* TBLTrainer.bestRule() choose rule with lower bad counts when two rules has the same score.
* New package structure.
* A hack was programed in Trainer. Now reinitialize the process when there are too many rules. Memory and execution time is better now.
* Now, Trainer only erase the rules that are new and its good count is zero.
* Files are saved with the given encoding.
* ASName parameter in DocumentConverter resource added.
* In Lexicon LR, key and classification initial parameters were added. Now it is possible to overwrite the mapping rule key => classification loaded from lexicon file (first line)
* In Initial Tagger, lowerCase and uppInitial initial parameters were added. These are used to annotate unseen attributes (not in lexicon)
* Trainer and CreateLexicon now can be executed at the end of a corpus pipeline. This helps preprocessing corpus with other gate pr's in a faster way. In order to add this functionality, the classes DocumentAnnotationManager and CorpusAnnotationManager were added (they both implement the new interface AnnotationManager).
* Fixed bug in Lexicon.java there was a problem with the complex matcher regexp
* Now DocumentConverter can read and convert all files in a directory (no recursion supported) so that you can load a whole corpus