(py25, py26, py27)
* call this in __init__.py (deciding whether to import align, which requires 2.6+)
* call this in corpus/reader/util.py (deciding whether to use followlinks flag)

Fixes Issue #294 

Added a regression test for SimpleGoodTuringProbDist to probability.doctest. Fixed another doctest. All tests in probability.doctest now pass

Fix to #229

(meaning that it will now fail 4 tests in Python 2.5 and 2.6)

Make MaltParser safe to use in parallel

Plug DependencyGraph file descriptor leak

Forgot a ".mco" in "${working_dir}/${mco}.mco".

As far as I could see, MaltParser does not take a full path as input for
mco models. You must specify the name of the model, which MaltParser
then uses to look for a file in ${working_dir}/${mco}.mco. It's a bit
clunky, but I couldn't find alternatives in their documentation.

Call gettempdir() at initialisation time instead of module load time.

This is not a major leak, since the garbage collector will destroy the
file object relatively soon, but it might still be problematic if
called in a tight loop.

Just moved the code inside the try block.

- Actually use self.working_dir
- Replace hard-coded temporary file names with
  tempfile.NamedTemporaryFile
- Use subprocess instead of os.system to avoid quoting issues when
  calling java
- Error out if subprocess fails

Bugfix for agreement.py removing dependence on ordered coder names.

Fix issue 285

Fixed bug in agr(self, cA, cB, i, data=None) where a wrong order of coder/item combinations in data could cause a result of 0.0, which affected results from kappa(), pi(), and S().