Please visit http://users.info.unicaen.fr/~alardill/anymalign/ for details.


================================== Change log =================================

2.5 (May 4th 2011)
- bug #5 corrected: program could run indefinitely when "-t" option value was
set too high.
- new option "-i" to improve the coverage of long n-grams, by generalizing
the indexation phase to n-grams rather than simple words. Default value is 1,
which corresponds to the original behavior. The larger, the more n-grams output,
but the slower the program and the more memory required at startup. To be
improved...
- replaced "-v" option (verbose) with "-q" (quiet). Default behavior is now
verbose. Also, now displays the average size of subcorpora processed in addition
to the number of subcorpora (info is now persistent).
- changed default value for "-N" option from 0 (no limit) to 7.
- set a limit on the size of subcorpora processed (100,000 sentences, hard
coded). This prevents the program from spending too much time on huge (though
rare) subcorpora, giving an impression of hanging.

2.4 (September 29th 2010)
- bug #4 corrected: hash collisions could make the program crash randomly when
computing translation probabilities.
- corrected inconsistency in long option names: "--min-gram" -> "--min-ngram".

2.3 (July 20th 2009)
- bug #3 corrected: under some conditions, computing lexical weights could make
the program crash.
- changed the "-n" and "-N" option behavior: now filter out alignments that
contain ANY n-gram that does not verify the specified length constraints
(formerly ONLY).

2.2 (June 9th 2009)
first public release


===============================================================================

List of nice people who helped in some way (hints, requests, bugs), starting
from former malign.py:

Yves Lepage
Julien Gosme
Francis Bond
Alex Yanishevsky
Sue Chen
Jonathan Chevelu
Aurélien Max
François Yvon
Kota Takeya