.Dd August 30, 2006
.Dt APERTIUM-TAGGER 1
.Os Apertium
.Sh NAME
.Nm apertium-tagger
.Nd part-of-speech tagger and trainer for Apertium
.Sh SYNOPSIS
.Nm apertium-tagger
.Op Ar options
.Fl g Ar serialized_tagger Op Ar input Op Ar output
.Nm apertium-tagger
.Op Ar options
.Fl r Ar iterations corpus serialized_tagger
.Nm apertium-tagger
.Op Ar options
.Fl s Ar iterations dictionary corpus tagger_spec serialized_tagger
.Ar tagged_corpus untagged_corpus
.Nm apertium-tagger
.Op Ar options
.Fl s Cm 0
.Ar dictionary tagger_spec serialized_tagger tagged_corpus untagged_corpus
.Nm apertium-tagger
.Op Ar options
.Fl s Cm 0 Fl u Ar model serialized_tagger tagged_corpus
.Nm apertium-tagger
.Op Ar options
.Fl t Ar iterations dictionary corpus tagger_spec serialized_tagger
.Sh DESCRIPTION
.Nm apertium-tagger
is the application responsible for the apertium part-of-speech tagger
training or tagging, depending on the calling options.
This command only reads from the standard input if the option
.Fl Fl tagger
or
.Fl g
is used.
.Sh OPTIONS
.Bl -tag -width Ds
.It Fl t Ar n , Fl Fl train Ar n
Initializes parameters through Kupiec's method (unsupervised),
then performs
.Ar n
iterations of the Baum-Welch training algorithm (unsupervised).
.It Fl s Ar n , Fl Fl supervised Ar n
Initializes parameters against a hand-tagged text (supervised) through
the maximum likelihood estimate method, then performs
.Ar n
iterations of the Baum-Welch training algorithm (unsupervised).
The CRP argument can be omitted only when
.Ar n
= 0.
.It Fl r Ar n , Fl Fl retrain n
Retrains the model with
.Ar n
additional Baum-Welch iterations (unsupervised).
.It Fl g , Fl Fl tagger
Tags input text by means of Viterbi algorithm.
.It Fl p , Fl Fl show-superficial
Prints the superficial form of the word along side the lexical form
in the output stream.
.It Fl f , Fl Fl first
Used in conjunction with
.Fl g
.Pq Fl Fl tagger
makes the tagger give all lexical forms of each word,
with the chosen one in the first place (after the lemma)
.It Fl d , Fl Fl debug
Print error (if any) or debug messages while operating.
.It Fl m , Fl Fl mark
Mark disambiguated words.
.It Fl h , Fl Fl help
Display a help message.
.It Fl z , Fl Fl null-flush        
Used in conjunction with
 .Fl g 
to flush the output after getting each null character.
.It Fl u , Fl Fl unigram=MODEL  
use unigram algorithm MODEL from <http://coltekin.net/cagri/papers/trmorph-tools.pdf>
.It Fl w , Fl Fl sliding-window 
use the Light Sliding Window algorithm  
.It Fl x , Fl Fl perceptron      
use the averaged perceptron algorithm
.It Fl e, Fl Fl skip-on-error  
Used with 
Fl xs 
to ignore certain types of errors with the training corpus

.El
.Sh FILES
These are the kinds of files used with each option:
.Bl -tag -width Ds
.It Ar dictionary
Full expanded dictionary file
.It Ar corpus
Training text corpus file
.It Ar tagger_spec
Tagger specification file, in XML format
.It Ar serialized_tagger
Tagger data file, built in the training and used while tagging
.It Ar tagged_corpus
Hand-tagged text corpus
.It Ar untagged_corpus
Untagged text corpus, morphological analysis of hand-tagged corpus to use
both jointly with
.Fl s
option
.It Ar input
Input file, stdin by default
.It Ar output
Output file, stdout by default
.El
.Sh SEE ALSO
.Xr apertium 1 ,
.Xr lt-comp 1 ,
.Xr lt-expand 1 ,
.Xr lt-proc 1
.Sh COPYRIGHT
Copyright \(co 2005, 2006 Universitat d'Alacant / Universidad de Alicante.
This is free software.
You may redistribute copies of it under the terms of
.Lk https://www.gnu.org/licenses/gpl.html the GNU General Public License .
.Sh BUGS
Many... lurking in the dark and waiting for you!