reverted changes to porter stemmer, because they inadvertently introduced different behaviour of the stemmer

Correction on the Pk segmentation metric

The first implementation I did was 100% wrong.
According to [1] Pk should provide a penalty of about 0.5 to all degenerated segmentations (empty, full, randomized), which is not statisfied by my last implementation.
It was corrected using the explantations provided in [2]

Sorry about that

[1] Beeferman D., Berger A., Lafferty J. (1999)
Statistical Models for Text Segmentation
Machine Learning, 34, 177-210

[2] Unbiased discourse segmentation evaluation
John Niekrasz and Johanna D. Moore
Spoken Language Technology Workshop (SLT), 2010 IEEE
Page(s): 43 - 48

incorporated David Doukhan's pull request https://github.com/nltk/nltk/pull/345 , specified float return type, added doctests, misc formatting fixes

fixes for tkinter in nltk.app and nltk.draw packages

Escape reserved LaTeX characters in nltk.tree.Tree.pprint_latex_qtree()

Fix for decision trees with binary=True

fixed doctype regex to handle doctype tag w/o square brackets

Fix the decision stumps so that they work on Python 3, and are robust to not seeing a positive or negative examples.

Correction of bugs in the proposed version

almost fixed it; want to make the decisiontrees demo work right with the binary attribute set to true

Two fixes for nltk.downloader

Some stability issues in resolution.doctest are fixed by making AbstractVariableExpression orderable and BindingDict.__str__ stable.