MassiveSumm: a very large-scale, very multilingual, news summarisation dataset

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10150–10161 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 10150 MassiveSumm: a very large-scale, very multilingual, newswire summarisation dataset Daniel Varab and Natalie Schluter IT University of Copenhagen Denmark {djam,natschluter}@itu.dk Abstract Current research in automatic summarisation is unapologetically anglo-centered–a persis- tent state-of-affairs, which also predates neu- ral net approaches. High-quality automatic summarisation datasets are notoriously expen- sive to create, posing a challenge for any lan- guage. However, with digitalisation, archiving, and social media advertising of newswire arti- cles, recent work has shown how, with careful methodology application, large-scale datasets can now be simply gathered instead of written. In this paper, we present a large-scale multi- lingual summarisation dataset containing arti- cles in 92 languages, spread across 28.8 mil- lion articles, in more than 35 writing scripts. This is both the largest, most inclusive, exist- ing automatic summarisation dataset, as well as one of the largest, most inclusive, ever pub- lished datasets for any NLP task. We present the first investigation on the efficacy of re- source building from news platforms in the low-resource language setting. Finally, we pro- vide some first insight on how low-resource language settings impact state-of-the-art auto- matic summarisation system performance. 1 Introduction Automatic summarisation datasets are generally ex- pensive to create, because they generally involve a human reading a document several times and then crafting a fluent piece of text that captures both the important information of the document and the intention of the resulting summary. Each datapoint in such a dataset could take hours to man- ually create. With digitalisation, archiving, and social media advertising of newswire articles, re- cent work has shown how, with dedicated time and methodology application, large-scale datasets can now be simply gathered instead of written (Grusky et al., 2018; Hermann et al., 2015). But the method development was carried out over English, and un- til the research presented here, the method has only been applied to a very limited number of relatively richly-resourced languages (Varab and Schluter, 2020; Scialom et al., 2020). We have extended the methodology further (Sec- tion 3) and applied it carefully and widely to gener- ate MassiveSumm: a very large-scale, very multi- lingual summarisation dataset of 28.8 million arti- cles, containing data in 92 languages, using more than 35 writing scripts. This is by far both the largest, most inclusive, existing automatic sum- marisation dataset, as well as one of the largest, most inclusive, ever published datasets for any NLP task. The bulk of this paper outlines the size, di- versity and inclusivity of the dataset as an auto- matic summarisation dataset, as well as simply raw text data in comparison with two other multilingual large-scale widely used datasets in NLP: Wikipedia and Common Crawl (Section 4). In light of extending and applying the data ac- quisition method under the low-resource setting, we identify some unreasonable conditions for lan- guage inclusion in automatic summarisation re- search, which stand to perpetuate a lack of …

Content truncated to 500 words