Recluse

..........................
* README has some dependency and installation information.

..........................
* utils now has a function to calculate precision, recall, and f-measure.

..........................
* utils.partition_by_list now returns a list, not a tuple.

..........................
* Fixed a unicode issue: the commandline version of the tokeniser must decode before sending to stdout.
* Updated a regression test of vocabulary generator that had slightly different output based on changes to the tokenisation.

..........................
* Changed the interface to nltk_based_segmenter_tokeniser.
** It still can return a generator of sentences, and prints out tokenised segmented sentences when used on the commandline.
** The generator now yields sentences in strings, either the original or the tokenised version
** Methods for doing the tokenisation are provided separately.
** This puts the burden of aligning tokenised text to original text in the client, while still allowing the client access to the information needed to do so.
* There is a change in dealing with contractions; we now follow the convention of separating a contraction into two parts with the apostrophe attached to the second part.

..........................
* data needed for testing should now be included.