Abstractive Summarization of Line Graphs from Popular Media

Proceedings of the Workshop on Automatic Summarization for Different Genres, Media, and Languages, pages 41–48, Portland, Oregon, June 23, 2011. c⃝2011 Association for Computational Linguistics Abstractive Summarization of Line Graphs from Popular Media Charles F. Greenbacker Peng Wu Sandra Carberry Kathleen F. McCoy Stephanie Elzer* Department of Computer and Information Sciences University of Delaware, Newark, Delaware, USA [charlieg|pwu|carberry|mccoy]@cis.udel.edu *Department of Computer Science Millersville University, Millersville, Pennsylvania, USA elzer@cs.millersville.edu Abstract Information graphics (bar charts, line graphs, etc.) in popular media generally have a dis- course goal that contributes to achieving the communicative intent of a multimodal docu- ment. This paper presents our work on ab- stractive summarization of line graphs. Our methodology involves hypothesizing the in- tended message of a line graph and using it as the core of a summary of the graphic. This core is then augmented with salient proposi- tions that elaborate on the intended message. 1 Introduction Summarization research has focused primarily on summarizing textual documents, and until recently, other kinds of communicative vehicles have been largely ignored. As noted by Clark (1996), language is more than just words — it is any signal that is intended to convey a message. Information graph- ics (non-pictorial graphics such as bar charts, line graphs, etc.) in popular media such as Newsweek, Businessweek, or newspapers, generally have a com- municative goal or intended message. For exam- ple, the graphic in Figure 1 is intended to convey a changing trend in sea levels — relatively flat from 1900 to 1930 and then rising from 1930 to 2003. Thus, using Clark’s view of language, information graphics are a means of communication. Research has shown that the content of informa- tion graphics in popular media is usually not re- peated in the text of the accompanying article (Car- berry et al., 2006). The captions of such graphics are also often uninformative or convey little of the graphic’s high-level message (Elzer et al., 2005). This contrasts with scientific documents in which graphics are often used to visualize data, with ex- plicit references to the graphic being used to explain their content (e.g., “As shown in Fig. A...”). Infor- mation graphics in popular media contribute to the overall communicative goal of a multimodal docu- ment and should not be ignored. Our work is concerned with the summarization of information graphics from popular media. Such summaries have several major applications: 1) they can be integrated with the summary of a multimodal document’s text, thereby producing a richer sum- mary of the overall document’s content; 2) they can be stored in a digital library along with the graphic itself and used to retrieve appropriate graphics in re- sponse to user queries; and 3) for individuals with sight impairments, they can be used along with a screen reader to convey not only the text of a docu- ment, but also the content of the document’s graph- ics. In this paper we present our work on summariz- ing line graphs. This builds on our previous efforts into summarizing bar …

Content truncated to 500 words