Controlling LLM Prose

Area VI · Controlled Natural Languages & Constrained Registers

Structured Topic Typing (DITA)

Concept · Task · Reference

Force each unit of content into one topic type instead of mixing modes.

Definition & mechanics

Topic-based authoring, as in DITA, sorts content into distinct types—concept (what it is and why), task (numbered steps to do something), and reference (lookup facts, tables, parameters). Keeping each topic to a single type stops explanations from tangling with procedures and makes content reusable and scannable.

A curiosity

Most confusing docs are confusing because they blend types—a tutorial that stops mid-step to philosophize. Separate the why from the how from the what, and each becomes easy.

Example

Before (mixed)Caching improves performance because repeated lookups are expensive, so click Settings, and note that the default TTL is 300 seconds, then enable the cache.
After (task topic)To enable caching: 1. Open Settings. 2. Turn on Cache. (Concept and the 300-second default TTL move to separate Concept and Reference topics.)

Copy-ready prompt

Copyable Prompt
Apply the concept of Structured Topic Typing (restructure the content into discrete topic types—separate conceptual explanation (what and why), task procedures (numbered, imperative steps), and reference material (lookup facts and tables); keep each resulting section to a single topic type and do not blend explanation into procedures) to the provided text and rewrite in place.

Apply the edits directly to the file as a clean, production-ready deliverable. Output ONLY the revised document text without embedding revision notes, changelogs, or meta-commentary in the file itself—put any explanations in your chat response only.