Software Overview

Software Overview#

This project is a frontend-first translation system designed for Qt Linguist .ts workflows. It integrates Large Language Models (LLMs) while keeping the format safe, the process controllable, and the output reliably compatible with the original project.

It supports two translation approaches:

Online GPT API
Translate directly in the browser via an OpenAI-compatible API.
Local LLM
Generate an executable Jupyter Notebook and run translation offline on the user’s machine (CPU / GPU).

Both approaches provide glossary guidance, format + placeholder protection, batch processing with progress tracking, and a live source/translation comparison to help convert strings in projects like QGIS into Traditional Chinese.

Which option should I choose?#

Option	How it works	Best for	Highlights & limits
Online GPT API	Calls your configured API in the browser	Quick setup, translate directly on the web	Built-in progress + comparison table, one-click `.ts` export
Local LLM	Generates a configurable `.ipynb` and runs locally	Offline use, your own GPU, testing open models	Highly customizable, but you must set up the environment

Quick Start#

A. Online GPT API (recommended to try first)#

Open the ChatGPT API Translation page (app_api).
Set API Key, Base URL, and Model.
Upload the .ts file to translate.
(Optional) Upload a Glossary (CSV / ODS, columns: en, zh or English Term, Chinese Term).
Set Batch and the maximum number of entries for a small test run.
Run translation and review the live source/translation table and progress.
After verifying format safety, download qgis_zh-Hant.ts.

B. Local LLM (offline)#

Open the Local LLM page (app_local).
Configure model settings, batch size, token limits, glossary path, etc.
Download the generated .ipynb.
Install dependencies and run the notebook in your local Jupyter environment.

Core Features#

Translation & terminology consistency#

Batch-based translation to keep input/output aligned.
Glossary + LCS matching to detect key terms in the source text and nudge the model to use fixed translations
(e.g., raster → 網格) to reduce terminology drift.

Format & structure protection#

Automatically protects and restores:
- HTML tags and entities
- Placeholders (%n, %1, {0}, etc.)
- Original .ts structure and DOCTYPE
Correct handling for numerus="yes" and <numerusform> output.

Batch processing & review#

Configure batch size and an entry limit.
Live progress bar and source/translation comparison table for quick spot checks.

Export & integration#

Exports .ts via browser download (no backend server required).
Works with .qm compilation and later merge/version workflows.

Pre-export validation (what the system guarantees)#

Before exporting the .ts file, the system checks:

XML is parseable
<numerusform> entries are complete for numerus="yes"
Placeholders are not missing or mismatched in count
HTML tags are not broken or unclosed

If validation fails, that entry will not be written back, preventing broken .ts outputs.

Note

This project runs in the browser via Pyodide and does not require a server or a local Python runtime.