File size: 2,785 Bytes
087c7f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ce527d
40cdaa1
 
 
 
 
 
e3fc83e
 
 
40cdaa1
 
 
 
 
 
 
 
 
 
 
 
 
 
e3488eb
 
 
 
 
 
 
 
 
 
 
 
 
 
6ee4163
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e3488eb
 
 
 
 
 
 
faea03a
e3488eb
 
 
 
 
 
 
 
 
40cdaa1
 
 
 
 
 
 
 
 
 
 
 
 
8ce527d
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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
---
task_categories:
- translation
language:
- en
- kn
tags:
- machine-translation
- nllb
- english
- kannada
- parallel-corpus
- multilingual
- low-resource
pretty_name: English-Kannada NLLB Machine Translation Dataset
size_categories:
- 10K<n<100K
---

# English-Kannada NLLB Machine Translation Dataset

This dataset contains English-Kannada parallel text from NLLB dataset along with new NLLB model translations.

## Dataset Structure

- Train: 16,702 examples
- Test: 8,295 examples  
- Validation: 4,017 examples

### Features

- `en`: Source English text (from NLLB Dataset)
- `kn`: Human-translated Kannada text (from NLLB Dataset)
- `kn_nllb`: Machine-translated Kannada text using facebook/nllb-200-distilled-600M model

While `kn` translations are available in the NLLB dataset, their quality is poor. Therefore, we created `kn_nllb` by translating the English source text using NLLB's distilled model to obtain cleaner translations.

## Preprocessing

- Filtered: Minimum 5 words in both English and NLLB-translated Kannada texts
- Train-test split: 2:1 ratio


## Sample Dataset

| en | kn | kn_nllb |
|---|---|---|
| The weather is beautiful today. | ಇಂದು ಹವಾಮಾನ ಅದ್ಭುತವಾಗಿದೆ. | ಇಂದು ಹವಾಮಾನ ಸುಂದರವಾಗಿದೆ. |
| I love reading interesting books. | ನಾನು ಆಸಕ್ತಿದಾಯಕ ಪುಸ್ತಕಗಳನ್ನು ಓದಲು ಇಷ್ಟಪಡುತ್ತೇನೆ. | ನಾನು ಆಸಕ್ತಿದಾಯಕ ಪುಸ್ತಕಗಳನ್ನು ಓದಲು ಪ್ರೀತಿಸುತ್ತೇನೆ. |

## Loading the Dataset

### Using Pandas
```python
import pandas as pd

splits = {
   'train': 'data/train-00000-of-00001.parquet',
   'validation': 'data/validation-00000-of-00001.parquet', 
   'test': 'data/test-00000-of-00001.parquet'
}

# Load all splits into DataFrames
dataframes = {}
for split, path in splits.items():
   dataframes[split] = pd.read_parquet(f"hf://datasets/pavan-naik/mt-nllb-en-kn/{path}")

# Access individual splits
train_data = dataframes['train']
test_data = dataframes['test']
validation_data = dataframes['validation']

```

### Using HuggingFace 🤗 Datasets
```python
from datasets import load_dataset

# Load from HuggingFace Hub
dataset = load_dataset("pavan-naik/mt-nllb-en-kn")

# Access splits
train_data = dataset["train"]
test_data = dataset["test"]
validation_data = dataset["validation"]

```


## Use Cases

- Evaluating NLLB translations for English-Kannada
- Training/fine-tuning MT models
- Analyzing translation quality: NLLB Dataset vs NLLB Model outputs

## Citation

- NLLB Team et al. "No Language Left Behind: Scaling Human-Centered Machine Translation"
- OPUS parallel corpus

## License

Same as NLLB license