Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Tags:
License:
wannaphong's picture
Update README.md
b80d99b verified
---
dataset_info:
features:
- name: id
dtype: int64
- name: q
dtype: string
- name: a
dtype: string
splits:
- name: th2en
num_bytes: 261404758
num_examples: 438008
- name: en2th
num_bytes: 261020179
num_examples: 437191
- name: mtinstruct_th2en
num_bytes: 194991586
num_examples: 438008
- name: mtinstruct_en2th
num_bytes: 194712580
num_examples: 437191
download_size: 349631371
dataset_size: 912129103
configs:
- config_name: default
data_files:
- split: th2en
path: data/th2en-*
- split: en2th
path: data/en2th-*
- split: mtinstruct_th2en
path: data/mtinstruct_th2en-*
- split: mtinstruct_en2th
path: data/mtinstruct_en2th-*
license: cc-by-3.0
language:
- th
- en
---
# Thai Aligninstruct Dataset
This project was try to make English-Thai Aligninstruct.
I use simalign for making the High Quality Word Alignments Without Parallel Training Data Using Static and Contextualized Embeddings and use [deepcut](https://github.com/rkcosmos/deepcut) (by [lekcut](https://github.com/PyThaiNLP/LEKCut)) to tokenize Thai word.
The data use [scb-mt-en-th-2020](https://huggingface.co/datasets/scb_mt_enth_2020).
## References
- Aligninstruct: [https://arxiv.org/abs/2401.05811v1](https://arxiv.org/abs/2401.05811v1)
- SimAlign: High Quality Word Alignments Without Parallel Training Data Using Static and Contextualized Embeddings: [https://aclanthology.org/2020.findings-emnlp.147/](https://aclanthology.org/2020.findings-emnlp.147/)
- scb-mt-en-th-2020: A Large English-Thai Parallel Corpus: [https://arxiv.org/abs/2007.03541](https://arxiv.org/abs/2007.03541)