Guessing the grammatical function of a non-root f-structure in LFG

Abstract

Lexical-Functional Grammar (Kaplan and Bresnan, 1982) f-structures are bilexical labelled dependency representations. We show that the Naive Bayes classifier is able to guess missing grammatical function labels (i.e. bilexical dependency labels) with reasonably high accuracy (82–91%). In the experiments we use f-structure parser output for English and German Europarl data, automatically “broken” by replacing grammatical function labels with a generic UNKNOWN label and asking the classifier to restore the label.