The LFG Parsebanker

Victoria Rosén, Paul Meurer, and Koenraad de Smedt

Abstract

We present the LFG Parsebanker, a comprehensive toolkit for the interactive incremental construction of a treebank as a parsed corpus. The LFG Parsebanker supports a process flow involving automatic parsing with XLE and manual disambiguation by means of discriminants. In the latter respect, the toolkit is somewhat similar to Carter's Treebanker and the tools developed in the context of the Alpino project and the LinGO Redwoods initiative. Our toolkit, however, is specifically designed for LFG grammars. The underlying design and implementation of our LFG discriminants is described in our paper 'Designing and Implementing Discriminants for LFG Grammars' (this volume). The LFG Parsebanker has the following components:

1. XLE-Web, an interface to the XLE parser on a web page, which includes a new display of packed structures and offers discriminants for disambiguation;
2. a parsebanking page which offers views and disambiguation as in XLE-Web, but also additional parsebank management operations and a search window based on TigerSearch extended for f-structures;
3. an overview page supporting administration of and navigation in a chosen subcorpus as a whole;
4. a discriminant statistics page displaying statistics on all chosen discriminants in a subcorpus.

Most of these components are implemented in Common Lisp and use XML, XSLT and Javascript to serve the interface web pages. C-structure trees (and graphs) are drawn using Scalable Vector Graphics (SVG). The treebank is stored in a database and is searchable with TigerSearch.

More information is available at http://gandalf.aksis.uib.no/trepil.