Improving Chronological Sentence Ordering by Precedence Relation

Improving Chronological Sentence Ordering by Precedence Relation Naoaki OKAZAKI The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan okazaki@miv.t.u-tokyo.ac.jp Yutaka MATSUO AIST 2-41-6 Aomi, Koto-ku, Tokyo 135-0064, Japan y.matsuo@carc.aist.go.jp Mitsuru ISHIZUKA The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan ishizuka@miv.t.u-tokyo.ac.jp Abstract It is necessary to find a proper arrangement of sen- tences in order to generate a well-organized sum- mary from multiple documents. In this paper we de- scribe an approach to coherent sentence ordering for summarizing newspaper articles. Since there is no guarantee that chronological ordering of extracted sentences, which is widely used by conventional sum- marization system, arranges each sentence behind presupposed information of the sentence, we improve chronological ordering by resolving antecedent sen- tences of arranged sentences. Combining the re- finement algorithm with topical segmentation and chronological ordering, we address our experiment to test the effectiveness of the proposed method. The results reveal that the proposed method improves chronological sentence ordering. 1 Introduction The growth of computerized documents enables us to find relevant information easily owing to technological advances in Information Retrieval. Although it is convenient that we can obtain a great number of documents with a search en- gine, this situation also presents the information pollution problem: “Who is willing to take the tedious burden of reading all those text docu- ments?” Automatic text summarization (Mani, 2001), is one solution to the problem, providing users with a condensed version of the original text. Most existing summarization systems make use of sentence or paragraph extraction, which finds significant textual segments in source doc- uments, and compile them in a summary. After we select significant sentences as a material for a summary, we must find a proper arrangement of the sentences and edit each sentence by delet- ing unnecessary parts or inserting necessary ex- pressions. Although there has been a great deal of research on extraction since the early stage of natural language processing (Luhn, 1958), research on post-processing of automatic sum- marization is relatively small in number. It is essential to pay attention to sentence ordering in case of multi-document summarization. Sen- tence position in the original document, which yields a good clue to sentence arrangement for single-document summarization, is not enough for multi-document summarization because we must consider inter-document order at the same time. In this paper we propose an approach to co- herent text structuring for summarizing news- paper articles. We improve chronological order- ing, which is widely used by conventional sum- marization system, complementing presupposed information of each sentence. The rest of this paper is organized as follows. We first review the sentence ordering problem and present our approach to generate an acceptable ordering in the light of coherence relation. The subsequent section (Section 3) addresses evaluation metrics and experiment results. In Section 4 we discuss future work and conclude this paper. 2 2.1Sentence Ordering Sentence ordering problem Our goal is to determine the most probable per- mutation of given sentences and to generate a well-structured text. When a human is …

Content truncated to 500 words