A Question/answer Explanation Based Approach for Customizable Text Expansion Vs Summarization

In this paper, we propose a text representation and a planning technique, on which can be based the generation of diierent versions of a same text. We focused our study on narration of events and states. This kind of text can be related in its minimal version, that we call kernel version, or by a hierarchy of versions containing more and more details, until the longest version that includes all the available details. We see scalable summarization as a process of text version generation. We deene an operation, the text expansion, that aims, from a kernel version of a text, at selecting and organizing the details to be introduced in order to obtain a particular version. These selection and organizing are realized in such a way that the resulting version contains enough information that answers to questions (how, when, where and why) that can be addressed to a text version (and more precisely, questions that are supported by utterances of a text version). This process requires: (1) a characterization of the kernel version, (2) a textual base that contains all the available information that we elaborate thanks to a decomposition of the text to summarize into elementary sentences, according to Harris's grammar, to which is associated syntactic knowledge, and (3) a question/answer structure that reeects a category of semantic knowledge linking these elementary sentences into the whole text. 1 Presentation In this paper, we propose a text representation and a planning technique, on which can be based the generation of diierent versions of a same text. We focused our study on narration of events and states. This kind of text can be related in its minimal version, that we call a kernel version, or by a hierarchy of versions containing more and more details, until the longest version that includes all the available details. We see scalable summa-rization as a process of text version generation. We deene an operation, the text expansion, that aims, from a kernel version of a text, at selecting and organizing the details to be introduced 0 in order to obtain a particular version. These selection and organizing are realized in such a way that the resulting version contains enough information answering to questions (how, when, where and why) that can be addressed to a text version (and more precisely, questions that are supported by utterances of a text version). The explanation is considered in the …