Regeneration Rules

This document covers content that is regenerated from scratch in the target language during course localization — as opposed to content that is translated or adapted (see Translation rules).

Why regenerate instead of translate

Some content formats produce better results when generated fresh from the blueprint in the target language, rather than translated from an existing source. This is the case when the output is tightly coupled to language-specific phrasing, timing, or assessment logic.

What is regenerated

Video

Video content is generated from scratch in the target language instead of being translated.

The video pipeline takes the same blueprint (section type, format, instructional intent) and produces a new video with:

voiceover scripts written natively in the target language;
slide content adapted to the target language;
the same template structure and visual layout.

This avoids the awkwardness of dubbed or subtitle-translated educational video, and produces content that sounds native to the target audience.

See docs/video/video.md for video pipeline details.

Questions

Section-level questions (ChoiceQuestions, FillTheBlanks, QuestionWithFeedback) are generated from the blueprint in the target language rather than translated from existing questions.

This ensures:

question phrasing is natural in the target language;
answer options and distractors make sense culturally and linguistically;
terminology follows the localization glossary.

See docs/question/questions.md for question format details.

What this means in practice

When localizing a course:

The outline and blueprints stay the same across all languages.
Text, images, and artifacts are translated (with localization quality rules).
Video and questions are regenerated from the same blueprints.
Publishing assets (description, objectives) are rewritten in the target language.

This is not a hard rule — it reflects the current system behavior. Future iterations may change the boundary between translation and regeneration.