wicho/kor_sae

---
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-description)
- [Dataset Summary](#dataset-summary)
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
- [Languages](#languages)
- [Dataset Structure](#dataset-structure)
- [Data Instances](#data-instances)
- [Data Fields](#data-fields)
- [Data Splits](#data-splits)
- [Dataset Creation](#dataset-creation)
- [Curation Rationale](#curation-rationale)
- [Source Data](#source-data)
- [Annotations](#annotations)
- [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
- [Social Impact of Dataset](#social-impact-of-dataset)
- [Discussion of Biases](#discussion-of-biases)
- [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
- [Dataset Curators](#dataset-curators)
- [Licensing Information](#licensing-information)
- [Citation Information](#citation-information)
- [Contributions](#contributions)

## Dataset Description

- **Homepage:** [Structured Argument Extraction for Korean](https://github.com/warnikchow/sae4k)
- **Repository:** [Structured Argument Extraction for Korean](https://github.com/warnikchow/sae4k)
- **Paper:** [Machines Getting with the Program: Understanding Intent Arguments of Non-Canonical Directives](https://arxiv.org/abs/1912.00342)
- **Point of Contact:** [Won Ik Cho]([e-mail])

### 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](https://huggingface.co/bert-base-uncased) 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 pair
* `intent_pair2`: a corresponding question/command pair
* `label`: determines the intent argument of the pair and can be one of `yes/no` (0), `alternative` (1), `wh- questions` (2), `prohibitions` (3), `requirements` (4) and `strong 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.](https://arxiv.org/abs/1811.04231), 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](https://github.com/stevhliu) for adding this dataset.

原始信息汇总

数据集概述

数据集名称

• 名称: Structured Argument Extraction for Korean
• 别名: SAE4K

数据集基本信息

• 语言: 韩语 (ko)
• 许可证: CC BY-SA-4.0
• 多语言性: 单语种
• 大小: 10K<n<100K
• 来源: 原始数据
• 任务类别: 文本分类
• 任务ID: 意图分类

数据集结构

• 特征:
  • intent_pair1: 字符串类型
  • intent_pair2: 字符串类型
  • label: 分类标签，包括 yes/no (0), alternative (1), wh- questions (2), prohibitions (3), requirements (4), strong requirements (5)
• 数据分割:
  • train: 30837个样本，数据大小2885167字节

数据集创建

• 注释创建者: 专家生成
• 语言创建者: 专家生成
• 数据收集: 来自Cho et al.构建的韩语单句语料库，用于识别指令/非指令
• 注释过程: 由三名具有计算语言学背景的韩语母语者进行分类和进一步的意图分类

使用考虑

• 许可证: 数据集根据CC BY-SA-4.0许可发布
• 引用信息: 引用时请参考Cho et al. (2019)的论文

数据集详细信息

数据实例

• 示例:

  { "intent_pair1": "내일 오후 다섯시 조별과제 일정 추가해줘", "intent_pair2": "내일 오후 다섯시 조별과제 일정 추가하기", "label": 4 }

数据字段

• intent_pair1: 问题/命令对
• intent_pair2: 对应的问题/命令对
• label: 意图分类标签，包括多种类型

数据分割

• train: 包含30837个样本，总数据大小为2885167字节