An efficient decomposition of human-written summary sentence

Abstract

The task of human written decomposition is based on the pair of original documents and its summary to define these components in a summary sentence come from somewhere in the document. This task aims to satisfy three requirements as follows: 1) Whether it is constructed from cutting and pasting? 2) Which components in the sentence come from the original document? 3) Where in the document do the components come from? The result of a decomposition program is considered as training data and data evaluation for reduction and combination steps in cut and paste summarization system. We propose a method to enhance the accuracy of decomposition task through checking position and semantic measure for each word within a summary sentence. The model we used in the paper is extended from the hidden Markov model described by Hongyan Jing and K.R. MacKeown, (1999). Experiment results in the DUC data shows that the proposed method is efficient.