Datasets:
annotations_creators:
- expert-generated
language_creators:
- expert-generated
language:
- ko
license:
- cc-by-sa-4.0
multilinguality:
- monolingual
size_categories:
- 10K<n<100K
source_datasets:
- original
task_categories:
- text-classification
task_ids:
- intent-classification
pretty_name: Structured Argument Extraction for Korean
dataset_info:
features:
- name: intent_pair1
dtype: string
- name: intent_pair2
dtype: string
- name: label
dtype:
class_label:
names:
'0': yes/no
'1': alternative
'2': wh- questions
'3': prohibitions
'4': requirements
'5': strong requirements
splits:
- name: train
num_bytes: 2885167
num_examples: 30837
download_size: 2545926
dataset_size: 2885167
Dataset Card for Structured Argument Extraction for Korean
Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Homepage: Structured Argument Extraction for Korean
- Repository: Structured Argument Extraction for Korean
- Paper: Machines Getting with the Program: Understanding Intent Arguments of Non-Canonical Directives
- Point of Contact: Won Ik Cho
Dataset Summary
The Structured Argument Extraction for Korean dataset is a set of question-argument and command-argument pairs with their respective question type label and negativeness label. Often times, agents like Alexa or Siri, encounter conversations without a clear objective from the user. The goal of this dataset is to extract the intent argument of a given utterance pair without a clear directive. This may yield a more robust agent capable of parsing more non-canonical forms of speech.
Supported Tasks and Leaderboards
intent_classification
: The dataset can be trained with a Transformer like BERT to classify the intent argument or a question/command pair in Korean, and it's performance can be measured by it's BERTScore.
Languages
The text in the dataset is in Korean and the associated is BCP-47 code is ko-KR
.
Dataset Structure
Data Instances
An example data instance contains a question or command pair and its label:
{
"intent_pair1": "내일 오후 다섯시 조별과제 일정 추가해줘"
"intent_pair2": "내일 오후 다섯시 조별과제 일정 추가하기"
"label": 4
}
Data Fields
intent_pair1
: a question/command pairintent_pair2
: a corresponding question/command pairlabel
: determines the intent argument of the pair and can be one ofyes/no
(0),alternative
(1),wh- questions
(2),prohibitions
(3),requirements
(4) andstrong requirements
(5)
Data Splits
The corpus contains 30,837 examples.
Dataset Creation
Curation Rationale
The Structured Argument Extraction for Korean dataset was curated to help train models extract intent arguments from utterances without a clear objective or when the user uses non-canonical forms of speech. This is especially helpful in Korean because in English, the Who, what, where, when and why
usually comes in the beginning, but this isn't necessarily the case in the Korean language. So for low-resource languages, this lack of data can be a bottleneck for comprehension performance.
Source Data
Initial Data Collection and Normalization
The corpus was taken from the one constructed by Cho et al., a Korean single utterance corpus for identifying directives/non-directives that contains a wide variety of non-canonical directives.
Who are the source language producers?
Korean speakers are the source language producers.
Annotations
Annotation process
Utterances were categorized as question or command arguments and then further classified according to their intent argument.
Who are the annotators?
The annotation was done by three Korean natives with a background in computational linguistics.
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
The dataset is curated by Won Ik Cho, Young Ki Moon, Sangwhan Moon, Seok Min Kim and Nam Soo Kim.
Licensing Information
The dataset is licensed under the CC BY-SA-4.0.
Citation Information
@article{cho2019machines,
title={Machines Getting with the Program: Understanding Intent Arguments of Non-Canonical Directives},
author={Cho, Won Ik and Moon, Young Ki and Moon, Sangwhan and Kim, Seok Min and Kim, Nam Soo},
journal={arXiv preprint arXiv:1912.00342},
year={2019}
}
Contributions
Thanks to @stevhliu for adding this dataset.