Leo is a project being initiated by RIALIST, NASA's natural language research group, in conjunction with the Center for the Study of Language and Information (CSLI) at Stanford. The initial goal of Leo is to provide an architecture for defining XML specifications of grammars for different natural language parsing systems such as Gemini (SRI), LKB (Stanford), and Grok (Edinburgh). Leo will thus act as a means for these different systems to share natural language grammar resources in a robust, automated manner. As part of its functionality, Leo will also facilitate the creation of speech recognition grammars for all of these systems to permit a tighter link between recognition and understanding grammars in spoken natural language dialogue interpretation applications.

Leo is entering life as part of the OpenNLP project. OpenNLP acts as an organizational center and banner under which various NLP projects will be carried out. It also defines a Java API for natural language components.

Ultimately, it is envisioned that Leo will become a natural language processing system in its own right, and will be comparable to Grok except that it will parse with phrase structure based grammars rather than categorial grammars. Having these two similar systems, built with interoperability in mind, will facilitate the computational evaluation of these different grammar formalisms and strategies for using them in parsing.

The initial code and XML specifications have been started only recently, but there is already a code base available with utilities for the Type Definition Language and Typed Unification Grammars (Gemini style). If you are interested in getting involved in this project as a developer or planner, please contact either Jason Baldridge (jmb@cogsci.ed.ac.uk) or John Dowding (jdowding@riacs.edu). You can also post questions or comments to the forums of the Leo project area on Sourceforge.

For more details, read Leo: an Architecture for Sharing Resources for Unification-Based Grammars