README.md

---
license: apache-2.0
language:
- fr
library_name: transformers
tags:
- nllb
- commonvoice
- pytorch
- pictograms
- translation
metrics:
- bleu
inference: false
---

# t2p-nllb-200-distilled-600M-commonvoice

*t2p-nllb-200-distilled-600M-commonvoice* is a text-to-pictograms translation model built by fine-tuning the [nllb-200-distilled-600M](https://huggingface.co/facebook/nllb-200-distilled-600M) model on a dataset of pairs of transcriptions / pictogram token sequence (each token is linked to a pictogram image from [ARASAAC](https://arasaac.org/)).
The model is used only for **inference**. 

## Training details

### Datasets

The [Propicto-commonvoice dataset](https://www.ortolang.fr/market/corpora/propicto) is used, which was created from the CommmonVoice v.15.0 corpus. 
This dataset was built with the method presented in the research paper titled ["A Multimodal French Corpus of Aligned Speech, Text, and Pictogram Sequences for Speech-to-Pictogram Machine Translation](https://aclanthology.org/2024.lrec-main.76/)" at LREC-Coling 2024. The dataset was split into training, validation, and test sets.
| **Split** | **Number of utterances** |
|:-----------:|:-----------------------:|
| train | 527,390 |
| valid | 16,124 |
| test | 16,120 |

### Parameters

A full list of the parameters is available in the config.json file. This is the arguments in the training pipeline :

```python
training_args = Seq2SeqTrainingArguments(
    output_dir="checkpoints_commonvoice/",
    evaluation_strategy="epoch",
    save_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=32,
    per_device_eval_batch_size=32,
    weight_decay=0.01,
    save_total_limit=3,
    num_train_epochs=40,
    predict_with_generate=True,
    fp16=True,
    load_best_model_at_end=True
)
```

### Evaluation

The model was evaluated with [sacreBLEU](https://huggingface.co/spaces/evaluate-metric/sacrebleu/blob/d94719691d29f7adf7151c8b1471de579a78a280/sacrebleu.py), where we compared the reference pictogram translation with the model hypothesis.

### Results

Comparison to other translation models :
| **Model** | **validation** | **test** |
|:-----------:|:-----------------------:|:-----------------------:|
| t2p-t5-large-commonvoice | 86.3 | 86.5 |
| t2p-nmt-commonvoice | 86.0 | 82.6 | 
| t2p-mbart-large-cc25-commonvoice | 72.3 | 72.3 |
| **t2p-nllb-200-distilled-600M-commonvoice** | **87.4** | **87.6** |

### Environmental Impact

Fine-tuning was performed using a single Nvidia V100 GPU with 32 GB of memory which took around 30 hours in total.

## Using t2p-nllb-200-distilled-600M-commonvoice model with HuggingFace transformers

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

source_lang = "fr"
target_lang = "frp"
max_input_length = 128
max_target_length = 128

tokenizer = AutoTokenizer.from_pretrained("Propicto/t2p-nllb-200-distilled-600M-commonvoice")
model = AutoModelForSeq2SeqLM.from_pretrained("Propicto/t2p-nllb-200-distilled-600M-commonvoice")

inputs = tokenizer("Je mange une pomme", return_tensors="pt").input_ids
outputs = model.generate(inputs.to("cuda:0"), max_new_tokens=40, do_sample=True, top_k=30, top_p=0.95)
pred = tokenizer.decode(outputs[0], skip_special_tokens=True)
```

## Linking the predicted sequence of tokens to the corresponding ARASAAC pictograms

```python
import pandas as pd

def process_output_trad(pred):
    return pred.split()

def read_lexicon(lexicon):
    df = pd.read_csv(lexicon, sep='\t')
    df['keyword_no_cat'] = df['lemma'].str.split(' #').str[0].str.strip().str.replace(' ', '_')
    return df

def get_id_picto_from_predicted_lemma(df_lexicon, lemma):
    id_picto = df_lexicon.loc[df_lexicon['keyword_no_cat'] == lemma, 'id_picto'].tolist()
    return (id_picto[0], lemma) if id_picto else (0, lemma)

lexicon = read_lexicon("lexicon.csv")
sentence_to_map = process_output_trad(pred)
pictogram_ids = [get_id_picto_from_predicted_lemma(lexicon, lemma) for lemma in sentence_to_map]
```

## Viewing the predicted sequence of ARASAAC pictograms in a HTML file

```python
def generate_html(ids):
    html_content = '<html><body>'
    for picto_id, lemma in ids:
        if picto_id != 0:  # ignore invalid IDs
            img_url = f"https://static.arasaac.org/pictograms/{picto_id}/{picto_id}_500.png"
            html_content += f'''
            <figure style="display:inline-block; margin:1px;">
                <img src="{img_url}" alt="{lemma}" width="200" height="200" />
                <figcaption>{lemma}</figcaption>
            </figure>
            '''
    html_content += '</body></html>'
    return html_content
    
html = generate_html(pictogram_ids)
with open("pictograms.html", "w") as file:
    file.write(html)
```

## Information

- **Language(s):** French
- **License:** Apache-2.0
- **Developed by:** Cécile Macaire
- **Funded by**
  - GENCI-IDRIS (Grant 2023-AD011013625R1)
  - PROPICTO ANR-20-CE93-0005
- **Authors**
  - Cécile Macaire
  - Chloé Dion
  - Emmanuelle Esperança-Rodier
  - Benjamin Lecouteux
  - Didier Schwab


## Citation

If you use this model for your own research work, please cite as follows:

```bibtex
@inproceedings{macaire_jeptaln2024,
  title = {{Approches cascade et de bout-en-bout pour la traduction automatique de la parole en pictogrammes}},
  author = {Macaire, C{\'e}cile and Dion, Chlo{\'e} and Schwab, Didier and Lecouteux, Benjamin and Esperan{\c c}a-Rodier, Emmanuelle},
  url = {https://inria.hal.science/hal-04623007},
  booktitle = {{35{\`e}mes Journ{\'e}es d'{\'E}tudes sur la Parole (JEP 2024) 31{\`e}me Conf{\'e}rence sur le Traitement Automatique des Langues Naturelles (TALN 2024) 26{\`e}me Rencontre des {\'E}tudiants Chercheurs en Informatique pour le Traitement Automatique des Langues (RECITAL 2024)}},
  address = {Toulouse, France},
  publisher = {{ATALA \& AFPC}},
  volume = {1 : articles longs et prises de position},
  pages = {22-35},
  year = {2024}
}
```