The arXiv as Dataset

The arXiv is a repository of over 1 million preprints in physics, mathematics and computer science. It is truly open access, and the preprints are an excellent dataset for testing out all sorts of language modelling / machine learning prototypes.