The Constraint Dependency Grammar Software

Introduction

The WCDG-System is based on the Weighted Constraint Dependency Grammar formalism which describes natural language exclusively as dependency structure, i.e. ordered, labeled pairs of words in the input text. It performs natural language analysis under the paradigm of constraint optimization, where the analysis that best conforms to all rules of the grammar is returned. The rules are explicit descriptions of well-formed tree structures, allowing a modular and fine grained description of grammatical knowledge. In general these constraints are defeasible, since many rules about language are not absolute, but can be preempted by more important rules. The strength of constraining information is controlled by the grammar writer: fundamental rules must always hold, principles of different import have to be weighted against each other and general preferences that only take effect when no other disambiguating knowledge is available can be formulated in a uniform way.

Features

• Supports multiple levels of dependencies (e.g., Syntactic, Semantic roles or References of relative pronouns)
• Arbitrary information sources can be integrated into the grammar as predictors (e.g. POS-Taggers, PP-Attachers, Chunkers, visual context information or even other parsers)
• Anytime capable, i.e., the parser provides an analysis even if interrupted early
• Supports interactive grammar development
• Can assist in corpora annotation
• Diagnostic features: provides the grammatical constraints violated by a dependency analysis
• Incremental parsing mode available
• Syntactic prediction for incomplete sentences

News

2012-03-08: Java Reimplementation Released!

For the last few months we, NielsBeuck and ArneKoehn, have worked on a reimplementation of cdg in Java, called jwcdg. This work is now usable (but doesn't contain all the features of cdg).

The following things are missing in jwcdg at the moment:

• predictors (except the tagger predictor)
• a GUI
• a command line shell, you only get a simple command line interface

You can download jwcdg from our git repository on gitorious. You only need maven to compile it (everybody who tried to compile cdg in the past should be relieved by this!) and everything else is documented in the README.

You probably also want to download the lexicon, unpack it and place it in the ressources/ folder.

If you have any question, please send us an e-mail to _email0

2011-03-23: New Version Released

We are proud to present the newest version of CDG. Version 0.97 features:

• incremental parsing
• syntactic predictions for incomplete sentences
• several fixes and optimizations

You can download this version from the download page.

Contact

Wolfgang Menzel
Natural Language Systems
Department of Computer Science
University of Hamburg
Vogt-Kölln-Straße 30
22527 Hamburg
_email1