Treebank-based acquisition of wide-coverage, probabilistic LFG resources: project overview, results and evaluation

Abstract

This paper presents an overview of a project to acquire wide-coverage, probabilistic Lexical-Functional Grammar (LFG) resources from treebanks. Our approach is based on an automatic annotation algorithm that annotates “raw” treebank trees with LFG f-structure information approximating to basic predicate-argument/dependency structure. From the f-structure-annotated treebank we extract probabilistic unification grammar resources. We present the annotation algorithm, the extraction of lexical information and the acquisition of wide-coverage and robust PCFG-based LFG approximations including long-distance dependency resolution. We show how the methodology can be applied to multilingual, treebank-based unification grammar acquisition. Finally we show how simple (quasi-)logical forms can be derived automatically from the f-structures generated for the treebank trees.