jdkato/prose

Built entirely in pure Go with no external dependencies, prose implements a modular NLP pipeline (tokenization → POS tagging → NE extraction) with functional options to disable stages as needed. Its sentence segmenter achieves 75% accuracy on the Golden Rules benchmark while executing 4× faster than Stanford CoreNLP, and its POS tagger outperforms NLTK's implementation (96.1% vs 89.3% accuracy) on the Treebank corpus. The tokenizer handles modern text artifacts like URLs, mentions, hashtags, and emoticons as distinct tokens.

3,069 stars. No commits in the last 6 months.