parser

Why we included this project

SuPar is a Python library for anyone who needs real parse trees instead of a sentence-level model. It reproduces a wide selection of academic parsers for dependency, constituency, and semantic dependency analysis, with pretrained models for more than nineteen languages. Rather than offering a single architecture, the package bundles several from the research literature: Biaffine, CRF-based, and tagging approaches all run against the same data, making side-by-side comparison straightforward. The same codebase includes highly parallelized implementations of structured-prediction algorithms such as linear-chain and semi-Markov CRFs and dependency CRFs, usable on their own without the parsers. For researchers and production teams that need precise grammatical structure, this is correct, reproducible code you do not have to reimplement from scratch.