self-attentive-parser

Why we included this project

The Berkeley Neural Parser (installed as benepar) recovers the nested phrase structure of sentences, the kind of grammatical analysis that search, text analytics, and NLP preprocessing pipelines often want on top of plain tokenization and tagging. It ships with pretrained models for eleven languages, drops into an existing spaCy pipeline with a single pip install, and also works through an NLTK interface. Because the parser uses a self-attentive encoder, it doubles as a useful reference for researchers looking at modern parsing architectures. Training paths are included, so you can retrain on your own treebank when the stock models do not fit, and the spaCy integration keeps the rest of your annotation workflow intact. If you need linguistically structured output rather than just embeddings, this package covers territory that general LLM tooling does not.