ISBN dependency parser (idp) - Incremental Sigmoid Belief Network Dependency Parser
(C) Ivan Titov and James Henderson, 2007.
University of Geneva and University of Edinburgh.
ISBN dependency parser (idp) is a configurable and trainable dependency
parser based on a generative statistical model, ISBNs (Incremental Sigmoid Belief
Networks).
The dependency parser is described in:
A Latent Variable Model for Generative Dependency Parsing
Ivan Titov, James Henderson
International Conference on Parsing Technologies (IWPT-07). Prague, Czech Republic, 2007.
Its evaluation in CoNLL-2007 shared task on dependency parsing is presented in:
Fast and Robust Multilingual Dependency Parsing with a Generative Latent Variable Model
Ivan Titov, James Henderson
CoNLL 2007 Shared Task. Joint Conf. on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL-07), Prague, Czech Republic, 2007. (3rd result out of 23)
Scores of idp in CoNLL 2007 Shared Task on Dependency parsing are presented here (see "Titov et al").
You can configure the feature model and the structure of the graphical model (the pattern of interconnections between latent state vectors) to reflect properties of the language or/and the treebank.
idp is free software, you can redistribute it and/or modify it under the terms of the GNU General Public License. Additionally, if you use the ISBN Dependency Parser (idp) in any your publication, you are required to cite (Titov and Henderson, IWPT 2007) as shown above. The GNU General Public License does not permit this software to be redistributed in proprietary programs. This software is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
[Download idp 0.0.4 and documentation]
If you downloaded version 0.0.3 (which was available between April 5 and April 10, 2008) please upgrade to 0.0.4. The version 0.0.3 introduced a major bug leading to substantial performance degradation.