Overview of AMALGUM - Large Silver Quality Annotations across English Genres

Corpus resources for Linguistics and NLP research on discourse phenomena, such as coreference and discourse trees, are limited by a lack of large scale, well-understood, annotated datasets: corpora are either very large (100M-10G tokens) but shallowly annotated and with unknown composition, or richly annotated, but smaller. Here, we present a resource that takes a middle path, combining some of the best features of scraped corpora - size, open licenses, lexical diversity - and high quality curated data for more interpretable inferences with complex annotations.