ol-openedx-course-translations 0.5.1

An Open edX plugin to translate courses

OL Open edX Course Translations

An Open edX plugin to manage course translations.

Purpose

Translate course content into multiple languages to enhance accessibility for a global audience.

Setup

For detailed installation instructions, please refer to the plugin installation guide.

Installation required in:

• Studio (CMS)

• LMS

Configuration

• Add the following configuration values to the config file in Open edX. For any release after Juniper, that config file is /edx/etc/lms.yml and /edx/etc/cms.yml. If you’re using private.py, add these values to lms/envs/private.py and cms/envs/private.py. These should be added to the top level. Ask a fellow developer for these values.

  # Output directory for translated courses
  # Default: /openedx/data/course_translations/
  COURSE_TRANSLATIONS_BASE_DIR: "/openedx/data/course_translations/"
  
  # Translation providers configuration
  TRANSLATIONS_PROVIDERS: {
      "default_provider": "mistral",  # Default provider to use
      "deepl": {
          "api_key": "<YOUR_DEEPL_API_KEY>",
      },
      "openai": {
          "api_key": "<YOUR_OPENAI_API_KEY>",
          "default_model": "gpt-5.2",
      },
      "gemini": {
          "api_key": "<YOUR_GEMINI_API_KEY>",
          "default_model": "gemini-3-pro-preview",
      },
      "mistral": {
          "api_key": "<YOUR_MISTRAL_API_KEY>",
          "default_model": "mistral-large-latest",
      },
  }
  TRANSLATIONS_GITHUB_TOKEN: <YOUR_GITHUB_TOKEN>
  TRANSLATIONS_REPO_PATH: ""
  TRANSLATIONS_REPO_URL: "https://github.com/mitodl/mitxonline-translations.git"
  LITE_LLM_REQUEST_TIMEOUT: 300  # Timeout for LLM API requests in seconds

• For Tutor installations, these values can also be managed through a custom Tutor plugin.

Translation Providers

The plugin supports multiple translation providers:

• DeepL

• OpenAI (GPT models)

• Gemini (Google)

• Mistral

Configuration

All providers are configured through the TRANSLATIONS_PROVIDERS dictionary in your settings:

TRANSLATIONS_PROVIDERS = {
    "default_provider": "mistral",  # Optional: default provider for commands
    "deepl": {
        "api_key": "<YOUR_DEEPL_API_KEY>",
    },
    "openai": {
        "api_key": "<YOUR_OPENAI_API_KEY>",
        "default_model": "gpt-5.2",  # Optional: used when model not specified
    },
    "gemini": {
        "api_key": "<YOUR_GEMINI_API_KEY>",
        "default_model": "gemini-3-pro-preview",
    },
    "mistral": {
        "api_key": "<YOUR_MISTRAL_API_KEY>",
        "default_model": "mistral-large-latest",
    },
}

Important Notes:

1. DeepL Configuration: DeepL must be configured in TRANSLATIONS_PROVIDERS['deepl']['api_key'].

2. DeepL for Subtitle Repair: DeepL is used as a fallback repair mechanism for subtitle translations when LLM providers fail validation. Even if you use LLM providers for primary translation, you should configure DeepL to enable automatic repair.

3. Default Models: The default_model in each provider’s configuration is used when you specify a provider without a model (e.g., openai instead of openai/gpt-5.2).

Provider Selection

You can specify providers in three ways:

1. Provider only (uses default model from settings):

./manage.py cms translate_course \
    --target-language ar \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider openai \
    --srt-translation-provider gemini

2. Provider with specific model:

./manage.py cms translate_course \
    --target-language ar \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider openai/gpt-5.2 \
    --srt-translation-provider gemini/gemini-3-pro-preview

3. DeepL (no model needed):

./manage.py cms translate_course \
    --target-language ar \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider deepl \
    --srt-translation-provider deepl

Note: If you specify a provider without a model (e.g., openai instead of openai/gpt-5.2), the system will use the default_model configured in TRANSLATIONS_PROVIDERS for that provider.

Translating a Course

1. Open the course in Studio.

2. Go to Tools -> Export Course.

3. Export the course as a .tar.gz file.

4. Go to the CMS shell

5. Run the management command to translate the course:

   ./manage.py cms translate_course \
       --source-language en \
       --target-language ar \
       --course-dir /path/to/course.tar.gz \
       --content-translation-provider openai \
       --srt-translation-provider gemini \
       --translation-validation-provider openai/gpt-5.2 \
       --content-glossary /path/to/content/glossary \
       --srt-glossary /path/to/srt/glossary

Command Options:

• --source-language: Source language code (default: en)

• --target-language: Target language code (required)

• --course-dir: Path to exported course tar.gz file (required)

• --content-translation-provider: Translation provider for content (XML/HTML and text) (required).

  Format:

  • deepl - uses DeepL (no model needed)

  • PROVIDER - uses provider with default model from settings (e.g., openai, gemini, mistral)

  • PROVIDER/MODEL - uses provider with specific model (e.g., openai/gpt-5.2, gemini/gemini-3-pro-preview, mistral/mistral-large-latest)

• --srt-translation-provider: Translation provider for SRT subtitles (required). Same format as --content-translation-provider

• --translation-validation-provider: Optional provider to validate/fix XML/HTML translations after translation.

• --content-glossary: Path to glossary directory for content (XML/HTML and text) translation (optional)

• --srt-glossary: Path to glossary directory for SRT subtitle translation (optional)

Examples:

# Use DeepL for both content and subtitles
./manage.py cms translate_course \
    --target-language ar \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider deepl \
    --srt-translation-provider deepl

# Use OpenAI and Gemini with default models from settings
./manage.py cms translate_course \
    --target-language fr \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider openai \
    --srt-translation-provider gemini

# Use OpenAI with specific model for content, Gemini with default for subtitles
./manage.py cms translate_course \
    --target-language fr \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider openai/gpt-5.2 \
    --srt-translation-provider gemini

# Use Mistral with specific model and separate glossaries for content and SRT
./manage.py cms translate_course \
    --target-language es \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider mistral/mistral-large-latest \
    --srt-translation-provider mistral/mistral-large-latest \
    --content-glossary /path/to/content/glossary \
    --srt-glossary /path/to/srt/glossary

# Use different glossaries for content vs subtitles
./manage.py cms translate_course \
    --target-language es \
    --course-dir /path/to/course.tar.gz \
    --content-translation-provider openai \
    --srt-translation-provider gemini \
    --content-glossary /path/to/technical/glossary \
    --srt-glossary /path/to/conversational/glossary

Glossary Support:

You can use separate glossaries for content and subtitle translation. This allows you to apply different terminology choices based on context:

• Content glossary (--content-glossary): Used for XML/HTML content, policy files, and text-based course materials. Typically contains more formal or technical terminology.

• SRT glossary (--srt-glossary): Used for subtitle translation. Can contain more conversational or context-specific terms appropriate for spoken content.

Create language-specific glossary files in each glossary directory:

# Content glossary structure
glossaries/technical/
├── ar.txt  # Arabic glossary
├── fr.txt  # French glossary
└── es.txt  # Spanish glossary

# SRT glossary structure
glossaries/conversational/
├── ar.txt  # Arabic glossary
├── fr.txt  # French glossary
└── es.txt  # Spanish glossary

Format: One term per line as “source_term : translated_term”

# es HINTS
## TERM MAPPINGS
These are preferred terminology choices for this language. Use them whenever they sound natural; adapt freely if context requires.

- 'accuracy' : 'exactitud'
- 'activation function' : 'función de activación'
- 'artificial intelligence' : 'inteligencia artificial'
- 'AUC' : 'AUC'

Note: Both glossary arguments are optional. If not provided, translation will proceed without glossary terms. You can provide one, both, or neither glossary as needed.

Subtitle Translation and Validation

The course translation system includes robust subtitle (SRT) translation with automatic validation and retry mechanisms to ensure high-quality translations with preserved timing information.

Translation Process

The subtitle translation follows a multi-stage process with built-in quality checks:

1. Initial Translation: Subtitles are translated using your configured provider (DeepL or LLM)

2. Validation: Timestamps, subtitle count, and content are validated to ensure integrity

3. Automatic Retry: If validation fails, the system automatically retries translation (up to 1 additional attempt)

4. Task Failure: If all retries fail validation, the translation task fails to prevent corrupted subtitle files

Validation Rules

The system validates subtitle translations against these criteria:

• Subtitle Count: Translated file must have the same number of subtitle blocks as the original

• Index Matching: Each subtitle block index must match the original (e.g., if original has blocks 1-100, translation must have blocks 1-100 in the same order)

• Timestamp Preservation: Start and end times for each subtitle block must remain unchanged

• Content Validation: Non-empty original subtitles must have non-empty translations (blank translations are flagged as errors)

Example Validation Process:

1. Initial Translation (using OpenAI):
   ✓ 150 subtitle blocks translated
   ✗ Validation failed: 3 blocks have mismatched timestamps

2. Retry Attempt:
   ✓ 150 subtitle blocks translated
   ✗ Validation failed: 2 blocks still have issues

3. Task Failure:
   ❌ Translation failed after all retries
   ❌ Task aborted to prevent corrupted subtitle files

Failure Handling

If subtitle translation fails after all attempts:

• The translation task will fail with a ValueError

• The entire course translation will be aborted to prevent incomplete translations

• The translated course directory will be automatically cleaned up

• An error message will indicate which subtitle file caused the failure

• No partial or corrupted translation files will be left behind

Generating static content translations

This command synchronizes translation keys from edx-platform and MFE’s, translates empty keys using LLM, and automatically creates a pull request in the translations repository.

What it does:

1. Syncs translation keys from edx-platform and MFE’s to the translations repository

2. Extracts empty translation keys that need translation

3. Translates empty keys using the specified LLM provider and model

4. Applies translations to JSON and PO files

5. Commits changes to a new branch

6. Creates a pull request with translation statistics

Usage:

1. Go to the CMS shell

2. Run the management command:

   ./manage.py cms sync_and_translate_language <LANGUAGE_CODE> [OPTIONS]

Required arguments:

• LANGUAGE_CODE: Language code (e.g., el, fr, es_ES)

Optional arguments:

• --iso-code: ISO code for JSON files (default: same as language code)

• --provider: Translation provider (openai, gemini, mistral). Default is taken from TRANSLATIONS_PROVIDERS['default_provider'] setting

• --model: LLM model name. If not specified, uses the default_model for the selected provider from TRANSLATIONS_PROVIDERS. Examples: gpt-5.2, gemini-3-pro-preview, mistral-large-latest

• --repo-path: Path to mitxonline-translations repository (can also be set via TRANSLATIONS_REPO_PATH setting or environment variable)

• --repo-url: GitHub repository URL (default: https://github.com/mitodl/mitxonline-translations.git, can also be set via TRANSLATIONS_REPO_URL setting or environment variable)

• --glossary: Path to glossary directory (optional). Should contain language-specific files (e.g. {iso_code}.txt).

• --batch-size: Number of keys to translate per API request (default: 200, recommended: 200-300 for most models)

• --mfe: Filter by specific MFE(s). Use edx-platform for backend translations

• --dry-run: Run without committing or creating PR

Examples:

# Use default provider (from TRANSLATIONS_PROVIDERS['default_provider']) with its default model
./manage.py cms sync_and_translate_language el

# Use OpenAI provider with its default model (gpt-5.2)
./manage.py cms sync_and_translate_language el --provider openai

# Use OpenAI provider with a specific model
./manage.py cms sync_and_translate_language el --provider openai --model gpt-5.2

# Use Mistral provider with a specific model and glossary
./manage.py cms sync_and_translate_language el --provider mistral --model mistral-large-latest --glossary /path/to/glossary --batch-size 250

License

The code in this repository is licensed under the AGPL 3.0 unless otherwise noted.

Please see LICENSE.txt for details.