hindirc-data's Introduction

HindiRC-Data

This repo contains data for HindiRC. Please cite our paper when using our dataset.

About HindiRC

HindiRC is the first Reading Comprehension dataset for Hindi.

The dataset has been split into grades 2-5 according the educational reading proficiency.

The data has been sourced from educational websites https://sandeepbarouli.com/ and 2classnotes. Hence, it is completely natural and formal.

Xml Schema of the Dataset

Reading Comprehension consists of a passage, <passage> and a set of questions, <q>.
For HindiRC, the answers <a> belong to the passage.
The index of these answers in the passage is given in <l>.
If an answer spans multiple sentences, we consider the most important sentence only. However, the dataset does provide the index of these sentences is given in <lg>.
Additionally, the dataset also provides a copy of the passage with all anaphors resolved.

The following tags have been used:

Data	Xml tags
Passage - raw data	<passage>
Passage data with anaphoras replaced with referent. This was done manually.	<anaphoraresolved>
Stemmed version of the passage	<stemmed>
Question - raw data	<q>
Answer - raw data	<a>
Answer - index in passage	<l>
Multi sentence answer - index in passage	<lg>

References

Please cite our paper when using this dataset:

Kaveri Anuranjana*, Vijjini Anvesh Rao*, Radhika Mamidi, "HindiRC: A Dataset for Reading Comprehension in Hindi", 20^th International Conference on Computational Linguistics and Intelligent Text Processing, CICLing, 2019.

@inproceedings{anuranjana2019hindirc,
  title     = {HindiRC: A Dataset for Reading Comprehension in Hindi},
  author    = {Anuranjana, Kaveri and Rao, Vijjini Anvesh and Mamidi, Radhika},
  booktitle = {20th International Conference on Computational Linguistics and Intelligent Text},
  year      = {2019}
}

hindirc-data's People