Michael Collins (home)

Software and Data Sets

• I've now released the source code for the parser described in my PhD thesis, under a GNU General Public License. Follow this link for a tar file that contains the code.

• You may also want to try Dan Bikel's parser. It has a number of great features: training of the original (WSJ treebank) models; training of Chinese and Arabic models; n-best parsing, and so on. In addition, Dan's code has a flag which gives output of the "events" files used in my original parser. If you need to retrain my parser, this is probably an excellent place to look.