Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
- hu
|
6 |
+
tags:
|
7 |
+
- translation
|
8 |
+
license: gpl
|
9 |
+
metrics:
|
10 |
+
- sacrebleu
|
11 |
+
- chrf
|
12 |
+
widget:
|
13 |
+
- text: "This may not make much sense to you, sir, but I'd like to ask your permission to date your daughter."
|
14 |
+
|
15 |
+
---
|
16 |
+
|
17 |
+
# mT5 Translation model
|
18 |
+
- Source language: English
|
19 |
+
- Target language: Hungarian
|
20 |
+
|
21 |
+
- Pretrained model used: mT5-small
|
22 |
+
- Finetuned on subcorpora from OPUS
|
23 |
+
- Segments: 56.837.602
|
24 |
+
|
25 |
+
|
26 |
+
## Limitations
|
27 |
+
|
28 |
+
- max_source_length = 128
|
29 |
+
- max_target_length = 128
|
30 |
+
|
31 |
+
## Results
|
32 |
+
|
33 |
+
| Model | BLEU | chrF-3 |
|
34 |
+
| ------------- | ------------- | ------------- | ------------- |
|
35 |
+
| Google en-hu | 25.30 | 54.08 | 49.06 |
|
36 |
+
| BART | 36.89 | 60.77 | 56.4 |
|
37 |
+
| **mT5** | **27.69 ** | ** 53.73** | **48.57** |
|
38 |
+
|
39 |
+
|
40 |
+
## Citation
|
41 |
+
If you use this model, please cite the following paper:
|
42 |
+
|
43 |
+
@inproceedings {laki-yang-mt,
|
44 |
+
title = {{Jobban fordítunk magyarra, mint a Google!}},
|
45 |
+
booktitle = {XVIII. Magyar Számítógépes Nyelvészeti Konferencia},
|
46 |
+
year = {2022},
|
47 |
+
publisher = {Szegedi Tudományegyetem, Informatikai Intézet},
|
48 |
+
address = {Szeged, Magyarország},
|
49 |
+
author = {Laki, László and Yang, Zijian Győző},
|
50 |
+
}
|
51 |
+
|
52 |
+
```
|