TATOO

- LAST CHANGES -

Version 3.00 - January 1998



Version 3.00

• The Matrix implementation has been completly reshaped to simplify the commands, to use less memory (sparse matrix) and allow to define more states and classes. That's improve the reusability of Matrices, since all declarations are included in one file, and if new states or classes appear in a new text, they will be automatically added.
• An history file is associated to each Matrices file.
• The preparation for the training and the tagging gives the same output (the -P option of mpreptxt is unavailable). This simplify the understanding and the comprehension of the sequence of the tools.
• A Web Page with documentation is now available.
  http://www.issco.unige.ch/staff/robert/tatoo/tatoo.html
• Known bugs fixed.

Version 2.22 (July 96)

• It's now possible to do shell-style pattern matching for ?, \, [], and * characters in the MSD conversion table.

Version 2.2

• The tagger can accept undefined states (and consequently classes). The mpreptxt program, with the option for tagging (-P), training (-T) and hand tagging (-H), provide the possibility for the concerned programs to adapt their results (append new states and classes, increase the matrices size and provide results).
• Possibility to print the results in the original tag set format.
• An user interface in tcl/tk/expectk5-19.

Version 2.1

• mpretag adapted to the last (October 1995) MULTEXT segmenter format.
• program 'mhandtag' general (format "independant")
• New biases possibilities.
• mdiffb can estimate the error rate when using the precision option in mtag.

Version 2.0

• Tag Conversion for specific words (in mpreptxt: -w -n options)
• No conversion applied if -c or -w options are not specified.
• The input matrices are checked to be in compiled form.
• Check the Tagset in function of both conversion tables.
• A new program 'mbiases' can take a list of biases and compute a new set of value for the matrices A[][] and PI[].
• Create matrices with an already tagged text.
• mtag can reestimate the matrices with an short hand tagged tag list, and this improve the performances.
• New Program to desambiguite by hand text (mhandtag)
• Program to convert new output segmenter in TAGGER format (mpretag).
• The possibility to prepare a new training text using an already defined formatted classes file, has been add (-T option in mpreptxt). And in this case, you can use it for the new training. If one or more tags do not appear in this text (which could change the matrices in the wrong way), and if you have specified initial compiled matrices, the values of this (those) tag(s) will be kept unchanged.
• Useful tools for performance checking (mdiff/mdiffb/mcontext).
• Makefiles modified. (Gnu make can be use)

Version 1.3

• In the MSD conversion table: Possibility to have more than one tag per conversion
• Notion of the precision in the 'mtag' program.
• The text preparation phase has been improved.

Version 1.2

• Reported Bugs are fixed
• The program 'mpreptxt' can be used as a filter for the tagger program 'mtag'.
• Portability between differents platforms has been checked. (SunOs 4.1.x / SunOs 5.3 )

Thanks

Nerea Ezeiza Ramos (si.ehu.es)
Holger Trebbe (muenster.de)
Petra Steiner (muenster.de)
Jules van Weerden (ruu.nl)
Andre Kempe (xerox.fr)
Anne Schiller (xerox.fr)
Pierrette Bouillon (issco.unige.ch)
Susan Armstrong (issco.unige.ch)
David McKelvie (edinburgh.ac)
monica monachini (CNUCE.CNR.it)