Incremental Sigmoid Belief Network Dependency Parser (idp)
This is a configurable and trainable syntactic parser which achieves state-of-the-art results on various languages. It is based on a generative dependency parsing model which uses binary latent variables to induce conditioning features. If you use idp, please refer to (Titov and Henderson, IWPT 2007). See also (Titov and Hendersong, CoNLL 2007) for the evaluation details.
Incremental Sigmoid Belief Network Dependency Parser (idp) for synchronous Syntactic Dependency Parsing and Semantic Role Labeling for Multiple Languages
This model is an adaptation of (Henderson et al., CoNLL 2008) and (Titov et al., IJCAI 2009) Dependercy Parser based on Sigmoid Belief Network. It can be used to solve jointly Syntactic Dependency Parsing and Semantic Role Labeling for any language. It requires the corpus to be in the CoNLL 2009 format. If you use this model, please refer to (Gesmundo et al. 2009).
Temporal Restricted Boltzmann Machines for Dependency Parsing
A dependency parser as described in (Garg and Henderson 2011). The parser uses a transition based algorithm to make parsing decisions and employs an RBM at each time step. Directed neural network style connections between time steps provide context information to the later decisions. The program accepts input in CoNLL 2009 format and outputs the result in the same format.
A toolkit for distributed perceptron training and prediction described in (Gesmundo and Tomeh, 2012). The software is designed within the Map-Reduce paradigm. It reduces substantially execution times in handling huge data sets guarantying the same performance as the serial Perceptron. It can be executed as stand-alone software, extended, or integrated in complex systems.
A system for different tagging tasks based on bidirectional sequence classification and guided learning techniques, described in (Gesmundo, 2011). The system can be trained for different languages providing a high-level performance, as shown in (Gesmundo and Samardžić 2012).