coocfreq - count co-occurrence frequencies for arbitrary features of nodes in a parallel treebank
coocfreq [OPTIONS] # count co-occurrence frequencies between category labels # in the parallel treebank of Sophie's World (Smultron) # and print the results in plain text files coocfreq -a sophie.xml -A sta -x cat -y cat -f cat.src -e cat.trg -c cat.cooc # count co-occurrences of 3-letter-suffix + category label of the parent node # of the source language tree with words from the target language tree # results will be stored in src.freq, trg.freq and cooc.freq coocfreq -a sophie.xml -A sta -x suffix=3:parent_cat -y word
This script counts frequencies and co-occurrence frequencies of source and target language features. It runs through the sentence aligned treebank and combines all node pairs. Note that co-occurrence frequencies in a sentence are
max( srcfreq(srcfeature) , trgfreq(trgfeature) ) to ensure Dice scores between 0 and 1!
- -f src.freq
Specify the name for the source language frequencies. The file will start with a line specifying the source language features used (starting with an initial '#'). All other lines have three TAB separated items: the feature string, a unique ID, and finally the frequency.
# word learned 682 4 stamp 722 3 hat 1056 5 what 399 20 again 220 14 of 27 118
- -e trg.freq
Specify the name for the target language frequencies. The format is the same as for the source language.
- -c cooc.freq
Specify the name for the co-occurrence frequencies. The first two lines specify the names of the files with the source and the target language frequencies and all other lines contain TAB separated source feature ID, target feature ID and co-occurrence frequency. Here is an example:
# source frequencies: word.src # target frequencies: word.trg 127 32 4 127 898 3 127 31 3 127 11 5 127 138 6 798 9 4 1250 1367 3
- -a align-file
Name of the alignment file (needs to include sentence alignment information). Parallel corpora without explicit sentence alignment files can also be used. For example, you can leave out this parameter if your parallel corpus is a plain text corpus with two separate files for source and target language and corresponding lines are aligned.
- -A align-file-format
This argument specifies the format of the sentence alignment file. For example, it can be OPUS (XCES format used in OPUS) or STA (Stockholm Tree Aligner format).
- -s src-file
Source language file of your parallel corpus.
- -S src-file-format
Format of the source language file. Default will be "plain text".
- -s trg-file
Target language file of your parallel corpus.
- -T trg-file-format
Format of the target language file. Default will be "plain text".
- -x srcfeatures
Features in the source language. Default feature is 'word' = surface words at each terminal node. All kinds of node attributes and combinations of features and contextual features can be used.
- -y trgfeatures
The same as -x but for the target language trees.
- -m freq-threshold
The frequency threshold. Default is 2.
A flag that enables storing the source and target language vocabulary in DB_FILE database files on disk to save memory when counting. This can be useful especially for complex (long) feature strings. Otherwise it doesn't save that much. The co-occurrence matrix is the big problem .....
Joerg Tiedemann, <email@example.com>
Copyright (C) 2009 by Joerg Tiedemann
This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.8.8 or, at your option, any later version of Perl 5 you may have available.