This repository contains the tokenized version of newstest2014 and newstest2015 data sets for English<->Czech language pair. We have used the newstest2015 for tuning and newstest2014 for testing the models available for wmt16 tuning shared task.

The baseline BLEU scores are as follows:

English-to-Czech: 0.2226 [0.2159,0.2291]
Czech-to-English: 0.3040 [0.2973,0.3108]

For more details about wmt16 tuning shared task: http://www.statmt.org/wmt16/tuning-task/