Understanding Human Sentence Compression Pattern for Malay Text Summarizer

Abstract

This paper focuses on understanding the discovered human sentence compression patterns from the developed Malay summary corpus to produce a more compact and informative sentence in an automated summary. The extractive method in text summarization tends to include unnecessary sentences alongside the important ones during the sentence selection process, which affected the summary's coherent and readability. Most techniques in Sentence Compression relies on heavy syntactic knowledge and external resources to decide on the compression decision, but still fail short in preserving the informativeness of the summary. From the automatic and manual evaluation, the compressed summary produced by our MYTextSum model by referring to the discovered human sentence compression pattern named Frequent Eliminated Pattern (FASPe) has shown promising results with the improvement as high as 23.5% against the uncompressed summary method.