OpenGrm NGram - Release 1.3 (compatible with OpenFst 1.6.6)
  * Templated Arc to provide new hist-arc counting, supporting fractional count
    based Katz backoff models. (1.3.0)
  * Updated error handling. (1.3.0)
  * Updated code to better conform to Google C++ guidelines. (1.3.0)
  * Added Bayesian model merging. (1.3.0)
  * Some re-factoring of functions to improve efficiency, particularly those
    for distributed training, such as ngram-complete and ngram-split. (1.3.0)
  * Updated unit tests to run more quickly and to be less bash-specific. (1.3.0)
  * Updated unit test scripts to use random file names and directories, to avoid
    collisions with multiple processes. (1.3.1)
  * Fixed ill-formatting in error handling. (1.3.1)
  * Improved reading of large files of ngram counts with ngramread. (1.3.1)
  * Compatibility with OpenFst version 1.6.0 (1.3.2)
  * Substantial rewrite of the ngram-marginalize methods. The algorithm now
    has improved efficiency and stability, and no longer needs the --iteration
    flag. (1.3.3)
  * Minor floating point stability and error handling updates. (1.3.3)
  * Update to ngram-shrink methods to allow specification of a minimum order
    to prune. (1.3.3)
  * Some changes to container types, following usage recommendations. (1.3.3)
  * Avoid copying in the context library (1.3.4)

OpenGrm NGram - Release 1.2
  * Port to C++11 (1.2.0)
  * Port to C++11 addenda (1.2.1)
  * Compatibility with OpenFst version 1.5.1 (1.2.2)

OpenGrm NGram - Release 1.1
  * Substantially refactored code to allow for easy distributed model building.
  * Included a 'convenience' shell script bin/ngram.sh that runs some common
    pipelines of commands, e.g., from text file to pruned language model.
    Shell script provides illustration of distributed model building utilities.
  * Functions to re-estimate smoothed n-gram models to impose marginal
    distribution constraints.

OpenGrm NGram - Release 1.0
  * Initial public release
