I Built a Meaning Engine Without Neural Networks

2026-07-15 · Source: Naturallanguageprocessing on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

A novel "meaning engine" has been developed that learns semantic relationships without relying on neural networks, transformers, or gradient descent. Instead, it builds an inspectable graph of word co-occurrence patterns, weighted by Pointwise Mutual Information (PMI), using a 2-bit ternary {-1, 0, +1} connection system. Trained on dictionaries and 65 public domain books, the system successfully learned word meanings, discovered 247 suffixes and 38 prefixes in Turkish morphology, and established cross-language bridges between English and Turkish with an average Jaccard similarity of 0.47 and a cosine bridge of 0.61 for 10 word pairs. Crucially, it outputs 0 when evidence is insufficient, indicating "I don't know" rather than hallucinating. This experimental system, which processed 1,234,706 sentences and identified 288,407 concepts with ~1.3 GB peak RAM, challenges the dominant "bigger models" AI narrative.

Key takeaway

For AI Architects or Machine Learning Engineers exploring alternatives to large language models for semantic understanding, this "meaning engine" offers a compelling architectural shift. You should consider its graph-based, non-neural approach for applications requiring inspectable relationships, explicit uncertainty handling, or low resource footprints. This system demonstrates that complex linguistic insights, including cross-language bridges and morphology, can emerge from simple co-occurrence patterns, challenging the assumption that scale is the sole path to advanced AI capabilities.

Key insights

A non-neural network "meaning engine" learns semantics via co-occurrence graphs, offering inspectable relationships and explicit uncertainty.

Principles

• Meaning can emerge from simple word co-occurrence patterns.
• Explicit uncertainty (e.g., "I don't know") prevents hallucination.
• Smaller, architecturally different systems can achieve complex learning.

Method

The system builds a navigable graph by observing word co-occurrences, weighting connections with Pointwise Mutual Information (PMI), and using a 2-bit ternary {-1, 0, +1} state for relationships.

In practice

• Train a semantic memory from scratch with custom data.
• Explore morphology discovery in agglutinative languages.
• Implement explicit "I don't know" states in knowledge systems.

Topics

• Meaning Engine
• Semantic Networks
• Pointwise Mutual Information
• Co-occurrence Graphs
• Morphology Discovery
• Cross-language Bridges

Code references

• arifkurnaz/ternary-semantic-brain-demo

Best for: NLP Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, AI Architect

Related on AIssential

• The Evolution of Meaning: NLP’s Long March from Syntax to Semantics — Meaning is a dynamic process requiring active computation and external anchoring, not an inherent property of a word.
• Conceptual Networks for Cross-Linguistic Idiomatic Expressions:A Feature-Based Graph Approach — Idioms organize by conceptual schema, not language, offering interpretable, cross-linguistically stable meaning representations.
• Geometry of Semantic Space: Comparative Study of Discrete and Continuous Models — Graph-based semantic models offer more interpretable and structured meaning representations than high-dimensional transformer embeddings.
• Can AI Truly Understand Human Language? — AI models recognize language patterns and approximate meaning, but lack true human-like understanding.
• Compositionality and the lexicon in evolutionary semantics — The framework integrates formal semantics with evolutionary modeling to explain semantic universals like conservativity.
• Language Models Learn Constructional Semantics, Not To Mention Syntax: Investigating LM Understanding of Paired-Focus Constructions — Modestly sized open-source LMs can learn rare Paired-Focus construction semantics, with understanding emerging later than syntax and correlating with world knowledge gains.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Naturallanguageprocessing on Medium.