Datasets:

Modalities:
Text
Formats:
text
Libraries:
Datasets
License:
File size: 1,907 Bytes
2a7b16c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ee5c52b
 
 
 
 
 
 
 
 
 
 
 
 
 
2a7b16c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
---
license: cc-by-4.0
language:
- zh
- ug
- bo
- en
---

# CUTE Dataset

CUTE (Chinese, Uyghur, Tibetan, English) 是一个大规模多语言数据集,专门设计用于增强低资源语言的跨语言知识迁移。数据集包含平行语料和非平行语料两部分,总规模约50GB。

## 数据集组成

### 平行语料 (24.70GB)
- 中文:2.62GB
- 英语:3.49GB
- 维吾尔语:7.37GB
- 藏语:11.22GB

### 非平行语料 (25.80GB)
- 中文:2.64GB
- 英语:3.49GB
- 维吾尔语:7.77GB
- 藏语:11.90GB

## 数据质量

数据集通过机器翻译生成,并经过人工评估验证:
- 中英翻译平均得分:9.1
- 中维翻译平均得分:8.5
- 中藏翻译平均得分:8.6

## 使用说明

### ⚠️ 平行语料重要提醒

**关于句对级对齐:**
- 四种语言的txt文件行数略有不同(如论文Table 2所示)
- 这是由于机器翻译过程中偶现的翻译行数不一致导致的(例如:一行中文可能翻译成两行维吾尔语)
- 目前只能保证**文档级平行**,无法实现严格的**句对级平行**

**数据特点:**
- 四种语言的文档整体平行度达到 99.98%
- 句子顺序可能存在错乱,但不影响大语言模型预训练
- 适用于多语言预训练任务,而非句对级翻译任务

## 相关链接

- 📝 [论文](https://aclanthology.org/2025.coling-main.670/)
- 🤖 [模型](https://huggingface.co/CMLI-NLP/CUTE-Llama)
- 📂 [GitHub](https://github.com/CMLI-NLP/CUTE)

## 引用

如果您使用了我们的数据集,请引用我们的论文:

```bibtex
@inproceedings{zhuang2025cute,
  title={CUTE: A Multilingual Dataset for Enhancing Cross-Lingual Knowledge Transfer in Low-Resource Languages},
  author={Zhuang, Wenhao and Sun, Yuan},
  booktitle={Proceedings of the 31st International Conference on Computational Linguistics},
  pages={10037--10046},
  year={2025}
}
```