lovit/soynlp

Based on the README, here's a technical summary: Employs unsupervised statistical approaches using cohesion scores, branching entropy, and accessor variety to extract word boundaries and identify nouns directly from corpus patterns without labeled training data. Provides multiple tokenization strategies (LTokenizer for left-right morpheme decomposition, MaxScoreTokenizer, RegexTokenizer) and includes a Point-wise Mutual Information module for analyzing word co-occurrence patterns, integrating with complementary libraries like soyspacing and KR-WordRank for spacing correction and keyword extraction.

984 stars and 122,443 monthly downloads. Available on PyPI.