What it does

MeTA is a C++ data-sciences toolkit for text analysis: tokenization with parse trees, compressed inverted/forward indexes, ranking functions, topic models, classification, graph algorithms, language models, and CRF-based POS tagging. It wraps liblinear and libsvm, supports UTF-8 for multilingual work, and runs multithreaded algorithms.

The interesting bit

The build guides are the real documentation here—exhaustive, platform-specific instructions for macOS, five Ubuntu versions, Arch, Fedora, CentOS, and Windows via AppVeyor. Someone clearly suffered through compiler hell so you don’t have to.

Key highlights

• Compressed indexes with pluggable caching strategies
• CRF implementation for POS tagging and shallow parsing
• UTF-8 support for non-English text analysis
• Multithreaded algorithms throughout
• Published ACL 2016 demo paper with official citation

Caveats

• Last meaningful activity appears to be 2016; Travis CI and AppVeyor badges suggest legacy CI infrastructure
• Requires jemalloc, ICU, and CMake 3.2+—not header-only or trivial to drop into existing projects

Verdict

Good fit if you’re doing reproducible NLP research in C++ and need a unified, citable toolkit. Skip it if you want Python bindings, GPU acceleration, or a project with active maintenance.