Implement corpus-based MFS baseline

[logological]

Originally reported on Google Code with ID 40

We should implement a corpus-based MFS baseline for the case where the WSD data set
includes training data.  (Note that this is separate from sense frequency information
which may be available from the sense inventory.)

Reported by [email protected] on 2013-10-15 14:35:11

[logological]

Such a baseline can also be used directly on the test data where no training data is
available.  In this case, it would be run *after* a knowledge-based annotator, and
would function as a one-sense-per-discourse annotator.

Reported by [email protected] on 2013-10-15 14:39:37