training/dataset_generation_guide.md

Dataset Generation Guide

This guide explains how to generate new training examples for the Glyphic Language.

---

1. Dictionary‑Driven Generation

All examples must be derived from:

• dictionary entries
• syntax rules
• BNF grammar

No freeform glyph usage is allowed.

---

2. Example Types

2.1 Atomic Examples

Single glyph → meaning
Meaning → single glyph

2.2 Scene Examples

Full sequences with:

• actor
• action
• object
• modifiers
• context

2.3 Negative Examples

Invalid sequences for syntax training.

2.4 Symbolic Examples

Mythic, emotional, sensory, or social scenes.

---

3. Generation Process

1. Select glyph(s) from dictionary
2. Build a valid sequence using syntax rules
3. Generate structured meaning
4. Generate natural language description
5. Add to appropriate dataset file

---

4. Quality Requirements

• No ambiguity
• No hallucinated glyphs
• No missing roles
• No invalid ordering
• No duplicate examples