Recent advances in natural language processing have enabled automation of a\nwide range of tasks, including machine translation, named entity recognition,\nand sentiment analysis. Automated summarization of documents, or groups of\ndocuments, however, has remained elusive, with many efforts limited to\nextraction of keywords, key phrases, or key sentences. Accurate abstractive\nsummarization has yet to be achieved due to the inherent difficulty of the\nproblem, and limited availability of training data. In this paper, we propose a\ntopic-centric unsupervised multi-document summarization framework to generate\nextractive and abstractive summaries for groups of scientific articles across\n20 Fields of Study (FoS) in Microsoft Academic Graph (MAG) and news articles\nfrom DUC-2004 Task 2. The proposed algorithm generates an abstractive summary\nby developing salient language unit selection and text generation techniques.\nOur approach matches the state-of-the-art when evaluated on automated\nextractive evaluation metrics and performs better for abstractive summarization\non five human evaluation metrics (entailment, coherence, conciseness,\nreadability, and grammar). We achieve a kappa score of 0.68 between two\nco-author linguists who evaluated our results. We plan to publicly share\nMAG-20, a human-validated gold standard dataset of topic-clustered research\narticles and their summaries to promote research in abstractive summarization.\n