/***************************************************************************/
/* How to use the TreeTagger                                               */
/* Author: Helmut Schmid, University of Stuttgart, Germany                 */
/***************************************************************************/


The TreeTagger consists of two programs: train-tree-tagger is used to 
create a parameter file from a lexicon and a handtagged corpus. 
tree-tagger expects a parameter file and a text file as arguments and
annotates the text with part-of-speech tags. The file formats are
described below. By default, the programs are located in the ./bin
sub-directory.

If either of the programs is called without arguments, it will print
information about its usage.


Tagging
-------

Tagging is done with the tree-tagger program. It requires at least one
command line argument, the parameter file. If no input file is specified,
input will be read from stdin. If neither an input file nor an output file
is specified, the tagger will print to stdout.

tree-tagger {-options-} <parameter file> {<input file> {<output file>}}

Description of the command line arguments:

* <parameter file>: Name of a parameter file which was created with the 
  train-tree-tagger program.
* <input file>: Name of the file which is to be tagged. Each token in this 
  file has to be on a separate line. Tokens may contain blanks. It is possible
  to override the lexical information contained in the parameter file of the
  tagger by specifying a list of possible tags after a token. This list has
  to be preceded by a tab character and the elements are separated by tab 
  characters. This pretagging feature could be used e.g. to ensure that
  certain text-specific expressions are tagged properly.
  Punctuation marks must be on separate lines as well. Clitics (like "'s",
  "'re", and "'d" in English or "-la" and "-t-elle" in French) should be
  separated if they were separated in the training data. (The French and
  English parameter files available by ftp expect separation of clitics).
  Sample input file:
    He
    moved
    to
    New York City	NP
    .
* <output file>: Name of the file to which the tagger should write its output.

Further optional command line arguments:

* -token: tells the tagger to print the words also.
* -lemma: tells the tagger to print the lemmas of the words also.
* -sgml: tells the tagger to ignore tokens starting with '<' and ending
  with '>' (SGML tags).
- -no-unknown: If an unknown word is encountered, emit the word form
  as lemma. This was previously the default behaviour. Now, the default 
  behaviour is to print "<unknown>" as lemma.
- -threshold <p>: This option tells the tagger to print all tags of a
  word with a probability higher than <p> times the largest probability.
  (The tagger will use a different algorithm in this case and the set of
  best tags might be different from the tags generated without this
  option.)
- -prob: Print tag probabilities (in combination with option -threshold)
- -pt-with-prob: If this option is specified, then each pretagging tag
  (see above) has to be followed by a whitespace and a tag probability 
  value.
- -pt-with-lemma: If this option is specified, then each pretagging tag
  (see above) has to be followed by a whitespace and a lemma. Lemmas may 
  contain blanks.
  If both -pt-with-prob and -pt-with-lemma have been specified, then each
  pretagging tag is followed by a probability and a lemma in that order.

The options below are for advanced users. Please, read the papers on the 
TreeTagger to fully understand their meaning.

* -proto: If this option is specified, the tagger creates a file named
  "lexicon-protocol.txt", which contains information about the degree of
  ambiguity and about the other possible tags of a word form. The part of
  the lexicon in which the word form has been found is also indicated. 'f'
  means fullform lexicon and 's' means affix lexicon. 'h' means that the
  word contains a hyphen and that the part of the word following the
  hyphen has been found in the fullform lexicon.
* -eps <epsilon>: Value which is used to replace zero lexical frequencies.
  This is the case if a word/tag pair is contained in the lexicon but not
  in the training corpus. The choice of this parameter has only minor
  influence on the tagging accuracy.
* -base: If this option is specified, only lexical information is used
  for tagging but no contextual information about the preceding tags.
  This option is only useful in order to obtain a baseline result
  to which to compare the actual tagger output.



Training
--------

Training is done with the *train-tree-tagger* program. It expects at least
four command line arguments which are described below.

train-tree-tagger {options} <lexicon> <open class file> <input file> <output file>

Description of the command line arguments:

* <lexicon>: name of a file which contains the fullform lexicon. Each line 
  of the lexicon corresponds to one word form and contains the word form 
  and a sequence of tag-lemma pairs. Each tag is preceded by a tab character
  and each lemma is preceded by a blank or tab character.
  Example:

aback	RB aback
abacuses	NNS abacus
abandon	VB abandon	VBP abandon
abandoned	JJ abandoned	VBD abandon	VBN abandon
abandoning	VBG abandon

  Attention: Ordinal and cardinal numbers which consist of digits
  (like 1, 13, 1278 or 2. and 75.) should not be included in the
  lexicon. Otherwise, the tagger will not be able to learn how to tag
  numbers which are not listed in the lexicon. Numbers with unusual
  tags should be added to the lexicon, however. If the training
  program reports an error because the POS tag used for numbers is
  unknown, you should add a lexicon entry for one number.

  Remark: The tagger doesn't need the lemmata for tagging actually. If
  you do not have the lemma information or if you do not plan to
  annotate corpora with lemmas, you can replace the lemma with a dummy
  value, e.g. "-".

* <open class file>: name of a file which contains a list of open class tags
  i.e. possible tags of unknown word forms separated by whitespace.
  The tagger will use this information when it encounters unknown words,
  i.e. words which are not contained in the lexicon.
  Example: (for Penn Treebank tagset)

FW JJ JJR JJS NN NNS NP NPS RB RBR RBS VB VBD VBG VBN VBP VBZ

* <input file>: name of a file which contains tagged training data. The data
  must be in one-word-per-line format. This means that each line contains 
  one token and one tag in that order separated by a tabulator. 
  Punctuation marks are considered as tokens and must be tagged as well.
  The file should neither contain empty lines nor untagged SGML markup.
  Example:

Pierre  NP
Vinken  NP
,       ,
61      CD
years   NNS

* <output file>: name of the file in which the resulting tagger parameters 
  are stored.

The following parameters are optional. Read the papers on the TreeTagger to 
fully understand their meaning.

* -st <sent. tag>: the end-of-sentence part-of-speech tag, i.e. the tag which
  is assigned to sentence punctuation like ".", "!", "?". 
  Default is "SENT". It is important to set this option properly, if your
  tag for sentence punctuation is not "SENT".
* -cl <context length>: number of preceding words forming the statistical
  context. The default is 2 which corresponds to a trigram context. For
  small training corpora and/or large tagsets, it could be useful to reduce
  this parameter to 1.
* -dtg <min. decision tree gain>: Threshold - If the information gain at a 
  leaf node of the decision tree is below this threshold, the node is deleted.
* -sw <weight>: A smoothing parameter, which determines how much the
  probability distribution of some decision tree node is smoothed with the
  probability distribution of the parent node.
* -ecw <eq. class weight>: weight of the equivalence class based probability
  estimates.
* -atg <affix tree gain> Threshold - If the information gain at a leaf of an
  affix tree is below this threshold, it is deleted. The default is 1.2.

The accuracy of the TreeTagger usually improves, if different settings
of the above parameters are tested and the best combination is chosen.


Caveat: Make sure that the lexicon and the training corpus contain no
extra blanks. If the word form, for instance, is followed by a blank
and a tab character, the blank will be considered part of the word.