Manual Typification of Source Texts and Multi-document Summaries Alignments

详细信息    查看全文

• 作者：Renata T. Camargo ; Ver么nica Agostini ; Ariani Di Felippo ; Thiago A.S. Pardo
• 关键词：typification ; alignment ; summarization
• 刊名：Procedia - Social and Behavioral Sciences
• 出版年：25 October, 2013
• 年：2013
• 卷：95
• 期：Complete
• 页码：498-506
• 全文大小：338 K

文摘

The Multi-document Summarization (MDS) has been focused in Natural Language Processing (NLP) and its aim is to produce automatic summaries from a collection of texts that deal with the same subject (). The alignment of human-written abstracts to their source documents makes explicit the correspondences that exist in such documents/abstract pairs and create a potentially rich data source to create of rules and models to support more linguistically motivated MDS methods. In this paper we describe the typification of such alignments in the CSTNews corpus. This work is part of two larger projects called Sucinto and Sustento, and it supports MSD researches of Brazilian Portuguese language. Specifically, the typification process consisted of assigning labels to the alignment between a summary sentence and its corresponding source sentence which codify formal and content aspects of the alignment. In order to present this work, we outline the alignment, and detail the typification process, the results of our work and some conclusions.