使用 Homebrew, MacPorts 安装 cabocha

项目链接

• 主页
• 上游文档
• 软件包管理器页面

命令和别名

• cabocha
• cabocha-config

项目历史

The official CaboCha page describes it as a Japanese dependency parser based on Support Vector Machines. Its cited research lineage includes the 2002 paper on Japanese dependency analysis using cascaded chunking by Taku Kudo and Yuji Matsumoto.

CaboCha offered more than a single CLI: the project page lists flexible input formats, user-redefinable features for dependency identification, user-trainable models, a Double-Array trie inside its dictionary machinery, named-entity analysis based on the IREX definition, and C, C++, Perl, and Ruby libraries.

The package sits in Taku Kudo's broader Japanese NLP tooling family. The install docs require MeCab and a dictionary such as mecab-ipadic, mecab-jumandic, or UniDic, and they require CRF++ 0.55 or later.

采用历史

CaboCha was widely known in Japanese NLP and corpus-processing workflows because it paired a scriptable CLI with trained models and published parser output formats. The official page's license notes explain that bundled model files were trained from Mainichi Newspaper CD-ROM data and carried separate usage constraints, while users could train their own models for other use cases.

In package-manager culture, CaboCha is the kind of older research-to-tooling artifact that survives because it remains useful for Japanese text-processing pipelines and because it integrates with MeCab, one of the best-known Japanese morphological analyzers.

使用方式

Users run cabocha on Japanese text from standard input and receive a simple dependency tree by default. The -f1 option emits a machine-oriented format with chunk and token information, and cabocharc controls parser, chunker, and named-entity model paths.

The install path supports Unix builds with configure, make, make check, and make install, and Windows builds that expect MeCab to be installed first. The config file can switch between IPA, JUMAN, and UniDic part-of-speech/model settings.

为什么软件包爱好者会关心

For package nerds, CaboCha is significant as a packaged research NLP tool: it wraps academic dependency parsing, model files, morphological analyzer dependencies, and C/C++ library bindings into a Unix-style command.

It also illustrates the packaging awkwardness of NLP tools with trained models: the code has an open software license, while bundled model data can carry corpus-specific redistribution and use constraints.

时间线

• 2002: Kudo and Matsumoto published the cascaded chunking Japanese dependency analysis work cited by CaboCha.
• 2013-12-31: CaboCha 0.67 improved analysis speed and fixed OS X and Windows issues.
• 2014-03-10: CaboCha 0.68 removed the tournament model and added a MeCab-compatible -u option.
• 2015-01-24: CaboCha 0.69 changed several Chunk fields to size_t and fixed a FreeList buffer overflow issue.

Related projects

• MeCab is the morphological analyzer required by CaboCha.
• CRF++ is required by the official CaboCha build instructions.
• ChaSen is explicitly noted as not supported by CaboCha's install instructions.
• JUMAN, mecab-ipadic, mecab-jumandic, and UniDic are related dictionary or part-of-speech ecosystems used around CaboCha models.

来源

• https://taku910.github.io/cabocha/
• https://taku910.github.io/crfpp/
• https://taku910.github.io/mecab/