RELEASE_NOTES

NILE - RELEASE NOTES
====================
v1.2  Bugfixes and usability improvements thanks to user feedback.
      (Code revision 33)
      (1) Fix a bug related to ignoring failed parse trees in training data.
      (2) New mechanism for specifying language-specific or user-supplied featuresets.
          See README section IV.
      (3) Updated README for clarity.
      (4) Modified the example train.sh training script to include flags for development data.
      (5) Eliminated unused code and commandline flags. General clean-up and style improvements.

v1.14 Bugfix release. (Code revision 28)
      (1) Fixes bug related to the processing of empty failed parse
          trees in training data.
      (2) Minor code cleanup and style fixes.

v1.12 Minor changes to the tree reader.
      (Code revision 18)
      (1) Minor edits to NLPTreeHelper.stringToTree_weakRef to accept standard Berkeley trees
          with no TOP symbols.
      (2) Tree reader will now also ignore blank lines in the tree input file in addition to
          lines with a 0 token on a line by itself. Useful in case you have sentences that
          failed to parse.

v1.11 Bugfix for importing features into model for new/unknown language pair.
      (Code revision 14)

v1.1 Low-level, but important bugfix release mostly relating to portability:
     (Code revision 12)
     (1) NLPTreeHelper.py:stringToTree now reads standard Penn Treebank style trees
         instead of only ISI-format trees. README adds description of tree format.
     (2) NLPTree.py cleaned up.
     (3) Fix stray newline in nile.py
     (4) Acquire tempdir location from tempfile.gettempdir() instead of using hardcoded path.
     (5) Added option to specify user-defined directory location for temporary files.
         Use flag --tempdir
     (6) Cleanup of train.sh and test.sh scripts.

v1.0 Initial public release -- 3/20/2012
     (Code revision 1)


ACKNOWLEDGEMENTS
================
Thanks very much to:
 David Chiang
 Philipp Koehn
 Sasa Hasan
 Zhongqiang Huang
 Ashish Vaswani
for helping to test this software and providing valuable feedback.