A Generative Approach for Multi-Document Summarization using the Noisy Channel Model

Multi-document summarization is the automatic production of a unique summary from a collection of texts. This task has become very important, since it assists the information processing in days where the amount of information is growing considerably. In this paper, we propose a statistical generative approach for multi-document summarization. In particular, we formulate the multi-document summarization task using a Noisy-Channel model. This approach is novel for multi-document summarization and it explores the process of summarization through the analysis of factors, such as redundancy, complementarity and contradiction. In this work, we model these factors using the Cross-document Structure Theory.