Improved Long Sentences Compression Method Using Semantic Role Labeling

Abstract

Sentence compression is the task of compressing a long sentence into a short one. It retains the important contents and in the meantime it generates grammatical short sentences. This research proposed an improved semantic-based method to investigate semantic role labels in sentence compression. Moreover, in the process of compression, the approach is applied in Ziff–Davis dataset. It includes short sentences as summary for a corresponding long ones. In the process of compression we proposed to design what so called semantic patterns, so each sentence has been compressed in three ways. To evaluate the result, we used Jaccard similarity to compare between our approach and human short sentences on the Ziff–Davis dataset. The results showed that our proposed semantic-based method is recommended to be used for sentence compression.