Skip to content

🧠 Semantic Communication Encoding · DRAFT v1

Semantic Communication Encoding (SCE)

🧠 A small, governed emoji ontology for humans, tools, and LLMs.

Semantic Communication Encoding (SCE) is a way to use Unicode emoji as semantic operators — a compact, standardized “mini-language” that both humans and machines can understand.

Instead of long, fragile instructions like:

“This is a non-negotiable fact that must not be contradicted…”

you can write:

📌 Student was first enrolled on 2024-11-06.

…and everyone knows that:

  • 📌 means pinned, non-negotiable fact
  • it belongs to the structure domain
  • its behavior and conflicts are defined in a formal ontology

SCE is designed for:

  • 🧭 LLM prompts & tools – structure, control flow, and state
  • ⚖️ Compliance & legal workflows – Title IX, FERPA, data access, audit trails
  • 🧑‍💻 Agent frameworks – shared symbolic “vocabulary” across tools
  • 📝 Human collaboration – readable, skimmable, explainable annotations

Why SCE?

Modern systems mix:

  • Narrative context
  • Legal / policy constraints
  • Workflow & task state
  • Roles and actors
  • Privacy & access rules

SCE gives you a small, opinionated, governed vocabulary for those concepts.

The key benefit? Replace verbose natural language with precise semantic symbols:

Before: "This is a non-negotiable fact that must remain true..." (28 tokens)
After: 📌 Student was injured on 11/06/24 (2 tokens for the symbol)

Result: ~93% token reduction while increasing semantic precision.

See the Prompting Guide for detailed examples and token savings analysis.

Domain Example symbols Purpose
structure 🗂️ 📌 📎 Sections, pinned facts, references
legalPolicy ⚖️ 📜 🧾 🏛️ Law, citation, evidence, authority
reasoning 🔍 🧠 🕵️ Analysis, insight, investigation
tasks 📝 ☐ ☑️ ✅ 🔁 Actions, task state
privacy 🔐 🗝️ 🔓 Restricted / authorized / open
actors 👤 🧑‍🎓 🧑‍🏫 🧑‍⚖️ 🏢 People & institutions
state ⏳ ❓ ⚠️ ❌ Pending, unknown, warning, prohibited
control 🔀 ⏭️ ⏮️ Decision points, next, back

Each symbol has a machine-readable definition with:

  • emoji – the actual Unicode grapheme
  • role – category / domain role
  • meaning – authoritative definition
  • allowedContext – HUMAN / LLM / TOOL
  • usage – REQUIRED / OPTIONAL / CONDITIONAL
  • conflictsWith – mutually exclusive symbols
  • example – canonical example usage

What’s in this repo?

SCE ships with a complete “stack”:

  • 🧬 Core ontology – ontology.ts is the normative symbol set
  • 🔎 Interpreter – emoji → JSON symbol definitions
  • ✅ Validator – checks for conflicts, duplicates, structural issues
  • 🧰 TypeScript API – drop-in library for your own tools
  • 🖥️ CLI – inspect symbols and validate the ontology from the terminal
  • 🔌 MCP server – stdio Model Context Protocol server exposing SCE tools

Quick start

1. Install

# once published
npm install semanticencoding

# or from a cloned repo
npm install .

2. Interpret symbols in text

import { getDefinitionsFromText } from "semanticencoding";

const text =
  "📌 First report was on 2024-11-06. ⏳ Investigation is still pending. ⚠️ Parent has raised safety concerns.";

const defs = getDefinitionsFromText(text);

console.log(defs[0]);
/*
{
  emoji: "📌",
  role: "STRUCTURE",
  meaning: "Pinned fact or non-negotiable constraint",
  allowedContext: ["HUMAN", "LLM"],
  usage: "REQUIRED",
  conflictsWith: ["📝", "🧠"],
  example: "📌 Timeline is determined by first reported contact (11/06/24)."
}
*/

3. Validate the ontology

import { validateOntology } from "semanticencoding";

const issues = validateOntology();
if (issues.length > 0) {
  console.error("Ontology has issues:", issues);
  process.exit(1);
}

MCP integration

SCE includes a stdio MCP server with three tools:

  • sce_explain – extract & explain SCE symbols from text
  • sce_validate_ontology – validate ontology structure & conflicts
  • sce_suggest_symbols – suggest SCE symbols for freeform text

Run the server

npm run build
node dist/sce-mcp-server.js
# or, if mapped in package.json:
sce-mcp

Each tool supports a structured format override:

{
  "format": {
    "type": "auto | pretty | json | hybrid",
    "color": true
  }
}
  • auto (default) – TTY → pretty, non-TTY → JSON
  • pretty – human output (optionally colored)
  • json – machine-first, strict JSON text
  • hybrid – pretty summary + JSON block

Learn more

Status

🏷️ SCE · v0.1: Draft

SCE is stable enough for:

  • real prompt engineering
  • compliance / legal workflows
  • MCP / agent framework integrations

…but will continue to evolve based on practical usage and contributor feedback.