BioNLP-Corpora is a repository of biologically and linguistically annotated corpora and biological datasets. It is one of the projects of the BioNLP initiative by the Center for Computational Pharmacology at the University of Colorado Denver Health Sciences Center to create and distribute code, software, and data for applying natural language processing techniques to biomedical texts. There are many resources available for download at BioNLP-Corpora:

CRAFT: Colorado Richly Annotated Full-Text Corpus
Protein Residue Corpora: several corpora relevant to extraction of protein residues in text
PICorpus: Protein Interaction Corpus, a corpus of annotated protein-protein interactions
GeneHomonym: Gene identifier values that mean different things or refer to multiple genes by inference
The Annotation Projects: human annotated text on biological subject matter
The Medline Mining Projects: automatically mined data from Medline
The Anaphora Corpus: sample of GeneRIFs annotated with pronominal anaphora and their antecedents
Test Suite Corpora: structured test suites for natural language processing applications