rparse is a data-driven parser for Probabilistic Linear Context-Free Rewriting Systems (PLCFRS). It has been developed at the Emmy Noether group of Prof. Dr. Laura Kallmeyer at the University of Tübingen, Germany and is now maintained at her project "Grammar Formalisms beyond Context-Free Grammars and their use for Machine Learning Tasks" at the Department for Computational Linguistics at the Institute for Language and Information at the University of Düsseldorf, Germany (see http://phil.hhu.de/beyond-cfg). The project is sponsored by Deutsche Forschungsgemeinschaft (DFG).
In case of questions or comments, please contact Wolfgang Maier (maierw@hhu.de).
Contents
In order to reference this parser, please cite
Laura Kallmeyer and Wolfgang Maier (2013): *Data-driven Parsing using Probabilistic Linear Context-Free Rewriting Systems." Computational Linguistics, 39(1).
for a general description of the parser and constituency parsing evaluation,
Wolfgang Maier and Laura Kallmeyer (2010): Discontinuity and Non-Projectivity: Using Mildly Context-Sensitive Formalisms for Data-Driven Parsing. In: Proceedings of the 10th International Conference on Tree Adjoining Grammars and Related Formalisms (TAG+10), Yale University.
for grammar-based parsing of non-projective dependencies and its evaluation, and
Wolfgang Maier, Miriam Kaeshammer and Laura Kallmeyer (2012): Data-Driven PLCFRS Parsing Revisited: Restricting the Fan-Out to Two. In: Proceedings of the Eleventh International Conference on Tree Adjoining Grammars and Related Formalisms (TAG+11), Paris, France.
for parsing with a (2,2)-PLCFRS.
The code is released under the GNU General Public Licence (GPL) 2.0 or higher. The release include a copy of the library jgrapht, which is licensed under the GNU Lesser General Public License (LGPL) 2.1. The full license texts of the GPL 2.0 and the LGPL 2.1 can be found at http://www.gnu.org/licenses/gpl-2.0 and http://www.gnu.org/licenses/old-licenses/lgpl-2.1.
The parser is written in Java, Java 7 is required. In order to run it, you need the jgrapht library. jgrapht (http://jgrapht.org) is included with this release in the /lib directory. You have to compile the parser against your copy of jgrapht. This can be done using the ant build file included in the rparse package, to which you have to pass the location of the compiled jgrapht library as follows:
$ ant -Djgraph.path="/path/to/jgrapht"
This will build a jar file rparse.jar in the rparse package. The classpath in the manifest of the jar file will contain the path to jgrapht. If you want to use a different copy of jgrapht, pass both jars in the class path and use the rparse entry point:
de.tuebingen.rparse.ui.Rparse
The default format for training is the export format. In order to train the parser on NeGra, run something like the following:
$ java -jar rparse.jar \
-doTrain \
-train path-to-training-corpus \
-headFinder negra \
-saveModel path-to-trained-model
In order to parser something with the trained model, you need to pass the parser a POS tagged sentence (rparse cannot do POS tagging on its own yet). The default input format is one word+POS tag combination per line, separated by a slash. A call to the parser would look something like the following:
$ java -jar rparse.jar \
-doParse \
-test path-to-test-corpus \
-readModel path-to-trained-model
In order to evaluate your output, use:
$ java -jar rparse.jar \
-doEval \
-test path-to-test-corpus \
-readModel path-to-trained-model
Make sure you also check the output of java -jar rparse.jar -help.
New in version 2.0
Fast parser for (2,2)-PLCFRS. Can be accessed via -parserType cyktwo. You also should use -binType optimal for obtaining a low fan-out during binarization. Default in mode -doParse remains old parser (-parserType cyk).