Generating Coherent Summaries of Scientific Articles Using Coherence Patterns

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 772–783, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics Generating Coherent Summaries of Scientific Articles Using Coherence Patterns Daraksha Parveen Mohsen Mesgar NLP Group and Research Training Group AIPHES Heidelberg Institute for Theoretical Studies gGmbH Heidelberg, Germany {daraksha.parveen|mohsen.mesgar|michael.strube}@h-its.org Michael Strube Abstract Previous work on automatic summarization does not thoroughly consider coherence while generating the summary. We introduce a graph-based approach to summarize scientific articles. We employ coherence patterns to en- sure that the generated summaries are coher- ent. The novelty of our model is twofold: we mine coherence patterns in a corpus of ab- stracts, and we propose a method to combine coherence, importance and non-redundancy to generate the summary. We optimize these fac- tors simultaneously using Mixed Integer Pro- gramming. Our approach significantly outper- forms baseline and state-of-the-art systems in terms of coherence (summary coherence as- sessment) and relevance (ROUGE scores). 1 Introduction The growth in the scientific output of many differ- ent fields makes the task of automatic summariza- tion imperative. Automatic summarizers assist re- searchers to have an informative and coherent gist of long scientific articles. An automatic summarizer produces summaries considering three properties: Importance: The summary should contain the im- portant information of the input document. Non-redundancy: The summary should contain non-redundant information. The information should be diverse in the summary. Coherence: Though the summary should comprise diverse and important information of the input doc- ument, its sentences should be connected to one an- other such that it becomes coherent and easy to read. If we do not ensure that a summary is coher- ent, its sentences may not be properly connected. This results in an obscure summary. In previous work coherence has not been thoroughly considered. Parveen and Strube (2015) use single sentence con- nectivity in the input document as a coherence mea- sure. They measure coherence by calculating the outdegree of a sentence in a graph representation of an input document. This has two disadvantages: first, since it is computed only based on one sen- tence, it is not sufficient to generate coherent sum- maries; second, it is obtained based on sentence con- nectivity in the input document rather than in the summary. In this work, we focus on the coherence aspect of summarization. We use discourse entities as the unit of information that relate sentences. Here, discourse entities are referred to as head nouns of noun phrases (see Section 2). The main goal is to extract sentences which refer to those entities which are important and unique, and also to entities which connect the ex- tracted sentences in a coherent manner. Entities in connected sentences can be used to create linguis- tically motivated coherence patterns (Daneˇs, 1974). Recently, Mesgar and Strube (2015) modeled these coherence patterns by subgraphs of the graph repre- sentation (nodes represent sentences and edges rep- resent entity connections among sentences) of doc- uments. They show that the frequency of coherence patterns can be used …

Content truncated to 500 words