adobe/NLP-Cube

Implements end-to-end processing through stacked RNNs trained on Universal Dependencies treebanks, supporting 60+ languages with swappable model "flavours" (e.g., `en_ewt`, `en_lines`) loaded at runtime. Outputs CoNLL-U formatted annotations including morphological attributes and dependency relations, with version 3.0 removing embedding files for reduced memory footprint while maintaining accuracy across language variants.

563 stars. No commits in the last 6 months.