In this assignment, your task is to take a sequence of Czech lemmas and produce the correct inflection for each of them. Since the data used in the assignment is licensed under the LDC, the assignment is written to require an account on the CLSP servers. Please contact Matt Post if you would like information on how to extract the PDT information from your own copy of the LDC catalog.

Originally developed for our Machine Translation course at Johns Hopkins University.

To setup the data, you need access to LDC2006T01. Set the environment variable $LDC2006T01 to point to the location of this data, then run

./scripts/extract.sh

You need Python 2.7. If you wish to parallelize the extraction (which takes a few hours), see that script for more info. Once the extraction is done, gather all the individually extracted files together:

mkdir data
for dir in train dtest etest; do 
  for type in lemma form tag tree; do 
    cat plain/$LDC2006T01/data/full/amw/$dir*/*.$type > data/$dir.$type
  done
done

This will create files in the data/ subdirectory corresponding to the files indicated in the assignment instructions.

Make sure to change the permissions of the test data so prevent peeking at the test data.

chmod 600 data/etest.form