README.md

---
license: apache-2.0
language:
- gl
- ca
metrics:
- bleu
library_name: fairseq
---
## Projecte Aina’s Galician-Catalan machine translation model
 
## Model description

This model was trained from scratch using the [Fairseq toolkit](https://fairseq.readthedocs.io/en/latest/) on a combination of Galician-Catalan datasets 
totalling 10.017.995 sentence pairs. 4.267.995 sentence pairs were parallel data collected from the web while the remaining 5.750.000 sentence pairs 
were parallel synthetic data created using the GL-ES translator of [Proxecto Nós](https://huggingface.co/proxectonos/Nos_MT-OpenNMT-es-gl). 
The model was evaluated on the Flores, TaCon and NTREX evaluation datasets. 

## Intended uses and limitations

You can use this model for machine translation from Galician to Catalan.

## How to use

### Usage
Required libraries:

```bash
pip install ctranslate2 pyonmttok
```

Translate a sentence using python
```python
import ctranslate2
import pyonmttok
from huggingface_hub import snapshot_download
model_dir = snapshot_download(repo_id="projecte-aina/aina-translator-gl-ca", revision="main")
tokenizer=pyonmttok.Tokenizer(mode="none", sp_model_path = model_dir + "/spm.model")
tokenized=tokenizer.tokenize("Benvido ao proxecto Ilenia.")
translator = ctranslate2.Translator(model_dir)
translated = translator.translate_batch([tokenized[0]])
print(tokenizer.detokenize(translated[0][0]['tokens']))
```

## Limitations and bias
At the time of submission, no measures have been taken to estimate the bias and toxicity embedded in the model. 
However, we are well aware that our models may be biased. We intend to conduct research in these areas in the future, and if completed, this model card will be updated. 

## Training

### Training data

The Catalan-Galician data is a combination of publicly available bilingual datasets collected from the web.
These datasets were concatenated before filtering to avoid intra-dataset duplicates and the final size was 4.267.995.
Additional 5.750.000 sentence pairs of synthetic parallel data were created from a random sampling 
of the [Projecte Aina ES-CA corpus](https://huggingface.co/projecte-aina/mt-aina-ca-es).

### Training procedure

### Data preparation

 All datasets are deduplicated and filtered to remove any sentence pairs with a cosine similarity of less than 0.75.
 This is done using sentence embeddings calculated using [LaBSE](https://huggingface.co/sentence-transformers/LaBSE). 
 The filtered datasets are then concatenated to form a final corpus of **10.017.995** and before training the punctuation is normalized using a 
 modified version of the join-single-file.py script from [SoftCatalà](https://github.com/Softcatala/nmt-models/blob/master/data-processing-tools/join-single-file.py)


#### Tokenization

 All data is tokenized using sentencepiece, with a 50 thousand token sentencepiece model  learned from the combination of all filtered training data. 
 This model is included.  

#### Hyperparameters

The model is based on the Transformer-XLarge proposed by [Subramanian et al.](https://aclanthology.org/2021.wmt-1.18.pdf)
The following hyperparameters were set on the Fairseq toolkit:

| Hyperparameter                 	| Value                        	|
|------------------------------------|----------------------------------|
| Architecture                   	| transformer_vaswani_wmt_en_de_big |
| Embedding size                 	| 1024                         	|
| Feedforward size               	| 4096                         	|
| Number of heads                	| 16                           	|
| Encoder layers                 	| 24                           	|
| Decoder layers                 	| 6                            	|
| Normalize before attention     	| True                         	|
| --share-decoder-input-output-embed | True                         	|
| --share-all-embeddings         	| True                         	|
| Effective batch size           	| 48.000                       	|
| Optimizer                      	| adam                         	|
| Adam betas                     	| (0.9, 0.980)                 	|
| Clip norm                      	| 0.0                          	|
| Learning rate                  	| 5e-4                         	|
| Lr. schedurer                  	| inverse sqrt                 	|
| Warmup updates                 	| 8000                         	|
| Dropout                        	| 0.1                          	|
| Label smoothing                	| 0.1                          	|

The model was trained for 24.000 updates on the parallel data collected from the web. 
This data was then concatenated with the synthetic parallel data and training continued for a total of 34.000 updates.
Weights were saved every 1000 updates and reported results are the average of the last 4 checkpoints.

## Evaluation

### Variable and metrics

We use the BLEU score for evaluation on test sets: [Flores-200](https://github.com/facebookresearch/flores/tree/main/flores200), 
[TaCon](https://elrc-share.eu/repository/browse/tacon-spanish-constitution-mt-test-set/84a96138b98611ec9c1a00155d02670628f3e6857b0f422abd82abc3795ec8c2/) 
and [NTREX](https://github.com/MicrosoftTranslator/NTREX).

### Evaluation results

Below are the evaluation results on the machine translation from Galician to Catalan compared to [Google Translate](https://translate.google.com/), 
[M2M100 1.2B](https://huggingface.co/facebook/m2m100_1.2B), [NLLB 200 3.3B](https://huggingface.co/facebook/nllb-200-3.3B) and 
[ NLLB-200's distilled 1.3B variant](https://huggingface.co/facebook/nllb-200-distilled-1.3B):

| Test set         	|Google Translate|M2M100 1.2B| NLLB 1.3B | NLLB 3.3 | aina-translator-gl-ca |
|----------------------|----|-------|-----------|------------------|---------------|
|Flores 101 devtest   	|**36,4**|32,6| 22,3   	| 34,3   	| 32,4     	|
| TaCON                 |48,4|56,5|32,2  	    | 54,1      	| **58,2**     	|
| NTREX                 |**34,7**|34,0|20,4    	| 34,2     	| 33,7     	|
| Average           	|39,0|41,0| 25,0 	| 40,9     	    | **41,4**      	|


## Additional information

### Author
The Language Technologies Unit from Barcelona Supercomputing Center.

### Contact
For further information, please send an email to <langtech@bsc.es>.

### Copyright
Copyright(c) 2023 by Language Technologies Unit, Barcelona Supercomputing Center.

### License
[Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE-2.0)

### Funding
This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU 
within the framework of the [project ILENIA](https://proyectoilenia.es/) 
with reference 2022/TL22/00215337, 2022/TL22/00215336, 2022/TL22/00215335 y 2022/TL22/00215334 

### Disclaimer

<details>
<summary>Click to expand</summary>

The model published in this repository is intended for a generalist purpose and is available to third parties under a permissive Apache License, Version 2.0. 

Be aware that the model may have biases and/or any other undesirable distortions.

When third parties deploy or provide systems and/or services to other parties using this model (or any system based on it) 
or become users of the model, they should note that it is their responsibility to mitigate the risks arising from its use and, 
in any event, to comply with applicable regulations, including regulations regarding the use of Artificial Intelligence.

In no event shall the owner and creator of the model (Barcelona Supercomputing Center) 
be liable for any results arising from the use made by third parties.

</details>