README_EN.md

# Trans-PolyDocs

Quickly translate Markdown documents or PDF documents (converted by Doc2X) into Markdown/Word while retaining the original format of formulas/tables/images.

Supports multiple translators and multi-threaded translation:

• DeepSeek (default translator)
• OpenAI (supports custom URL, must end with /v1)
• Ollama
• Google Translate (experimental, may be unstable, provided by py-googletrans)
• DeepL (official API)
• DeeLX

Main Interface	LLM Settings	Multiple Translators
Drag or click to import Markdown/PDF files, supports automatic dark mode switching	Detailed configuration for LLM, offering more customization	Supports multiple translators

## Translating using LLM

If you wish to use LLM for translation, there are additional settings to aid in the translation process.

### Text Extraction Method for Translation

Due to the uncertainty of LLM output, three methods are built-in for extracting translated text: json, markdown, and direct. The preset prompt output requires using the markdown method for extraction.

• json method extracts content with the key "translated" from JSON format.
• markdown method extracts text enclosed in ``` (snips documents between the first and last ```).
• direct method uses the raw response text directly.

### Prompts

You can use some variables in the input prompt to fill in. In the GUI, you can click the corresponding button to copy the variable name. Supported variables include:

• {{text}} Text to be translated (REQUIRED)
• {{prev_text}} Previous text context, if none, then an empty string
• {{next_text}} Next text context, if none, then an empty string
• {{dest}} Target language for translation

For example, here is a sample input prompt using variables:

Translate the following text to {{dest}}:
{{text}}

You do not need to emphasize retaining formula structure in the prompt; the program will replace formulas with emoji and restore them post-translation. This ensures that any translator (such as DeepL) can retain formula translation.

## Running the GUI

Important

If you wish to output the translated document as a Word file, please install pandoc before running the program.

Windows:

Download installer or enter winget install --source winget --exact --id JohnMacFarlane.Pandoc in Powershell

MacOS: Run brew install pandoc in Terminal

Ubuntu/Debian: Run sudo apt install pandoc in Terminal

Arch/Manjaro: Run sudo pacman -S pandoc-cli in Terminal

### Precompiled Program

You can click on releases on the right to download the precompiled program, download the version corresponding to your operating system, unzip the package, and run the program.

### Adjusting Output Word Styles

Adjust the styles in the reference.docx located at the root directory after extraction.

### Running from Source

After cloning the current repository, execute it from the repository path:

conda create -n translate python=3.12
conda activate translate
pip install uv
uv pip install -r requirements.txt
python app.py

## CLI Program

If you wish to use the CLI program, after cloning the current repository, execute the following in the repository path to copy sample environment variables:

cp example.env .env

Then modify it according to the instructions in .env, configure the environment, and run:

conda create -n translate python=3.12
conda activate translate
pip install uv
uv pip install -r requirements.txt
python Main.py

## Custom Translator

If you want to use your own translation API, you can customize a translator. A sample translator is as follows:

def translate(text: str, prev_text: str, next_text: str) -> str:
    try:
        return "This is an example!"
    except Exception as e:
        print(f"Error: {e}")
        return text

Then import MD_Translate.py in your program for use:

# Your defined translate function
from MD_Translate import Process_MD
file_path = "path"  # Path to the MD file
threads = 10  # Number of translation threads
Process_MD(md_file=file_path, translate=translator, thread=threads)

graph TD
A[Input MD or PDF file] --> B{File Type}
B -->|MD File| C[Parse MD file]
B -->|PDF File| D[Use Doc2X to convert to MD file]
D --> C
C --> E[Segment based on type]
E --> F[Call translator for translation]
F --> G{Retain formula translation successful?}
G -->|Yes| H[Retain formula translation]
G -->|No| I[Rollback to plain text translation]
H --> J[Combine translated document]
I --> J
J --> K[Use pandoc to convert to Word]

Loading

## Packaging

Use pyinstaller to package. Install with pip install pyinstaller. Run the following command:

pyinstaller -w --onefile -i icon.png app.py

And copy the reference.docx and example.env from the project to the same directory as the generated binary file.