Lute AI Translation Panel

This project lets you launch Lute (Learning Using Texts) with an intuitive translation panel powered by local AI models.

Below you can find:

1. How to launch the app
2. How to configure the panel in Lute
3. Motivation
4. Credits

1. How to launch the app

1.1 Configure environment variables

Create .env from .env.example and set the model you want by updating the values of MT_MODEL, SOURCE_LANGUAGE_DEFAULT and TARGET_LANGUAGE_DEFAULT:

PORT=8000
HOST=0.0.0.0
MT_MODEL=Helsinki-NLP/opus-mt-fi-en
MODEL_CACHE_DIR=/model_cache
SOURCE_LANGUAGE_DEFAULT=Finnish
TARGET_LANGUAGE_DEFAULT=English
MAX_INPUT_CHARS=1000
LOG_LEVEL=INFO

Example model values:

Translation	MT_MODEL	SOURCE_LANGUAGE_DEFAULT	TARGET_LANGUAGE_DEFAULT
Finnish to English	Helsinki-NLP/opus-mt-fi-en	Finnish	English
English to Finnish	Helsinki-NLP/opus-mt-en-fi	English	Finnish
Spanish to English	Helsinki-NLP/opus-mt-es-en	Spanish	English
English to Spanish	Helsinki-NLP/opus-mt-en-es	English	Spanish
French to English	Helsinki-NLP/opus-mt-fr-en	French	English
English to French	Helsinki-NLP/opus-mt-en-fr	English	French
Italian to English	Helsinki-NLP/opus-mt-it-en	Italian	English
English to Italian	Helsinki-NLP/opus-mt-en-it	English	Italian

1.2 Launch Docker containers

Launch Docker containers with Docker Compose:

$ docker compose up -d --build

Note: The model is prefetched during the Docker image build. For this reason, to switch models, update .env, rebuild and recreate the service:

$ docker compose up -d --build --force-recreate

Lute will be available at localhost:5001 and the translation panel at localhost:{PORT} (e.g. localhost:8000).

Try visiting localhost:8000/translate?payload=Tervetuloa. Tämä on Lute AI panel and changing the value of payload for a quick demo.

2. How to configure the panel in Lute

• Open language settings from the header menu

• Select your language

• Add a dictionary entry by clicking the + icon and set the following values: Terms, Embedded and http://localhost:8000/translate?payload=[LUTE] (replace 8000 with your port if you changed it)
• Move the entry to the desired position by dragging the icon on the left side of the entry
• Save

3. Motivation

I like studying languages and I am currently learning Finnish. In addition to traditional learning methods, I discovered the Lute project and immediately liked the idea behind it.

I wanted to try integrating an AI translation panel into Lute for a few reasons:

• AI translations can sometimes be more accurate than regular online translators
• I wanted to experiment with local AI models that can run on a CPU
• I was curious to see what could be built
• I hope to inspire other developers to contribute to Lute and create useful plugins

4. Credits

• Lute for the amazing project: https://github.com/LuteOrg/lute-v3
• Helsinki-NLP and the OPUS-MT project for the amazing models: https://github.com/Helsinki-NLP/Opus-MT
• OPUS-MT reference papers:

@article{tiedemann2023democratizing,
  title={Democratizing neural machine translation with {OPUS-MT}},
  author={Tiedemann, J{\"o}rg and Aulamo, Mikko and Bakshandaeva, Daria and Boggia, Michele and Gr{\"o}nroos, Stig-Arne and Nieminen, Tommi and Raganato\
, Alessandro and Scherrer, Yves and Vazquez, Raul and Virpioja, Sami},
  journal={Language Resources and Evaluation},
  number={58},
  pages={713--755},
  year={2023},
  publisher={Springer Nature},
  issn={1574-0218},
  doi={10.1007/s10579-023-09704-w}
}

@InProceedings{TiedemannThottingal:EAMT2020,
  author = {J{\"o}rg Tiedemann and Santhosh Thottingal},
  title = {{OPUS-MT} — {B}uilding open translation services for the {W}orld},
  booktitle = {Proceedings of the 22nd Annual Conferenec of the European Association for Machine Translation (EAMT)},
  year = {2020},
  address = {Lisbon, Portugal}
 }