Universal segmentation of Text with SUMO formalism

Abstract

We propose a universal formalism for the segmentation of text documents called Sumo. Its main purpose is to help creating segmentation systems for documents in any language. Because the processing is independent of the language, any level of segmentation (be it character, word, sentence, paragraph, etc.) can be considered. We will argue about the usefulness of such a formalism, describe the framework for segmentation on which Sumo relies, and give detailed examples to demonstrate some of its features.