Abstract
Summarization of legal case judgement documents is a practical and challenging problem, for which many summarization algorithms of different varieties have been tried. In this work, rather than developing yet another summarization algorithm, we investigate if intelligently ensembling (combining) the outputs of multiple (base) summarization algorithms can lead to better summaries of legal case judgements than any of the base algorithms. Using two datasets of case judgement documents from the Indian Supreme Court, one with extractive gold standard summaries and the other with abstractive gold standard summaries, we apply various ensembling techniques on summaries generated by a wide variety of summarization algorithms. The ensembling methods applied range from simple voting-based methods to ranking-based and graph-based ensembling methods. We show that many of our ensembling methods yield summaries that are better than the summaries produced by any of the individual base algorithms, in terms of ROUGE and METEOR scores.