Creative Commons BY-NC-SA 3.0 license. The lexicon and the lemmatization model are distributed under the Creative Commons BY-SA 4.0 license.