LexEd


Summary

Lexicon creation and maintenance are necessary aspects of natural language generation systems. Since there is a large amount of redundancy in the information contained in lexicons, computer assistance should be useful in containing the amount of work required. Given the lexicon data structures and relationships, such a system must be able to robustly retrieve and store lexicon information, and allow for the creation and maintenance of new forms unanticipated by the system's designers. In addition, in some cases it is desirable to have the Knowledge Engineer take over the duties of lexicon maintenance. In these cases, this system must also be able to be easily manipulated by a domain expert who in all likelihood knows nothing about linguistics.

I am developing LexEd, a robust Lexicon Creation and Maintenance System that will be able to retrieve and store lexical information from a lexicon that is integrated into a knowledge base. Its main feature is a graphical user interface for manipulating phrase structure grammar representations. By pointing and clicking, a user is able to create, from scratch or from template, the PSG representation for any type of clause or phrase. At any point the user can request the current surface string generated by this deep structure. When satisfied, the user can elect to have this representation stored into the lexicon with just one push of a button.

Theoretically this is an interesting problem because although the user has created a phrase structure representation, the actual lexicon is composed of functional descriptions, the unit of expression in Michael Elhadad's FUF/SURGE system. Thus in order to store a phrase tree into the lexicon, a transformation must be performed (PSG->SURGE). In addition, in order to edit an element of the lexicon, its SURGE representation must be transformed (SURGE->PSG) before it can be displayed.

LexEd is just such a system. Its PSG is based upon Dr. C. L. Baker's (Professor of Linguistics at UT-Austin) Phrase Structure Rules described in his book English Syntax, soon to be in its second edition. At this writing, all of Dr. Baker's rules have been entered into the system, and transformations from PSG to SURGE have been entered for roughly half of those. The following is an example of what we are currently able to transform:

The leaf, which is attached to the stem, ...

Screen capture:


Back to KBS Group Home Page
Creator:
theorist@cs.utexas.edu