摘要：AbstractThe 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 (Mani, 2001). 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 CSTNewscorpus. 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.
