OpenCCG: The OpenNLP CCG Library / News: Recent posts

The jsgf_compiler extension, developed for Knexus Research Corporation, has been released open source!

This extension adds the ccg2jsgf command line tool to OpenCCG 0.9.5 for compiling an OpenCCG grammar into a context-free grammar in the Java Speech Grammar Format used by the Sphinx speech recognizer. See the full README in the download for details.

By Jason's suggestion, development beyond version 0.9.5 is moving to github. Releases and binaries will continue to appear here on sourceforge.

See https://github.com/OpenCCG/openccg/wiki

A new version is out, with new features for dependency ordering and dependency length minimization in realization and support for using a very large 5-gram memory-mapped language model with KenLM on linux, as well as a new disjunctivizer package, for creating a disjunctive LF XML structure based on an LF graph difference.

The git repository has been updated to be in sync with the CVS for version 0.9.4, and so (at long last) CVS access has been disabled. From here, use git.

The latest release contains code and models for parsing and realizing English with CCGbank-trained models. There is also a new visualization tool for semantic dependency graphs and various bug fixes and other improvements.

We have just released a new version of OpenCCG that contains a lot of recent updates that are building tools around the core OpenCCG code, and which focuses on making it easier to create, edit, and visualize grammars. And this time around, there is actually documentation and tutorials for these things (thanks to an instructional technology grant from the College of Liberal Arts at the University of Texas at Austin). Most of it comes in the form of a wiki page which contains various tutorials and example grammars:... read more

I've implemented a more conventional two-stage packing/unpacking approach to realization, for direct comparison to the anytime approach. Initial experiments indicate that the anytime approach generally works better, though more comprehensive experiments are planned soon to verify these results. A nice feature of the packing approach though is that you can now turn unpacking off (in tccg) and view all possible realizations in the generation forest, pretty-printed as a regexp-style string. (NB: some effort is taken to "compact" the gen forest, but it's hard to eliminate all equivalent derivations without completely unpacking the structure.)... read more

The E.W. Beth Dissertation Prize for the year 2003 has been awarded to Jason Baldridge for his 2002 dissertation ``Lexically Specified Derivational Control in Combinatory Categorial Grammar''.

Congratulations Jason!

Read more about it at http://www.folli.uva.nl/beth.html

Congrats to Jason and GJ for winning the best paper award at EACL 2003, for their foundational work on multi-modal CCG!

Welcome to Gunes Erkan, who will be joining the dev team to work on implementing simple type hierarchies, which may also serve as a way to implement Cem Bozsahin's morphosyntactic modalities.

Version 0.8.1 has been released, which contains new XML schemas for validating the contents of the grammar files -- PLUS, these schemas have a reasonable amount of comments in them, and thus serve the dual role of providing reference documentation for the grammar formats (about time, you say? :-).

The first baseline release derived from the Grok codebase has been uploaded as OpenCCG version 0.8.0. Stay tuned for the development of some practical grammars!

Mike is repurposing the Grok library for primary use as a CCG realizer in limited domain dialogue systems. CCG parsing will of course remain part of the picture too, but at present this will not be emphasized in ongoing development.
See http://www.iccs.informatics.ed.ac.uk/~mwhite/White-Baldridge-ENLG-2003-to-appear.pdf
for a description of the effort so far.