One problem in CS is that a lot of interesting datasets are encumbered. When work is done in collaboration with a company (which is common), the company may oft

One problem in CS is that a lot of interesting datasets are encumbered. When work is done in collaboration with a company (which is common), the company may often not agree to publish the dataset, and retains a veto right on what gets made public. Among many areas, this is getting quite normal in natural-language processing and machine translation, where a lot of the good datasets are owned by companies. A lot of research lately has come either from, or in collaboration with, Google in particular, because they have a treasure-trove of data that powers Google Translate. They are not likely to agree to release that, because it's one of the keys to their competitive advantage in that area. Even for in-house data, there are increasingly explicit demands from universities that scientists work with the university technology-transfer office to commercialize their work. Everyone wants to be the next Stanford, with spinoff startups and a stream of licensing revenue. So even if you collected a nice dataset in-house, there are financial pressures that rather than just giving it away, you should instead talk to your local venture capitalist about your unique, hard-to-replicate dataset with promising commercial value...