File size: 5,160 Bytes
f0c8b83
 
8ec480b
 
 
 
 
 
 
 
 
 
 
 
 
 
f0c8b83
8ec480b
6814ce3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6868b4d
 
 
 
 
 
 
8ec480b
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
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
---
license: apache-2.0
task_categories:
  - text-generation
  - question-answering
  - conversational
language:
  - ta
size_categories:
  - 1K<n<2K
language_creators:
  - expert-generated
  - machine-generated
multilinguality:
  - monolingual
pretty_name: tamil_stories
---

# Summary
`tamil_stories` is an open source dataset of instruct-style records generated by scraping publicly available short stories on the following websites.
- [Siruvarmalar](https://www.siruvarmalar.com/)
- [Tamilsurangam](https://www.tamilsurangam.in/)

Apart from scraping and automated cleaning, the data was also tagged manually by a group of volunteers. 
This dataset created as part of [Aya Open Science Initiative](https://sites.google.com/cohere.com/aya-en/home) by Cohere For AI.
This dataset can be used for any purpose, whether academic or commercial, under the terms of the [Apache 2.0](https://opensource.org/license/apache-2-0) License.

Supported Tasks:
- Training LLMs
- Synthetic Data Generation
- Data Augmentation
- Question Answering

Languages: Tamil Version: 1.0

# Dataset Overview
`tamil_stories` is a corpus of 1202 records generated by converting scraped stories into instruction-style. This Dataset can be used for the following tasks:
- Given the story, generate the appropriate title for the story.
- Given some prominent characters/words from the story along with the title, generate the complete story.


# Intended Uses
While immediately valuable for instruction fine tuning large language models, as a corpus of instruction prompts, this dataset also presents a valuable opportunity for synthetic data generation. 
For example, prompt-completions could be submitted as few-shot examples to a large open language model to generate new stories in a similar style.

# Dataset

## Load with Datasets
To load this dataset with Datasets, you'll just need to install Datasets as `pip install datasets --upgrade` and then use the following code:

```python
from datasets import load_dataset
ds = load_dataset('aitamilnadu/tamil_stories')
```

## Purpose of Collection
Tamil is a low-resource language (inspite of having rich literature) where there are no instruct-style dataset to the best of my knowledge. 
This was created as a part of [Aya Open Science Initiative](https://sites.google.com/cohere.com/aya-en/home) from Cohere For AI to make sure Tamil is well represented in the space of AI/ML. 
Unlike other datasets that are limited to non-commercial use, this dataset can be used, modified, and extended for any purpose, including academic or commercial applications.

## Sources
The following are the websites from which the data is scraped.
- [Siruvarmalar](https://www.siruvarmalar.com/)
- [Tamilsurangam](https://www.tamilsurangam.in/)

The scripts used for scraping and cleaning the data can be found here:
[Tamilsurangam Webscraper](https://github.com/miluckshan-j/tamilsurangam-webscraper), 
[Siruvarmalar Webscraper](https://github.com/miluckshan-j/siruvarmalar-webscraper)
[Dataset Creation](https://github.com/miluckshan-j/tamil-dataset-creation-tools)

- The scraped data is carefully analysed making sure there are no missed words, spelling mistakes and the data is in Tamil only.
- Words/characters are manually added to stories by volunteers.
- Finally, converted the pre-processed data into instruct-style prompts and completions.


## Templates
For the creation of instruct-style prompts and completions from the scraped data, the following templates were used:
1. Given the story, generate the appropriate title for the story.

```python
Prompt:
கீழே கொடுக்கப்பட்டுள்ள கதைக்குப் பொருத்தமான தலைப்பைக் கொடு.
கதை: {Story}

Completion:
கொடுக்கப்பட்டுள்ள கதைக்குப் பொருத்தமான தலைப்பு '{Title}' என்பதாகும்.
```

2. Given some prominent characters/words from the story along with the title, generate the complete story.
```python
Prompt:
கீழே கொடுக்கப்பட்டுள்ள வார்த்தைகளையும் தலைப்பையும் பயன்படுத்தி சிறு கதை எழுதுக.
வார்த்தைகள்: {Comma_Seperated_Words}
தலைப்பு: {Title}

Completion:
{Story}
```

## Personal or Sensitive Data
This dataset contains public information. To my knowledge, there are no private person’s personal identifiers or sensitive information.

## Language
Tamil

# Known Limitations
- The meanings used in the prompts/completions are chosen randomly based on the availability of complete sentences and this may reflect some bias by ignoring other meanings written by other scholars.

# Contributors
- [miluckshan-j](https://github.com/miluckshan-j)
- [chezhian0599](https://github.com/chezhian0599)
- [AbinayaM02](https://github.com/AbinayaM02)
- [anitha67](https://github.com/anitha67)
- [vijiv11](https://github.com/vijiv11)
- [Lagstill](https://github.com/Lagstill)
- [Merlinecodes](https://github.com/Merlinecodes)