Translation Rules

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

What is translated

Section text and artifacts

Section text content and interactive artifacts are translated from the source language into the target language. The blueprint (section type, format, and instructional intent) remains the same — only the language of the output changes.

Translation follows the localization glossary and language quality rules defined in Localization principles.

For artifact payloads specifically, the localization step preserves the exact JSON structure and translates only learner-facing values inside that JSON. It does not redesign the interaction, reinterpret the checking logic, or fact-check the content.

Images with text

Images that contain text are localized:

preserve meaning, not literal wording;
keep the translated text grammatically correct;
allow minor layout changes caused by text length;
keep brand names, acronyms, and widely accepted English industry terms when appropriate.

The image localization pipeline uses a translation prompt followed by a binary quality checker. See docs/image/images.md for pipeline details.

Representation assets (publishing assets)

The following publishing assets are regenerated in the target language using the source content plus the localization glossary:

Course description — via flow_asset_localize_description
Learning objectives — via flow_asset_localize_learning_objectives

These are not literal translations. They are rewritten for the target language so the result reads naturally and stays terminologically consistent.

Localization asset itself

The localization asset (flow_asset_localization) produces:

localized course, module, topic, and section titles;
a terminology glossary for the target language.

This asset is the baseline that all other translation operations consume.