|
|
---
|
|
|
license: cc-by-sa-4.0
|
|
|
language:
|
|
|
- ja
|
|
|
tags:
|
|
|
- definition generation
|
|
|
extra_gated_prompt: >-
|
|
|
To prevent leaking the dataset to the training data of future models, it should not be distributed on the open web.
|
|
|
extra_gated_fields:
|
|
|
I agree not to make the D3J dataset, or any part of it, available on the open web: checkbox
|
|
|
---
|
|
|
|
|
|
This repository presents two resources:
|
|
|
|
|
|
- D3J, the Dictionary Definition Dataset for Japanese. D3J provides lexically simple definitions for 325 Japanese headwords, created in collaboration with a professional lexicographer.
|
|
|
- TUBE16K, a defining vocabulary used for building D3J. TUBE16K consists of (1) the 16,000 most frequent words from the Japanese section of [the TUBELEX corpus](https://github.com/naist-nlp/tubelex), and (2) linguistic terms to concisely describe meanings or usages of headwords. Note that words in TUBE16K are SUWs (short unit words).
|
|
|
|
|
|
Details of D3J and TUBE16K are described in [our paper](https://arxiv.org/abs/2601.01842). Consider citing it when using the resources.
|
|
|
|
|
|
```
|
|
|
@misc{ide2026,
|
|
|
title={Towards Automated Lexicography: Generating and Evaluating Definitions for Learner's Dictionaries},
|
|
|
author={Yusuke Ide and Adam Nohejl and Joshua Tanner and Hitomi Yanaka and Christopher Lindsay and Taro Watanabe},
|
|
|
year={2026},
|
|
|
eprint={2601.01842},
|
|
|
archivePrefix={arXiv},
|
|
|
primaryClass={cs.CL},
|
|
|
url={https://arxiv.org/abs/2601.01842},
|
|
|
}
|
|
|
```
|
|
|
|
