More documentation over reference outputs

[jnothman]

Given the number of variant tasks, we need to document the task setting for each system, in a structured form. Ideally, we would like to know:

• does the system take gold mentions as input?
• if so, do they include gold nils?
• does the system output intend to reflect exactly CoNLL 2003 entities (PER, LOC, ORG, MISC) or does it perform additional wikification? (perhaps: does it annotate only a subset of PER, LOC, ORG?)
• what objective metric / corpus was used to develop/optimise the system?

We should additionally encourage README files including bibtex references.

Finally, I'm not sure that .analysis files should be stored in the repo.