中英平行语料链接

# This is far from being the real WMT17 task - only toyset here
# you need to register to get UN data and CWT data. Also, by convention,
# this is EN to ZH - use translate_enzh_wmt8k_rev for ZH to EN task
#
# News Commentary, around 220k lines
# This dataset is only a small fraction of full WMT17 task

http://data.statmt.org/wmt17/translation-task/training-parallel-nc-v12.tgz

# Test set from News Commentary. 2000 lines

http://data.statmt.org/wmt17/translation-task/dev.tgz



# UN parallel corpus. 15,886,041 lines

# NOTE: You need to register to download dataset from official source


https://s3-us-west-2.amazonaws.com/twairball.wmt17.zh-en/UNv1.0.en-zh.tar.gz


# CWMT corpus

# casia2015: 1,050,000 lines
# casict2015: 2,036,833 lines
# datum2015:  1,000,003 lines
# datum2017: 1,999,968 lines
# NEU2017:  2,000,000 lines