Gtbabel extracts on every page load any page into logical paragraph tokens. Static and dynamic content is deliberately treated the same. All tokens are replaced (if available) by it’s translation before rendered. The tokens get dumped (if not available) into gettext, where they can be translated.