Boosting Algorithm for Empty Node Recovery in a Sentence

Abstract

This paper presents a boosting algorithm for recovering of empty nodes in the
analysis of a sentence which are very essential in perceiving syntactic parsing.
The inputs of the boosting algorithm are sentences that are converted to parse
trees that lacked empty nodes (outputs of broad coverage syntactic parsers
such as Charniak’s parser and Collin’s parser) which are being taken by the
algorithm to produce parse trees with different kinds of empty nodes and their
antecedents. Evaluation metrics (precision, recall and F-score) were use in
order to highlight the differences on performance of recovering empty nodes on
the output of the parser with Penn Treebank analysis. The evaluation of the
boosting classifier tree (boosting algorithm) on the output of broad coverage
syntactic parsers and Penn Treebank achieves high F-score on most types of
empty nodes which leads to high parsing accuracy.