A sentence compaction experiment based on phrase significance and inter-phrase dependency

摘要

Most of text summarization methods reported so far are based on the idea of selecting important sentences from a set of given sentences. However, it is sometimes necessary to shorten each sentence. In such a case, it is important that a shortened sentence is natural as a Japanese sentence. In this paper we present results of sentence compaction experiments using phrase significance score and inter-phrase dependency score. The results show that information of phrase category frequency is effective for inter-phrase dependency score calculation, and that phrase significance score has more influence on the results in higher compaction rate cases.