metadata

language: et
dataset_info:
  features:
    - name: context
      dtype: string
    - name: question
      dtype: string
    - name: answers
      list:
        - name: answer_start
          dtype: int64
        - name: text
          dtype: string
    - name: title
      dtype: string
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 772885
      num_examples: 776
    - name: test
      num_bytes: 590275
      num_examples: 603
  download_size: 257799
  dataset_size: 1363160
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
license: cc-by-4.0
task_categories:
  - question-answering
pretty_name: EstQA

Estonian Question Answering dataset

Dataset for extractive question answering in Estonian. It is based on Wikipedia articles, pre-filtered via PageRank. Annotation was done by one person.
Train set includes 776 context-question-answer triplets. There are several possible answers per question, each in a separate triplet. Number of different questions is 512.
Test set includes 603 samples. Each sample contains one or more golden answers. Altogether there are 892 golden ansewrs.

Change log

Version 1.2: restructured from https://huggingface.co/datasets/anukaver/EstQA to use flattened structure
Version 1.1: Test set v1.1 adds some more golden answers.

Reference

If you use this dataset for research, please cite the following paper:

@mastersthesis{mastersthesis,
  author       = {Anu Käver}, 
  title        = {Extractive Question Answering for Estonian Language},
  school       = {Tallinn University of Technology (TalTech)},
  year         = 2021
}