JoanneAB/translation_fr-als

Translation from French to Alsatian.

What is it ?

This repository proposes a model for a translation task from French to Alsatian (dialiect in Alsace, North-East France) languages. Because Alsatian is a spoken and non-standard language with significant regional variation across Alsace, the development of a robust and accurate models is not expected. The objective of this project is to explore and apply the concepts learned during the deep-learning course on a challenging application. Words and sentences in French and their translation in Alsatian have been downloaded from the Internet from several sources.

A full report is availaible report.pdf.

How to use it ?

This study uses python libraries and can be run using the jupyter-notebook main_translator_fr-als.ipynb. All datafile and local functions are available in this repository.

Model evaluation from people

The evaluation of some translations has been proposed to people using this form : human_evaluation_form. Summary of the evaluation is presented in the table output_human_evaluation_table

The model is available in model/t5-base_essai6_evaluation

Technology stack

Data processing

• numpy
• pandas
• datasets

Modeling

• sklearn

Deep learning

• transformers (AutoTokenizer, AutoConfig, NllbTokenizer, AutoModelForSeq2SeqLM, DataCollatorForSeq2Seq, Seq2SeqTrainingArguments, Seq2SeqTrainer, pipeline)

Model evaluation:

• nltk
• evaluate
• bleu
• sacrebleu

Credits

This model has been developed as part of a project for the "Deep Learning with Python" class for Data Science Master's course in Data ScienceTech Institute (DSTI, France). Source codes and supplementary files are available at https://github.com/JoanneAB/Translator_fr-als.

If you mention this project in a publication, please include the citations below.

• Adam, J.M.-C. (2025). Translator model from French to Alsatian. Master's project at DSTI, France. https://github.com/JoanneAB/Translator_fr-als

Licence

This work is under a GNU General Public License.