Open Translate

Welcome to Open-Translate, where innovation meets seamless communication. Our project serves as a user-friendly interface for machine translation, complemented by a robust Restful API implementation.

Table of Contents

• Open Translate
  • Table of Contents
  • Frontend
    • Installation and Setup
    • Customization
      • Adding Languages
      • API Configuration
  • Backend
    • Setup and Run
    • API Payload
    • Language Capabilities Enhancement
    • User-Submitted Translations

Frontend

The provided code is intentionally generic, allowing you to seamlessly integrate the frontend with your unique project requirements.

Installation and Setup

git clone https://github.com/danielinux7/open-translate.git
cd open-translate
npm install
ng serve

Customization

To tailor the frontend to your specific requirements, consider the following customization options:

Adding Languages

Expand the language options by adding a new language to the src/app/languages.ts file. Simply include it in the languages object list.

API Configuration

Adjust API calls to suit your needs by modifying the relevant sections in the src/app/translate.service.ts file. For local testing of the API server, set the endpoint to localhost.

Backend

The backend comprises two servers: a Flask server responsible for handling RESTful API calls, and a TensorFlow server dedicated to managing Neural Machine Translation (NMT) models. When a call is initiated, the Flask server manages the requests and seamlessly passes them to the TensorFlow server.

Setup and Run

git clone https://github.com/danielinux7/open-translate.git
cd open-translate/server
bash setup_prod.sh # use setup.sh for localhost
bash run_prod.sh # use run.sh for localhost

You should configure your production server name in setup_prod.sh and run_prod.sh. You can stop the servers by running bash stop.sh.

API Payload

The payload for API calls includes the following key components:

• langSrc: Source language identifier.
• langTgt: Target language identifier.
• source: Representing the source content.

The response includes a target key-value pair, providing the translated content.

Language Capabilities Enhancement

Expanding the server's language capabilities is a straightforward process. To integrate your TensorFlow saved model:

• Place the model under the /models/<langSrc-langTgt>/<Version> folder.
  • Example: /models/en-fr/1/ for English to French translation, version 1.
• If you introduce a new version, TensorFlow serves the latest version.

For tokenization, the backend uses Sentencepiece. Include the src.model and tgt.model models in the /models/<langSrc-langTgt>/<Version>/assets folder.

You can unpuck server/models.zip to get the available models running.

User-Submitted Translations

The /starred folder serves as the destination for user-submitted edited or high-quality translations via the frontend.