Datasets:
dataset_info:
features:
- name: Document
dtype: string
- name: Sentence
dtype: string
- name: is_vague
dtype: int64
splits:
- name: train
num_bytes: 4033144
num_examples: 12647
download_size: 1623870
dataset_size: 4033144
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- text-classification
- feature-extraction
language:
- ru
pretty_name: RuVS
size_categories:
- 10K<n<100K
The dataset contains 12647 manually annotated Russian sentences annotated according to vagueness criterion. The value of vagueness is represented in the column is_vague where 0 means the absence of vague expressions in the sentence, while 1 indicates the presence of vagueness.
DATA
The data are represented by the interviews with Russian civil servants conducted from 2004 to 2021. The interviews were collected from the websites ConsultantPlus and Garant.ru as well as from the website of Echo of Moscow - a radio station switched off 04.03.2022 at the request of the prosecutor's office.
ANNOTATION
The dataset was annotated and checked twice afterwards by one expert with an interwal of two weeks. The annotation was limited to five categories of the extracted contexts: gradable adjectives (tall, high, beautiful), gradable adverbs (deeply, reliably), fuzzy quantifiers (around NUM, approximately NUM, more than NUM), fuzzy time expressions (after august 2016, often, seldom) and degree terms (very, extremely, slightly). Further work on this data implies the search for new annotators to evaluate inter-annotator agreement. If you are fluent in Russian and would like to contribute, please write to the email address below. Your participation will be inestimably significant.
USE
This dataset can serve as training data for binary classification model to detect vagueness in Russian texts.
CONTACT INFO
Please send all questions and suggestions to schepovetskaya@hse.ru