Datasets:
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
task_categories:
|
3 |
+
- translation
|
4 |
+
language:
|
5 |
+
- ha
|
6 |
+
- kr
|
7 |
+
- en
|
8 |
+
- fr
|
9 |
+
- sw
|
10 |
+
- swc
|
11 |
+
- ln
|
12 |
+
- nnd
|
13 |
+
- rhg
|
14 |
+
- ti
|
15 |
+
size_categories:
|
16 |
+
- 10K<n<100K
|
17 |
+
---
|
18 |
+
# Gamayun Language Data Kits
|
19 |
+
|
20 |
+
There are more than 7,000 languages in the world, yet only a small proportion of them have language data presence in public. Translators without Borders’(TWB) Gamayun kits are a starting point for developing audio and text corpora for languages without pre-existing data resources. We create parallel data for a language by translating a pre-compiled set of general-domain sentences in English. If audio data is needed, these translated sentences are recorded by native speakers.
|
21 |
+
|
22 |
+
To scale corpus production, we offer four dataset versions:
|
23 |
+
|
24 |
+
- Mini-kit of 5,000 sentences (`kit5k`)
|
25 |
+
- Small-kit of 10,000 sentences (`kit10k`)
|
26 |
+
- Medium-kit of 15,000 sentences (`kit15k`)
|
27 |
+
- Large-kit of 30,000 sentences (`kit30k`)
|
28 |
+
|
29 |
+
Since each kit contains a unique set of sentences, it is possible to combine them to reach a maximum dataset size of 60,000 sentences. Recording these sentences result approximately in 75 hours of audio data.
|
30 |
+
|
31 |
+
## Source sentences (`core`)
|
32 |
+
|
33 |
+
Sentences in `core` directory are in English, French and Spanish and are sourced from the [Tatoeba repository](https://tatoeba.org). Sentence selection algorithm ensures representation of most frequently used words in the language. For more information, please refer to [corepus-gen repository](https://github.com/translatorswb/corepus-gen). `etc` directories contain sentence id's as used in the Tatoeba corpus.
|
34 |
+
|
35 |
+
## Parallel corpora (`parallel`)
|
36 |
+
|
37 |
+
Translations of the kits are performed by professionals and volunteers of TWB's translator community. A complete list of translated sentences are:
|
38 |
+
|
39 |
+
| Language | Pair | # Segments | Source |
|
40 |
+
|------|--------|--------|--------|
|
41 |
+
| Hausa | English | 15,000 | Tatoeba |
|
42 |
+
| Kanuri | English | 5,000 | Tatoeba |
|
43 |
+
| Nande | French | 15,000 | Tatoeba |
|
44 |
+
| Rohingya | English | 5,000 | Tatoeba |
|
45 |
+
| Swahili (Coastal) | English | 5,000 | Tatoeba |
|
46 |
+
| Swahili (Congolese) | French | 25,302 | Tatoeba |
|
47 |
+
|
48 |
+
## Reference
|
49 |
+
|
50 |
+
More on [Gamayun, language equity initiative](https://translatorswithoutborders.org/gamayun/)
|
51 |
+
|
52 |
+
Gamayun kits are officially published in the [Gamayun portal](https://gamayun.translatorswb.org/data/). Conditions for use are described in `LICENSE.txt`.
|
53 |
+
|
54 |
+
If you need to cite Gamayun kits:
|
55 |
+
|
56 |
+
```
|
57 |
+
Alp Öktem, Muhannad Albayk Jaam, Eric DeLuca, Grace Tang
|
58 |
+
Gamayun – Language Technology for Humanitarian Response
|
59 |
+
In: 2020 IEEE Global Humanitarian Technology Conference (GHTC)
|
60 |
+
2020 October 29 - November 1; Virtual.
|
61 |
+
Link: https://ieeexplore.ieee.org/document/9342939
|
62 |
+
```
|