Sentence Compression for Minutes Creation Using Dependency Structure-based Relation Extraction

Abstract

We present a rule-based method for sentence compression by utilizing relation extraction which relies on a dependency tree structure. To create good meeting minutes, it is required to summarize important items such as decision and policy with minimum number of words. Our method first extracts a relation between text phrases as subtrees in the dependency tree, and then generates a sentence by combining the extracted subtrees, which results in creating a shorter sentence that contains only important items. In evaluation, we showed that the method has capability to create shorten sentences with high precision compared to a baseline method and an original sentence.