PAC is a predicate-argument clustering software which is trained on tuples such as
2 | abandon | SUBJ:NP | company | project | |
1 | abstain | SUBJ | legislator | ||
1 | accede | SUBJ:P:NP | government | to | pact |
and induces a statistical soft clustering of these (and other) tuples. The statistical PAC model is described in the following paper:
Sabine Schulte im Walde, Christian
Hying, Christian Scheible, and Helmut Schmid:
Combining EM Training
and the MDL Principle for an Automatic Verb Classification
Incorporating Selectional Preferences,
ACL-HLT 2008, Columbus, Ohio. (pdf)
The source code of PAC can be downloaded here as a gzip-compressed tar file. It is freely available for education, research and other non-commercial purposes.