
1.5.0
          Major size saving improvements in CFSA2. Built in Polish dictionary size decreased from
          2,811,345 to 1,806,661 (CFSA2 format).

          FSABuilder returns a ready-to-be-used FSA (ConstantArcSizeFSA). Construction overhead
          for this automaton is a round zero (it is immediately serialized in-memory).

          Polish dictionary updated to Morfologik 1.7. [19.11.2010]

          Added an option to serialize automaton to CFSA2 or FSA5 directly from fsa_build.

          CFSA is now deprecated for serialization (the code still reads CFSA automata, but will
          no be able to serialize them). Use CFSA2.

          Added immediate state interning. 
          Speedup in automaton construction by about 30%, memory use 
          decreased significantly (did not perform exact measurements, but incremental 
          construction from presorted data should consume way less memory).

          Added an option to build FSA from already sorted data (--sorted). Avoids in-memory sorting. 
          Pipe the input through shell sort if building FSA from large data.

          Changed the default ordering from Java signed-byte to C-like unsigned byte value.
          This lets one use GNU sort to sort the input using 'export LC_ALL=C; sort input'.  

          Added traversal routines to calculate perfect hashing based on FSA with NUMBERS.

          Changed the order of serialized arcs in the binary serializer for FSA5 to lexicographic 
          (consistent with the input). Depth-first traversal recreates the input, in other words.

          Removed character-based automata.

          Incompatible API changes to FSA builders (moved to morfologik.fsa).

          Incompatible API changes to FSATraversalHelper. Cleaned up match types, added
          unit tests. 

          ################################################################
          - Incompatible API changes have been made in release 1.5.0. See above.
          - An external dependency HPPC (high performance primitive collections) is now required
              for compiling FSAs (it is optional for traversals).
          ################################################################

1.4.1
          Upgrade of the built-in Morfologik dictionary for Polish (in CFSA format).

          Added options to define custom FILLER and ANNOT_SEPARATOR bytes in the fsa_build
          tool.

          Corrected an inconsistency with the C fsa package -- FILLER and ANNOT_SEPARATOR
          characters are now identical with the C version.
          
          Cleanups to the tools' launcher -- will complain about missing JARs, if any.

1.4.0
          Added FSA5 construction in Java (on byte sequences). Added preliminary support for
          character sequences. Added a command line tool for FSA5 
          construction from unsorted data (sorting is done in-memory).

          Added a tool to encode tab-delimited dictionaries to the format accepted by
          fsa_build and FSA5 construction tool.

          Added a new version of Morfologik dictionary for Polish (in CFSA format).

1.3.0      
          Added runtime checking for tools availability so that unavailable
          tools don't show up in the list.

          Recompressed the built-in Polish dictionary to CFSA. 

          Cleaned up FSA/Dictionary separation. FSAs don't store encoding any more (because
          it does not make sense for them to do so). The FSA is a purely abstract class
          pushing functionality to sub-classes. Input stream reading cleaned up.

          Added initial code for CFSA (compressed FSA). Reduces automata size about 10%. 

          Changes in the public API. Implementation classes renamed 
          (FSAVer5Impl into FSA5). Major tweaks and tunes to the API.

          Added support for version 5 automata built with NUMBERS flag (an extra field
          stored for each node).

          ################################################################
          Incompatible API changes have been made in release 1.3.0.
          ################################################################

1.2.2     License switch to plain BSD (removed the patent clause which did not make much sense 
          anyway).

          The build ZIP now includes licenses for individual JARs (prevents confusion). 

1.2.1     Fixed tool launching routines.

1.2       Package hierarchy reorganized.

          Removed stempel (heuristic stemmer for Polish).

          Code updated to Java 1.5. 

          The API has changed in many places (enums instead of constants, 
          generics, iterables, removed explicit Arc and Node classes and replaced
          by int pointers).

          FSA traversal in version 1.2 is implemented on top of primitive
          data structures (int pointers) to keep memory usage minimal. The speed
          boost gained from this is enormous and justifies less readable code. We
          strongly advise to use the provided iterators and helper functions
          for matching state sequences in the FSA.

          Tools updated. Dumping existing FSAs is much, much faster now.        

          ################################################################
          Incompatible API changes have been made in release 1.2.
          Java 1.5 or later is required from this version on.
          ################################################################

1.1.4     * Fixed a bug that caused UTF-8 dictionaries to be garbled. Now it should be relatively
          safe to use UTF-8 dictionaries (note: separators cannot be multibyte UTF-8 characters,
          yet this is probably a very rare case).

1.1.3     * Fixed a bug causing NPE when the library is called with null context class loader 
          (happens when JVM is invoked from an JNI-attached thread). Thanks to
          Patrick Luby for report and detailed analysis.

          Updated the built-in dictionary to the newest version available. 

1.1.2     * Fixed a bug causing JAR file locking (by implementing a workaround).

          Fixed the build script (manifest file was broken).

1.1.1     Distribution script fixes. The final JAR does not contain test classes and resources. Size
          trimmed almost twice compared to release 1.1.

          Updated the dump tool to accept dictionary metadata files.

1.1       Introduced an auxiliary "meta" information files about compressed dictionaries.
          Such information include delimiter symbol, encoding and infix/prefix/postfix
          decoding info.
          
          The API has changed (repackaging). Some deprecated methods have been removed.
          This is a major redesign/ upgrade, you will have to adjust your source code.
          
          Cleaned up APIs and interfaces.

          Added infrastructure for command-line tool launching.

          Cleaned up tests.

          Changed project name to morfologik-stemmers and ownership to (c) Morfologik.

1.0.7     Removed one bug in fsa 'compression' decoding.

1.0.6     Customized version of stempel replaced with a standard distribution.

          Removed deprecated methods and classes.
          
          Added infix and prefix encoding support for fsa dictionaries.

1.0.5     Added filler and separator char dumps to FSADump.
          
          * A major bug in automaton traversal corrected. Upgrade when possible.
          
          Certain API changes were introduced; older methods are now deprecated
          and will be removed in the future.

1.0.4     Licenses for full and no-dict versions.

1.0.3     Project code moved to SourceForge (subproject of Morfologik).
          LICENSE CHANGED FROM PUBLIC DOMAIN TO BSD (doesn't change much, but clarifies
          legal issues).

1.0.2     Added a Lametyzator constructor which allows custom dictionary stream, field
          delimiters and encoding. Added an option for building stand-alone 
          JAR that does not include the default polish dictionary.

1.0.1     Code cleanups. Added a method that returns the third automaton's column (form).

1.0       Initial release