Franz Och's mkcls

mkcls is a suite of algorithms for clustering words based on bigram contextual similarity written by Franz Och. The quality of the clusters is quite good relative those produced by similar clustering algorithms.

The code was originally developed in the early 1990's, and it has been fixed to work with modern C++ compilers, but it is otherwise unchanged.

make
./run-mkcls.pl --help

This wrapper script helps set up the arguments to mkcls properly, and puts the output into a format that is similar to the one used by Percy Liang's Brown clustering implemention, which several tools use. This lets you use mkcls output in place of wcluster's, which can give slightly better results in some applications.

Details about what's going on inside mkcls, as well as citation information, can be found here.