Some corpora built out over the years.
We're excited to share the first release of the Evidence Inference website, part of NSF CAREER project. The idea here is to infer the reported findings concerning a given treatment, comparator, and outcome of interest. More details on the corpus website, here: http://evidence-inference.ebm-nlp.com/
The EBM-NLP dataset comprises about 5,000 abstracts of articles describing clinical trials, the text is annotated to indicated descriptions of patients/populations, interventions, and outcomes. This corpus is described at length in our ACL 2018 paper, "A Corpus with Multi-Level Annotations of Patients, Interventions and Outcomes to Support Language Processing for Medical Literature". The corpus website is here: http://www.ccs.neu.edu/home/bennye/EBM-NLP/
This is a corpus of online reviews of physicians from RateMDs.com. We used this in our JAMIA 2014 paper. Here's the data http://www.byronwallace.com/static/ratemd-dr-ratings.csv.zip
This is a dataset built in 2014; it comprises reddit comments annotated as being sarcastic (or not). One version is available here: https://github.com/bwallace/ACL-2014-irony; there is also a version on Kaggle: https://www.kaggle.com/rohandoshi/sarcasm-classification.