Abstractive Unsupervised Multi-Document Summarization using Paraphrastic Sentence Fusion

Proceedings of the 27th International Conference on Computational Linguistics, pages 1191–1204 Santa Fe, New Mexico, USA, August 20-26, 2018. 1191 Abstractive Unsupervised Multi-Document Summarization using Paraphrastic Sentence Fusion Mir Tafseer Nayeem University of Lethbridge Lethbridge, AB, Canada mir.nayeem@uleth.ca Tanvir Ahmed Fuad University of Lethbridge Lethbridge, AB, Canada t.fuad@uleth.ca Yllias Chali University of Lethbridge Lethbridge, AB, Canada chali@cs.uleth.ca Abstract In this work, we aim at developing an unsupervised abstractive summarization system in the multi-document setting. We design a paraphrastic sentence fusion model which jointly performs sentence fusion and paraphrasing using skip-gram word embedding model at the sentence level. Our model improves the information coverage and at the same time abstractiveness of the gen- erated sentences. We conduct our experiments on the human-generated multi-sentence com- pression datasets and evaluate our system on several newly proposed Machine Translation (MT) evaluation metrics. Furthermore, we apply our sentence level model to implement an abstrac- tive multi-document summarization system where documents usually contain a related set of sentences. We also propose an optimal solution for the classical summary length limit problem which was not addressed in the past research. For the document level summary, we conduct experiments on the datasets of two different domains (e.g., news article and user reviews) which are well suited for multi-document abstractive summarization. Our experiments demonstrate that the methods bring significant improvements over the state-of-the-art methods. 1 Introduction The task of automatic document summarization aims at finding the most relevant informations in a text and presenting them in a condensed form. A good summary should retain the most important con- tents of the original document or a cluster of related documents, while being coherent, non-redundant and grammatically readable. There are two types of summarizations: abstractive summarization and extractive summarization. Abstractive methods need extensive natural language generation to rewrite the sentences (Chali et al., 2017). Therefore, research community is focusing more on extractive sum- maries, which selects salient (important) sentences from the source document without any modification to create a summary. The abstractive techniques which are traditionally used are sentence compression, syntactic reorganization and lexical paraphrasing. Summarization is classified as single-document or multi-document based upon the number of source document. The information overlap between the doc- uments from the same topic makes the multi-document summarization more challenging than the task of summarizing single documents. However, in case of multi-document summarization where source doc- uments usually contain similar information, the extractive methods would produce redundant summary or biased towards specific source document (Nayeem and Chali, 2017a). Multi-sentence compression (MSC) can be a useful solution for the above problem. It usually takes a group of related sentences and produces an output sentence through merging the sentences about the same topic, retaining the most important information and still maintain the grammaticality of the generated sentence. MSC is a text-to-text generation process in which a novel sentence is produced as a result of summarizing a set of similar sentences originally called sentence fusion (Barzilay and McKeown, 2005). On the other hand, lexical paraphrasing aims at replacing …

Content truncated to 500 words