Version 0.8:
- Sparse matrix representation enabled, so that fadd is in sync with fsa.
- Corrected a bug that showed only when NEXTBIT was not on (thanks to
  Gertjan van Noord)
Version 0.9:
- No Idea
Version 0.10:
- Corrected a bug resulting from an incorrect behaviour of the new I/O
  library.
Version 0.11:
- The library now compiles under gcc 4.0.1 (the compiler people are quite
  creative in creating new syntax errors).
- Sparse matrix representation did not work, and it looks like it has
  never worked here (I forgot to finish it), or I messed some
  versions. Now it should be corrected.
- LGPL licence.
Version 0.12:
- Set the compile options to those of the current fsa package (thanks
  to Martin Forst and Gertjan van Noord).
- Corrected compile error when -DNEXTBIT was absent (thanks to Martin
  Forst and Gertjan van Noord).
Version 0.13
- Corrected a buffer overflow error in morphological analysis detected
  by Daniel de Kok.
Version 0.14
- Corrected a memory leak in pro_init_tuple, pro_word_tuple_grams, and
  pro_word_tuple_fpgrams in pro_fadd.c by applying a patch supplied by
  Daniel de Kok.
Version 0.15
- Removed 'register' keyword from fsa.h as it is now forbidden by a
  new 'standard'
- Renamed tuple class to ftuple as developers of iostream or rather of
  exceptions used there chose to use the name for their own purpose
  even if 'tuple' has nothing to do with exceptions.
- Introduced a new target 'c_plus_plus' in the Makefile to compile
  only the C++ library (so users don't have to search for their
  target).
- Added a Python wrapper to be used with pybind11
