Syntactic Parsing
Syntactic parsing is the process of clarifying the structure of sentences and the dependency relationships between words. In Japanese NLP, two approaches coexist: the traditional dependency parsing and the more recent Universal Dependencies (UD) framework-based analysis.
3.1 Dependency Parsing
CaboCha (taku910/cabocha): A Japanese dependency parser by the same author as MeCab, Taku Kudo, based on SVM (Support Vector Machine). Stars: 121. Implemented in C++ with Python bindings available.
KNP (ku-nlp/knp): A Japanese syntactic, case, and coreference analysis system developed by Kyoto University. Used in combination with Juman++. Stars: 35. Its ability to perform case relation analysis and coreference resolution is a distinctive feature.
J.DepP (jdepp-python, jdeppy): Python bindings for a fast Japanese dependency parser.
3.2 Universal Dependencies Support
Universal Dependencies (UD) is a cross-lingual dependency structure framework that has become the standard for multilingual NLP.
| Tool | Weekly DL | Total DL | Stars | Description |
|---|---|---|---|---|
| GiNZA | 11k | 3M | 878 | spaCy-based UD-compliant pipeline |
| UniDic2UD | 72 | 342k | 38 | UD conversion using UniDic dictionary |
| esupar | 384 | 189k | 56 | BERT/RoBERTa/DeBERTa-based |
| SuPar-UniDic | 145 | 126k | 21 | UD parsing with BERT model |
GiNZA is a Japanese NLP library built on the spaCy framework, providing UD-compliant dependency parsing. With Stars: 878, it is the highest-rated among parsing tools, and 18k weekly downloads demonstrate its wide adoption. It is developed and maintained by Megagon Labs (Recruit Institute of Technology).
esupar (by Yasunori Yasuoka) is a tokenizer, POS tagger, and dependency parser leveraging BERT/RoBERTa/DeBERTa models, and also supports languages other than Japanese. Active updates continue.
3.3 CCG Parsers
Combinatory Categorial Grammar (CCG) is a parsing framework based on formal grammar theory.
depccg (masashi-y/depccg): An A* CCG parser equipped with supertag and dependency factoring models. Stars: 104.
lightblue (daisukebekki/lightblue): A CCG parser using DTS (Dependent Type Semantics) representations. Stars: 29. Updates continue as recently as February 2026.
3.4 BERT-based Parsing
bertknp (ku-nlp/bertknp): A BERT-based Japanese dependency parser, extending the KNP approach with BERT. Stars: 23.
KWJA (ku-nlp/kwja): A unified Japanese language analyzer from Kyoto University, integrating morphological analysis, syntactic parsing, predicate-argument structure analysis, and more. Stars: 150. Weekly downloads: 612.