Datasets:
annotations_creators:
- crowdsourced
language_creators:
- crowdsourced
language:
- ro-RO
license:
- cc-by-4.0
multilinguality:
- monolingual
pretty_name: 'RoITD: Romanian IT Question Answering Dataset'
size_categories:
- unknown
source_datasets:
- original
task_categories:
- question-answering
task_ids:
- extractive-qa
Dataset Summary
We introduce a Romanian IT Dataset (RoITD) resembling SQuAD 1.1. RoITD consists of 9575 Romanian QA pairs formulated by crowd workers. QA pairs are based on 5043 articles from Romanian Wikipedia articles describing IT and household products. Of the total number of questions, 5103 are possible (i.e. the correct answer can be found within the paragraph) and 4472 are not possible (i.e. the given answer is a "plausible answer" and not correct)
Dataset Structure
The data structure follows the format of SQuAD, which contains several attributes such as question, id, text, `answer_start, is_impossible and context. The paragraph provided to crowd sourcing workers is stored in the field context. This incorporates manually-selected paragraphs from Wikipedia. The field id is comprised of a randomly assigned unique identification number for the answer-question pair. Only the numbers "0" and "1" are allowed in the is_impossible field. The category "A" is assigned the value "0", indicating that the answer is correct. The value "1" corresponds to the category "U", indicating a plausible answer. The question posed by the source crowd source worker is represented by the field question. The field answer_start keeps track of the character index marking the beginning of an answer.