File size: 1,385 Bytes
24e5b4e
aeebad0
 
 
24e5b4e
 
aeebad0
24e5b4e
 
 
 
aeebad0
 
 
24e5b4e
 
 
5083d70
4c41635
5083d70
24e5b4e
 
 
 
 
 
 
 
 
 
650b1b7
24e5b4e
 
 
 
 
ca4aa0e
24e5b4e
 
 
6cee675
24e5b4e
 
 
 
 
0ab3892
 
24e5b4e
 
f2601a7
 
 
 
 
 
24e5b4e
 
 
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
---
language:
- en
- hu
tags:
- translation
license: apache-2.0
metrics:
- sacrebleu
- chrf
widget:
- text: >-
    This may not make much sense to you, sir, but I'd like to ask your
    permission to date your daughter.
---

# mT5 Translation model

For further models, scripts and details, see [our repository](https://github.com/nytud/machine-translation) or [our demo site](https://juniper.nytud.hu/demo/nlp).

- Source language: English
- Target language: Hungarian

- Pretrained model used: mT5-small
- Finetuned on subcorpora from OPUS
	- Segments: 56.837.602


## Limitations

- tokenized input text (tokenizer: [HuSpaCy](https://huggingface.co/huspacy))
- max_source_length = 128
- max_target_length = 128

## Results

| Model | BLEU | chrF-3 | chrF-6 |
| ------------- | ------------- | ------------- | ------------- |
| Google en-hu  | 25.30  | 54.08 | 49.06 |
| BART | 36.89 | 60.77 | 56.4 |
| **mT5** | **27.69**  | **53.73** | **48.57** |


## Citation
If you use this model, please cite the following paper:

```

@inproceedings {laki-yang-mt,
    title = {{Jobban fordítunk magyarra, mint a Google!}},
	booktitle = {XVIII. Magyar Számítógépes Nyelvészeti Konferencia},
	year = {2022},
	publisher = {Szegedi Tudományegyetem, Informatikai Intézet},
	address = {Szeged, Magyarország},
	author = {Laki, László and Yang, Zijian Győző},
	pages = {357--372}
}

```