Automatic generation of summaries

To satisfy their information needs or acquire knowledge, humans often have to read documents whose contents are complex and time-consuming to assimilate. The aim of a summary is to reduce the amount of effort required to acquire the knowledge contained in a document. A summary can be defined as a condensed, human-intelligible and non-critical representation of the content of another document:

• condensed representation: a summary gives a quick idea of the content of the original document;

• human-readable: a summary reduces the effort required by a human to acquire the knowledge contained in a document. The human is the direct user of a summary. In this way, a summary differs from an index or a representation used for knowledge extraction or reasoning;

• uncritical: a summary contains no comments or viewpoints on the original document.

We demonstrate the benefits of automatic document summarization, analyze the problem and present the main solutions currently in use. We focus on the main applications and concepts of automatic summary generation (ASG). After explaining the principle of GAR, we outline the different types of improvements that can be made. The problems and methods used for GAR can be applied to the summarization of several texts. Finally, we detail the issues and methods involved in summarizing documents in formats other than text: images, sound and video.