PD&CFT: A Chinese Reading Comprehension Dataset

We release the first Chinese reading comprehension dataset, which includes People Daily and Children's Fairy Tale (PD&CFT). We hope this would speed up the process for future research in machine comprehension.

Directory Guide

• people_daily

  • pd.zip
    • pd.train (training file)
    • pd.valid (validation file)
    • pd.test (test file)

• children_fairy_tale

  • cft.zip
    • cft.test.auto (automatically generated test set)
    • cft.test.human (human evaluated test set)

NOTE: As we have illustrated in the paper, the human evaluation test set is NOT the query proposed by human. The human evaluation set is also the Cloze-style queries, but those easy ones are eliminated.

Statistics

The statistics of the dataset is listed as below.

Data Format

Here is a sample of People Daily data,

1 |||  1 1             2013                         
2 |||         500    29.6%   1997      
3 |||      26.5%   1996      
4 |||   38.3% 
5 |||  12 31                 
6 |||     12     78.1    11  72 
7 |||         2013  1995          
8 |||                   
9 |||      XXXXX               
10 |||                      
11 |||      XXXXX                ||| 

This document consists of 10 sentences, each sentence is in the form of

sentence_id(space)|||(space)sentence

and the last line indicate the Query and Answer

sentence_id(space)|||(space)Query(space)|||(space)Answer

Downloads

Our data is avaliable through Github

People Daily & Children's Fairy Tale (PD&CFT)

Paper

Our paper is avaliable through

ACL Anthology

arXiv Pre-print

International Standard Language Resource Number (ISLRN)

ISLRN: 343-112-755-039-0

http://www.islrn.org/resources/resources_info/7838/

Licence

Our data is under CC-BY-SA-4.0 licence.

Reference

If you wish to use this data in your work, please cite

@InProceedings{cui-etal-2016-consensus,
  title		= {Consensus Attention-based Neural Networks for Chinese Reading Comprehension},
  author	= {Cui, Yiming and Liu, Ting and Chen, Zhipeng and Wang, Shijin and Hu, Guoping},
  booktitle = {Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers},
  year      = {2016},
  address   = {Osaka, Japan},
  pages     = {1777--1786},
}

You may also interested in ...

DeepMind CNN / Daily Mail data Pre-processed Data (recommended) Original Data

Children's Book Test (CBTest) Original Data

Contact

For any problems concerning the paper or data, please contact: admin [AT] ymcui [dot] com