Automatic Slide Generation Based on Discourse Structure Analysis

Automatic Slide Generation Based on Discourse Structure Analysis Tomohide Shibata and Sadao Kurohashi Graduate School of Information Science and Technology, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan Abstract. In this paper, we describe a method of automatically gener- ating summary slides from a text. The slides are generated by itemizing topic/non-topic parts that are extracted from the text based on syntac- tic/case analysis. The indentations of the items are controlled according to the discourse structure, which is detected by cue phrases, identifica- tion of word chain and similarity between two sentences. Our experiments demonstrates generated slides are far easier to read in comparison with original texts. 1 Introduction A presentation with slides is so effective to pass information to people in many situations, such as an academic conference or business. Although some softwares, such as PowerPoint and Keynote, help us with making presentation slides, it is still cumbersome to make them from scratch. Some researchers have developed a method of (semi-)automatically making presentation slides from a technical paper or a news article [1, 2]. However, input texts of their systems were supposed to be documents whose structure is anno- tated: in [1], TEX source and in [2], semantically annotated documents by GDA tag. In this paper, we propose a method of automatically generating summary slides from a raw text. An example of a text is shown in Figure 1 and an example slide that is generated from the text is shown in Figure 2 (the translated slide is shown in Figure 3). In a slide, topic/non-topic parts that are extracted from the original text are itemized and each item is indented based on the discourse structure of the text. In particular, a big contrast/list structure in the text is an important clue for producing an easy-to-read slide. The outline of our procedure is as follows: 1. Inputsentences are processedbyJapanese morphologicalanalyzer,JUMAN[3], and are parsed by Japanese syntactic analyzer, KNP [4]. 2. Each sentence is segmented into clauses and the discourse structure of the text is analyzed. 3. Topic/non-topic parts are extracted from the text. 4. Summary slides are generated by displaying the topic/non-topic parts based on the discourse structure. R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 754–766, 2005. c ⃝ Springer-Verlag Berlin Heidelberg 2005 {shibata, kuro}@kc.t.u-tokyo.ac.jp Fig. 1. An example of a text Fig. 2. An example of a slide Our method not only helps us with making presentation slides but also creates a full-automatic presentation. That is to say, input texts are spoken via text- to-speech engine while presenting automatically generated summary slides. We call this system “text-to-presentation”, as illustrated in Figure 4. For the input of text-to-speech engine, written texts are not appropriate, because unnatural speech might be produced due to difficult words or long compound nouns, which are unsuitable for speech synthesis. Therefore, written texts are automatically converted into spoken texts based on paraphrasing technique [5, 6] and then are inputted into speech synthesis. The rest of this paper is organized as follows. …

Content truncated to 500 words