wicho/kor_sae

Dataset Card for Structured Argument Extraction for Korean

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 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., 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.