Structural Semantics Constraint Geometry, Scope Calculus, and Resonant Models of Representation
Flyxion
A Research Monograph
Version 1.0
August 1, 2026
ii
Copyright ยฉ 2026 Flyxion All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form without permission, except for brief quotations used in scholarly review. This work develops original mathematical constructions, formal definitions, and computational models. Where established results are discussed, appropriate references are provided.
iv
โThe aim of science is not things themselves, as the dogmatists in their simplicity imagine, but the relations among things.โ
โ Henri Poincarรฉ
vi
Contents
Preface xiii
Methodological Statement xv
How to Read This Monograph xvii
Notation and Conventions xix
I Epistemic and Historical Boundaries 1
1 Epistemic Framing, Disentanglement, and Primary Research Questions 3 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2 Motivating the Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Epistemic Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.4 Formal Foundations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.4.1 Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.4.2 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.4.3 Motivation for Scope Geometry . . . . . . . . . . . . . . . . . . . . 6 1.5 The Disentanglement Operator . . . . . . . . . . . . . . . . . . . . . . . . 6 1.5.1 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.6 Boundary Constraints and Bounded Contexts . . . . . . . . . . . . . . . . 7 1.6.1 Bounded contexts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.6.2 Motivation for scope geometry . . . . . . . . . . . . . . . . . . . . 8 1.6.3 A preview of Structural Semantics . . . . . . . . . . . . . . . . . . 8 1.7 The Research Program . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.8 Statement of Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.9 Methodological Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.10 Why Structure Rather Than Meaning? . . . . . . . . . . . . . . . . . . . . 10 1.11 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2 Historical Foundations 13 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.2 The Structuralist Tradition . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.3 Ferdinand de Saussure and Relational Structure . . . . . . . . . . . . . . 14 2.4 Louis Hjelmslev and Formal Structural Systems . . . . . . . . . . . . . . 15 2.5 John Lyons and Structural Semantics . . . . . . . . . . . . . . . . . . . . . 15 2.6 The Shift from Static Structures to Dynamic Structures . . . . . . . . . . . 16 2.7 Gottlob Frege and Compositional Semantics . . . . . . . . . . . . . . . . . 16 2.8 Charles Sanders Peirce and Triadic Semiotics . . . . . . . . . . . . . . . . 17
vii
viii CONTENTS
2.9 Alfred North Whitehead and Process Philosophy . . . . . . . . . . . . . . 17 2.10 John Searle and the Chinese Room . . . . . . . . . . . . . . . . . . . . . . 18 2.11 Ned Block and the Separation of Consciousness . . . . . . . . . . . . . . 18 2.12 Predictive Processing and the Free Energy Principle . . . . . . . . . . . . 19 2.13 Mechanistic Interpretability . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.14 Historical Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 Three Definitions of Understanding 23 3.1 Motivating the Taxonomy . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 3.2 Phenomenological Understanding . . . . . . . . . . . . . . . . . . . . . . 24 3.3 Functional Understanding . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.4 Structural Understanding . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 3.5 Comparing the Three Conceptions . . . . . . . . . . . . . . . . . . . . . . 27 3.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 The Whiteheadian Framework 29 4.1 Process Rather Than Substance . . . . . . . . . . . . . . . . . . . . . . . . 29 4.2 Actual Occasions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 4.3 Prehension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 4.4 Concrescence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 4.5 Subjective Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 4.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 The Phenomenological Boundary 33 5.1 Structural and Phenomenological Components . . . . . . . . . . . . . . . 33 5.2 Axiomatic Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 5.3 Definitions and Empirical Criteria . . . . . . . . . . . . . . . . . . . . . . 34 5.4 The Structural Research Program . . . . . . . . . . . . . . . . . . . . . . . 34 5.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
II Scope Geometry, Algebra, and Categories 37
6 Scope Geometry 39 6.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 6.2 Primitive Intuition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 6.3 Why Boundaries Are Fundamental . . . . . . . . . . . . . . . . . . . . . . 40 6.4 State Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 6.5 Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 6.6 Interior Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 6.7 Nested Scope Geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 6.8 Hierarchy, Containment, and Inheritance . . . . . . . . . . . . . . . . . . 42 6.8.1 Local and Global Admissibility . . . . . . . . . . . . . . . . . . . . 43 6.8.2 Boundary Violations . . . . . . . . . . . . . . . . . . . . . . . . . . 43 6.8.3 The Scope Hierarchy as a Partially Ordered Structure . . . . . . . 43 6.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 6.10 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
CONTENTS ix
7 Primitive Operations and the Scope Calculus 45 7.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 7.2 Primitive Alphabet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 7.2.1 The Pop Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 7.2.2 The Refuse Operator . . . . . . . . . . . . . . . . . . . . . . . . . . 46 7.2.3 The Bind Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 7.2.4 The Collapse Operator . . . . . . . . . . . . . . . . . . . . . . . . . 47 7.2.5 Operational Completeness . . . . . . . . . . . . . . . . . . . . . . . 47 7.3 Composition of Primitive Operations . . . . . . . . . . . . . . . . . . . . . 47 7.4 The Primitive Rewriting System . . . . . . . . . . . . . . . . . . . . . . . . 49 7.5 Algebraic Properties of the Scope Calculus . . . . . . . . . . . . . . . . . 50 7.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 7.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
8 Category-Theoretic Foundations of Scope Geometry 53 8.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 8.2 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 8.3 Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 8.4 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 8.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
9 Foundational Theorems of Structural Semantics 57 9.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 9.2 Preliminary Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 9.3 Closure of Scope Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 58 9.4 Canonical Primitive Decomposition . . . . . . . . . . . . . . . . . . . . . 58 9.5 Canonical Normal Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 9.6 Admissibility Preservation . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 9.7 Representation Invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 9.7.1 Interpretation of Representation Invariance . . . . . . . . . . . . . 61 9.7.2 Representations as Coordinates Rather Than Meaning . . . . . . 62 9.7.3 Limitations of the Present Results . . . . . . . . . . . . . . . . . . 63 9.8 Levels of Structural Equivalence . . . . . . . . . . . . . . . . . . . . . . . . 63 9.9 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
10 Comparative Ontology and the Translation Functor 67 10.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 10.2 The Category of Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 10.3 Operational Correspondence . . . . . . . . . . . . . . . . . . . . . . . . . 68 10.4 Functorial Preservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 10.5 The Image of Subjective Form . . . . . . . . . . . . . . . . . . . . . . . . . 70 10.6 Interpretive Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 10.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
III Computational Semantics: Transformer Diagnostics 73
11 The Mathematical Structure of Transformer Architectures 75 11.1 Historical Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 11.2 Residual Stream Representation . . . . . . . . . . . . . . . . . . . . . . . . 76
x CONTENTS
11.3 Attention Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 11.4 Feed-Forward Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 11.5 Activation Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 11.6 Current Directions in Mechanistic Interpretability . . . . . . . . . . . . . 77 11.7 Relationship to Scope Geometry . . . . . . . . . . . . . . . . . . . . . . . 77 11.8 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
12 Dynamic Geometry of Activation Trajectories 79 12.1 Activation Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 12.2 Discrete Dynamical Systems Interpretation . . . . . . . . . . . . . . . . . 80 12.3 Layerwise Jacobians and Local Linearization . . . . . . . . . . . . . . . . 80 12.4 Curvature and Trajectory Length . . . . . . . . . . . . . . . . . . . . . . . 80 12.5 Local Stability and Lyapunov Analysis . . . . . . . . . . . . . . . . . . . . 80 12.6 Relationship to Scope Geometry . . . . . . . . . . . . . . . . . . . . . . . 81 12.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
13 Resonant Analysis of Activation Trajectories 83 13.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 13.2 Analytic Signal Representation . . . . . . . . . . . . . . . . . . . . . . . . 83 13.3 Amplitude, Instantaneous Phase, and Synchronization . . . . . . . . . . 84 13.4 Semantic Coherence Hypothesis . . . . . . . . . . . . . . . . . . . . . . . 84 13.5 The Marine Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 13.5.1 Doppler Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 13.5.2 The Salience Gate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 13.5.3 Energy Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . 86 13.6 Interpretive Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 13.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
14 Transformer Diagnostics and Algorithmic Architecture 89 14.1 Diagnostic Pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 14.2 Admissibility Estimation and Context Collapse . . . . . . . . . . . . . . . 90 14.3 Computational Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . 90 14.4 Case Study: Multi-Head Attention Residuals . . . . . . . . . . . . . . . . 91 14.4.1 The Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 14.4.2 Reading the Architecture as Scope Retrieval . . . . . . . . . . . . 91 14.4.3 What the Reading Clarifies, and What It Does Not . . . . . . . . . 92 14.5 Relationship to Existing Interpretability Methods . . . . . . . . . . . . . . 92 14.6 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 14.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
15 Experimental Predictions, Controls, and Falsification Protocols 95 15.1 Scientific Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 15.2 General Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . 95 15.3 Hypothesis I: Phase Coherence and Semantic Stability . . . . . . . . . . . 96 15.4 Hypothesis II: Frequency Drift and Context Collapse . . . . . . . . . . . 96 15.5 Hypothesis III: Admissibility and Error Prediction . . . . . . . . . . . . . 96 15.6 Control Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 15.7 Baseline Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 15.8 Failure Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 15.9 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
CONTENTS xi
15.10Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
IV Synthesis 99
16 Cross-Disciplinary Integration and Comparative Landscape 101 16.1 Predictive Processing and Active Inference . . . . . . . . . . . . . . . . . 101 16.2 Mechanistic Interpretability . . . . . . . . . . . . . . . . . . . . . . . . . . 101 16.3 Information Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 16.4 Category Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 16.5 Formal Language Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 16.6 Classical Structural Semantics . . . . . . . . . . . . . . . . . . . . . . . . . 102 16.7 Process Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 16.8 Large Language Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 16.9 Synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 16.10Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
17 Scope Limits, Boundary Conditions, and Non-Claims 105 17.1 The Scope of Structural Semantics . . . . . . . . . . . . . . . . . . . . . . 105 17.2 Consciousness, Truth, Agency, and Ethics . . . . . . . . . . . . . . . . . . 105 17.3 General Intelligence and Biological Realism . . . . . . . . . . . . . . . . . 106 17.4 Completeness and Boundary Conditions . . . . . . . . . . . . . . . . . . 106 17.5 Research Programme Versus Finished Theory . . . . . . . . . . . . . . . . 107 17.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 17.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
18 Conclusion: The Architecture of Structural Semantics 109 18.1 From Philosophy to Mathematics . . . . . . . . . . . . . . . . . . . . . . . 109 18.2 The Scope Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 18.3 Geometry of Semantic Organization . . . . . . . . . . . . . . . . . . . . . 110 18.4 Computational Realization . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 18.5 Scientific Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 18.6 Relationship to Existing Research . . . . . . . . . . . . . . . . . . . . . . . 110 18.7 Limitations and Future Directions . . . . . . . . . . . . . . . . . . . . . . 111 18.8 What Structural Semantics Does Not Claim . . . . . . . . . . . . . . . . . 111 18.9 Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
A Lambda Calculus and Scope Calculus 115 A.1 Translation of Lambda Terms . . . . . . . . . . . . . . . . . . . . . . . . . 115 A.2 The Correspondence Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 115
B The Scope Calculus as an Abstract Reduction System 117 B.1 Reduction Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 B.2 Primitive Rewrite Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 B.3 Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 B.4 Confluence and Canonical Normal Forms . . . . . . . . . . . . . . . . . . 118 B.5 Categorical Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 B.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
xii CONTENTS
C Cartesian Closed Categories and the Scope Calculus 121 C.1 Cartesian Closed Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 C.2 Interpretation of Types and Terms . . . . . . . . . . . . . . . . . . . . . . 121 C.3 CurryโHowardโLambek Interpretation . . . . . . . . . . . . . . . . . . . 122 C.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
D Higher Categories, Double Categories, and Scope Geometry 125 D.1 Double Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 D.2 Bicategorical Relaxation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 D.3 Computational Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . 126 D.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
E Topological Semantics of Scope Geometry 129 E.1 Scope Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 E.2 Continuity and Admissible Maps . . . . . . . . . . . . . . . . . . . . . . . 129 E.3 Fixed Points and Homotopy . . . . . . . . . . . . . . . . . . . . . . . . . . 130 E.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
F Sheaf Theory and Local-to-Global Reconstruction 131 F.1 Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 F.2 Reconstruction as Gluing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 F.3 Cohomological Obstructions . . . . . . . . . . . . . . . . . . . . . . . . . . 132 F.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
G Measure Theory, Entropy, and Admissibility 133 G.1 Measures and Admissibility . . . . . . . . . . . . . . . . . . . . . . . . . . 133 G.2 Entropy and Mutual Information . . . . . . . . . . . . . . . . . . . . . . . 133 G.3 Measure-Preserving Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 134 G.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
H Notation and Mathematical Conventions 135 H.1 Scope Calculus (Chapters 1, 6โ10) . . . . . . . . . . . . . . . . . . . . . . 135 H.2 Whitehead Reconstruction (Chapters 4โ5, 10) . . . . . . . . . . . . . . . . 135 H.3 Transformer Analysis (Chapters 11โ15) . . . . . . . . . . . . . . . . . . . 135 H.4 General Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
I Collected Proofs 137 I.1 Operational Completeness (Theorem 7.6) . . . . . . . . . . . . . . . . . . 137 I.2 Local Confluence Verification (Chapter 7, ยง7 and Appendix B) . . . . . . 137 I.3 Closure, Decomposition, and Reconstruction (Chapter 9) . . . . . . . . . 138
Preface
The rapid development of modern machine learning has revived questions that have occupied philosophy, mathematics, linguistics, and cognitive science for more than a century. Systems capable of producing coherent mathematical arguments, computer programs, scientific summaries, and extended dialogue have blurred traditional distinc- tions between syntax, semantics, reasoning, and understanding. At the same time, the conceptual vocabulary used to discuss these systems has become increasingly heteroge- neous: terms such as understanding, representation, meaning, reasoning, and intelligence are often employed without a common mathematical framework, leading to disagreements that arise as much from differences in definition as from differences in evidence. The purpose of this monograph is to construct such a framework. Rather than begin- ning with phenomenological assumptions, behavioral benchmarks, or biological com- mitments, we begin with the mathematics of constrained transformation. The central question investigated throughout this work is whether semantic capacity can be charac- terized as a structural property of admissible transformations over bounded contexts. The resulting framework, called Structural Semantics, combines ideas from topology, algebra, category theory, dynamical systems, signal processing, and machine learning, and engages process philosophy directly through a formal correspondence with White- headโs ontology. The mathematical developments are separated carefully from their philosophical interpretations, allowing each to be evaluated on its own terms. Where a result depends on mathematics beyond what a given chapter needs to state it, the corre- sponding construction โ from the ๐-calculus and abstract rewriting theory through categorical, topological, sheaf-theoretic, and measure-theoretic foundations โ is de- veloped independently in the appendices rather than assumed. The work is intended equally for mathematicians, computer scientists, philosophers of mind, and researchers in mechanistic interpretability. No commitment to any particular metaphysical position is assumed.
Relation to earlier structuralism. This monograph does not replace the structural se- mantics developed within twentieth-century linguistics. Rather, it generalizes the struc- turalist methodology from systems of linguistic signs to systems of admissible computa- tional transformations. The resulting framework shares with classical structuralism an emphasis on relational organization while introducing a different mathematical founda- tion based upon topology, algebra, category theory, and dynamical systems. The precise relationship between the two uses of the term is stated in remark 1.1 and developed at length in section 2.2.
xiii
xiv CONTENTS
Methodological Statement
Throughout this monograph, three distinct classes of claims are carefully distinguished, and every chapter closes with a status summary that sorts its content among them. Mathematical claims consist of definitions, lemmas, theorems, and proofs concern- ing topology, scope algebra, category theory, and related formal structures; these are established by proof from stated hypotheses and by nothing else. Computational claims concern algorithms, diagnostic procedures, and models of transformer representations; these are empirical hypotheses whose validity depends upon experimental evaluation, and they are presented as such rather than as consequences of the mathematics. Philo- sophical claims concern interpretation: they discuss possible implications of the mathe- matical and computational developments but are not presented as logical consequences of the formal theory. Maintaining this distinction is a central methodological principle of the present work. A theorem in Volume II never licenses a claim about consciousness or moral status; a diagnostic algorithm in Volume III is never treated as proven merely because it is well motivated; and a philosophical reading offered in Volume IV is never mistaken for a mathematical consequence of what precedes it.
xv
xvi CONTENTS
How to Read This Monograph
Although the eighteen chapters are arranged sequentially, the book is organized into four relatively independent volumes, and readers are encouraged to enter at the vol- ume suited to their interests rather than proceed straight through. Readers primarily interested in philosophical questions may begin with Volume I, which establishes the epistemic framing, historical lineage, and the boundary separating structural claims from phenomenological ones. Readers interested in the mathematical framework itself should concentrate on Volume II, where bounded scope geometries, the primitive al- gebra, the category of scopes, and the representation-invariance and process-geometry results are developed; the notation introduced in ?? is sufficient preparation, and the historical material is not a prerequisite. Readers interested in transformer analysis and computational diagnostics may read Volume III after becoming familiar with the basic scope formalism, since the diagnostics there are applications of, rather than alternatives to, the mathematics of Volume II. Volume IV situates the resulting framework within the broader landscape of philosophy, cognitive science, and applied mathematics, and states its limitations and non-claims explicitly. The appendices provide complete proofs, nota- tion tables, implementation details, and supplementary derivations referenced through- out the text, and are consulted as needed rather than read in sequence.
xvii
xviii CONTENTS
Notation and Conventions
Bold uppercase symbols denote categories or structured spaces, calligraphic symbols denote operators, constraint systems, or families of sets, and lowercase Greek letters denote morphisms, transformations, or parameters. Once a symbol is assigned a for- mal meaning it is preserved throughout the monograph and is not reassigned in a later chapter; appendix H collects the complete registry. Unless stated otherwise, S = โจโฌ, โฐ, ๐ฎchild โฉ denotes a bounded scope geometry, ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } denotes the primitive event alphabet, and ๐ โถ S๐ โ S๐ denotes an admissi- ble scope morphism. These three lines anticipate Chapter 6 and are given here only so that forward references elsewhere in the front matter are legible; their full definitions, and the argument for why the alphabet has exactly four elements, are deferred to Part II. The reconstruction of Whiteheadโs process ontology in Chapters 4 and 5 uses a sec- ond, disjoint set of symbols native to that literature: ๐ธ for an actual occasion, ๐ท for an- tecedent data, ๐ for a prehension, ๐ฃ for valuation polarity, ๐ for a satisfaction state, and ๐ด for Subjective Form. These are kept visually distinct from the Scope Calculus notation deliberately: no identification between the two systems is assumed before Chapter 10, where the translation functor is constructed and any correspondence between them is established as a proposition rather than presupposed by shared notation. Finally, a computational system is written ๐ = (๐, ๐ฏ), where ๐ is a state space and ๐ฏ is the set of admissible transition operators acting on ๐, each individual operator being written ๐ โ ๐ฏ with ๐ โถ ๐ โ ๐. This corrects an earlier draft in which ๐ฏ was written as if it were itself a single function; the set reading is what the rest of the text relies on whenever it writes ๐๐ โ ๐ฏ.
xix
xx CONTENTS
Part I
Epistemic and Historical Boundaries
1
Chapter 1
Epistemic Framing, Disentanglement, and Primary Research Questions
Chapter Roadmap. This chapter establishes the epistemic foundations of the monograph. It defines the central object of study, separates structural seman- tic capacity from neighboring philosophical concepts, introduces the primary research questions, summarizes the principal contributions, and outlines the methodological strategy that governs all subsequent chapters.
The limits of my language mean the limits of my world. โ Ludwig Wittgenstein, Tractatus Logico-Philosophicus
1.1 Introduction Few concepts have become simultaneously more important and more ambiguous than the notion of understanding. Within philosophy, cognitive science, linguistics, neuro- science, and artificial intelligence, the same word often refers to fundamentally differ- ent phenomena: subjective conscious awareness in one context, successful behavioral performance in another, internal representation, semantic competence, or predictive ca- pability in others. Debates regarding machine intelligence frequently proceed without agreement concerning the object under discussion itself. Recent advances in transformer-based language models have intensified this ambi- guity. Systems capable of producing mathematically coherent proofs, generating soft- ware, translating between natural languages, summarizing scientific literature, and par- ticipating in extended technical dialogue have challenged long-standing intuitions con- cerning the relationship between computation and semantic understanding [8, 45, 55]. At the same time, philosophical objections remain influential, including the Chinese Room [50], the distinction between phenomenal and access consciousness [7], and vari- ous forms of embodied cognition questioning whether statistical computation alone can constitute genuine understanding.
3
CHAPTER 1. EPISTEMIC FRAMING, DISENTANGLEMENT, AND PRIMARY 4 RESEARCH QUESTIONS
The present work does not attempt to resolve the metaphysical question of conscious- ness. Instead it addresses a more precise question: is there a mathematically definable notion of semantic capacity that depends only upon the preservation of structural re- lationships during computation? If such a notion exists, it should admit formal defi- nitions, rigorous proofs, computational implementations, and empirical evaluation in- dependently of debates concerning subjective experience. We refer to this proposed framework throughout as Structural Semantics. The central hypothesis is that semantic capacity may be understood as the preservation of admissible structural relationships across transformations of bounded contexts; later chapters develop this proposal using topology, rewriting systems, category theory, dynamical systems, and signal processing. Remark 1.1 (Terminological note). The phrase structural semantics has an established meaning in linguistics, referring to structuralist theories of lexical meaning developed by Saussure, Hjelmslev, Lyons, Greimas, Coseriu, and related authors. Throughout this monograph the term is used in a different, explicitly defined sense: a mathematical framework for the preservation of admissible structural transformations over bounded contexts. The present theory is inspired by the general structuralist tradition but is not a development of lexical field theory, componential analysis, or structural linguistics; section 2.2 situates the two usages against one another in detail.
1.2 Motivating the Problem The history of theories of meaning exhibits a recurring pattern: philosophical accounts generally begin from one of three starting points. The first begins from conscious expe- rience, regarding understanding as inseparable from phenomenal awareness or inten- tionality, as in phenomenology, process philosophy, and much contemporary discussion of consciousness. The second begins from observable behavior, as in functionalist ap- proaches that characterize understanding in terms of successful interaction, prediction, or problem solving, treating internal implementation as secondary to externally measur- able performance. The third begins from representation itself, asking whether internal structures preserve relationships sufficiently to support inference, abstraction, and gen- eralization independently of their physical realization. Each perspective captures an important aspect of cognition, but they address dif- ferent questions, and difficulties arise when conclusions appropriate to one framework are transferred without modification into another: a behavioral criterion cannot by it- self establish phenomenology, and the absence of phenomenological evidence does not immediately imply the absence of mathematically characterizable semantic structure. This observation motivates the methodological separation adopted here. Rather than collapsing the three viewpoints into a single definition, we introduce a formal frame- work that treats structural semantic capacity as an independent object of investigation, deliberately separated from the philosophical questions concerning consciousness that remain important but lie outside its scope.
1.3 Epistemic Separation The first objective of the monograph is therefore methodological rather than mathemati- cal. Before introducing scope geometry or category theory, it is necessary to distinguish several concepts that are frequently conflated in the literature: phenomenal conscious- ness, behavioral competence, veridical truth, intentional agency, and structural semantic
capacity. Only the last of these constitutes the primary object of mathematical investi- gation developed in subsequent chapters. The remaining concepts may interact with structural semantic capacity, but they are not identified with it; results proved within the Scope Calculus should not automatically be interpreted as results concerning con- sciousness, moral agency, or epistemic truth, which belong to different theoretical do- mains and require independent justification. This separation serves two purposes. First, it permits the mathematical development to proceed independently of unresolved philosophical debates. Second, it makes empir- ical evaluation possible, since structural properties may be investigated directly through computational observation without requiring access to subjective experience.
1.4 Formal Foundations Having established the methodological scope of the monograph, we now introduce the mathematical objects on which the remainder of the development is built, identifying only the minimal structures required to state the principal research questions precisely. Throughout the text a system is understood abstractly, with no assumption regarding its physical realization: it may be implemented biologically, digitally, symbolically, me- chanically, or by any other computational substrate, and the definitions below are corre- spondingly representation-independent.
Definition 1.2 (Computational system). A computational system is an ordered pair ๐ = (๐, ๐ฏ), where ๐ is a non-empty state space and ๐ฏ is a set of admissible state- transition operators ๐ โถ ๐ โ ๐ acting on ๐. No assumptions are made regarding continuity, discreteness, determinism, or stochasticity unless explicitly stated.
Remark 1.3. The abstraction deliberately encompasses finite-state automata, symbolic rewriting systems, neural networks, probabilistic graphical models, dynamical systems, and biological cognitive architectures. The mathematical framework developed through- out this monograph applies at this level of generality whenever the stated hypotheses are satisfied.
1.4.1 Representations The notion of representation is equally general. Rather than committing to a particular implementation, we define a representation only through its structural role.
Definition 1.4 (Representation). Let ๐ = (๐, ๐ฏ) be a computational system. A repre- sentation is an element ๐ฅ โ ๐ together with the admissible transition structure induced by ๐ฏ. Two representations need not possess identical internal encodings in order to support identical structural behavior.
This definition intentionally avoids identifying representations with vectors, sym- bols, neurons, graphs, or linguistic expressions; a representation refers only to the math- ematical state on which the transition operators act.
1.4.2 Transformation Understanding, regardless of philosophical commitment, concerns not isolated states but the evolution of states under transformation, so transformations rather than static representations become the primary object of investigation.
CHAPTER 1. EPISTEMIC FRAMING, DISENTANGLEMENT, AND PRIMARY 6 RESEARCH QUESTIONS
Definition 1.5 (Transformation sequence). Let ๐ = (๐, ๐ฏ) be a computational sys- tem. A transformation sequence is a finite ordered family ๐ = (๐ฅ0 , ๐ฅ1 , โฆ , ๐ฅ๐ ) such that ๐ฅ๐+1 = ๐๐ (๐ฅ๐ ) for some ๐๐ โ ๐ฏ. The sequence ๐ records the structural evolution of a representation through the state space.
1.4.3 Motivation for Scope Geometry Transformation sequences alone are insufficient for characterizing semantic preserva- tion, since two systems may undergo identical numbers of state transitions while pre- serving fundamentally different contextual relationships. A symbolic proof assistant may reject an invalid inference by enforcing logical typing constraints, whereas a neu- ral language model may continue generating fluent text despite violating assumptions introduced earlier in the reasoning process; both systems perform transformations, yet only one explicitly preserves the contextual constraints governing the computation. The central mathematical idea developed in this monograph is that such constraints may themselves be treated as geometric objects. Rather than asking whether a repre- sentation โcontains meaning,โ we ask whether successive transformations preserve the admissible boundaries within which meaning is defined. This observation motivates bounded scope geometries, developed fully in Chapters 6 through 10; for the present chap- ter it suffices to observe that semantic preservation appears to depend not upon individ- ual states but upon the structural integrity of the permissible transformations connect- ing them.
1.5 The Disentanglement Operator The preceding discussion suggests that much of the disagreement surrounding seman- tic capacity originates not from conflicting mathematical results but from differences in the domains over which the term understanding is defined. Before constructing an alge- bra of semantic transformations, it is therefore necessary to specify the epistemic domain within which subsequent definitions operate. The purpose of the Disentanglement Op- erator is precisely this: rather than measuring semantic capacity, it separates semantic capacity from other properties that may coexist with it but are not logically implied by it, allowing later chapters to formulate structural results without simultaneously making claims concerning consciousness, truth, or agency.
Definition 1.6 (Disentanglement operator). Let ๐ซ = {ฮฆ, T, A, F, S} denote the collec- tion of properties commonly associated with understanding, where ฮฆ denotes phenom- enal consciousness, T denotes veridical truth, A denotes intentional agency, F denotes syntactic fluency, and S denotes structural semantic capacity. The Disentanglement Op- erator ๐ โถ ๐ซ โ 2๐ซ identifies structural semantic capacity as an independent object of mathematical investigation by separating S from the remaining properties: throughout this monograph, S โข ฮฆ, S โข T, S โข A, and S โข F.
Remark 1.7. Definition 1.6 establishes only logical independence; it does not imply statis- tical, causal, or empirical independence. Systems possessing high structural semantic capacity may also exhibit high behavioral competence, and biological systems may si- multaneously possess structural semantic capacity and phenomenal experience. The present framework merely asserts that these properties are mathematically distinguish- able and should not be identified without additional argument.
1.6. BOUNDARY CONSTRAINTS AND BOUNDED CONTEXTS 7
1.5.1 Interpretation The distinction above serves as a recurring methodological principle throughout the monograph. Structural Semantics is not presented as a theory of consciousness: whether subjective awareness accompanies a particular computational architecture remains out- side the scope of the present formalism and is revisited only in the philosophical discus- sion of Chapters 16 and 17. Nor is it a theory of objective truth: a system may preserve contextual relationships perfectly while reasoning within an entirely fictional, hypothet- ical, or mathematically inconsistent universe, since the framework evaluates the preser- vation of internal constraint structures rather than correspondence with external reality. Nor is it a theory of agency: goal-directed behavior, preference formation, intentional action, and moral status require additional theoretical machinery that is deliberately ex- cluded from the present development. Finally, Structural Semantics is not equivalent to linguistic fluency: contemporary language models routinely generate syntactically well- formed sentences while simultaneously violating assumptions established only a few paragraphs earlier, a discrepancy that motivates the distinction between surface gram- matical regularity and the preservation of admissible contextual structure.
Example 1.8 (Structural coherence without truth). Consider a narrative set entirely within a fictional world governed by its own internal rules โ for instance, that a par- ticular journey between two named locations cannot be completed within a stated span of time. A system that continues to respect this constraint throughout an extended nar- rative, refusing continuations that would violate it, is exhibiting structural semantic ca- pacity with respect to that narrativeโs admissibility conditions, regardless of the fact that neither the locations nor the constraint correspond to anything in physical reality. Con- versely, a system that abruptly violates a constraint it had itself established earlier in the same narrative โ introducing a character incompatible with rules already fixed for that fictional world โ has produced a boundary violation in the sense of Chapter 6, indepen- dently of whether either version of the narrative is judged more โrealistic.โ Structural Semantics evaluates the second kind of failure. It has nothing to say about the first.
1.6 Boundary Constraints and Bounded Contexts The Disentanglement Operator isolates the domain of investigation but does not yet specify the mathematical structures through which semantic preservation is expressed. We therefore introduce the notion of a boundary constraint: informally, the conditions under which a state may enter, remain within, or leave a contextual region. Such con- straints appear in type systems, formal grammars, topological boundaries, logical the- ories, and dynamical systems, and the present formulation deliberately abstracts over these examples, treating a boundary not as a physical surface but as a collection of ad- missibility conditions governing transformations of representations.
Definition 1.9 (Boundary constraint). Let ๐ = (๐, ๐ฏ) be a computational system. A boundary constraint is a predicate โฌ โถ ๐ ร ๐ฏ โ {0, 1}, where โฌ(๐ฅ, ๐ ) = 1 indicates ๐ that the transition ๐ฅ โ โ ๐ (๐ฅ) is admissible, and โฌ(๐ฅ, ๐ ) = 0 indicates that it violates the structural constraints governing the current context.
Definition 2.7 intentionally avoids assigning any particular computational interpre- tation to โฌ: in different applications it may represent typing constraints, logical consis- tency conditions, physical conservation laws, permission systems, grammatical restric-
CHAPTER 1. EPISTEMIC FRAMING, DISENTANGLEMENT, AND PRIMARY 8 RESEARCH QUESTIONS
tions, or learned admissibility relations extracted from data. The subsequent mathemat- ics depends only on the existence of such a predicate and not on its implementation.
1.6.1 Bounded contexts Semantic reasoning rarely occurs within an unrestricted universe of states. Mathemati- cal proofs introduce hypotheses, computer programs create lexical environments, con- versations establish topics, and scientific models operate under explicitly stated assump- tions; in each case reasoning proceeds within a bounded context whose admissibility conditions determine which transformations remain valid.
Definition 1.10 (Bounded context). A bounded context is a pair (๐, โฌ) consisting of a state domain together with a boundary constraint predicate. Its admissible region is
Adm(๐, โฌ) = {๐ฅ โ ๐ โฃ โฌ(๐ฅ, ๐ ) = 1 for every admissible ๐ }.
The notation Adm(๐, โฌ) recurs throughout the monograph, denoting informally the collection of states that satisfy every structural constraint currently imposed by the con- text.
1.6.2 Motivation for scope geometry Bounded contexts remain insufficient for describing nested reasoning processes. Mod- ern computational systems routinely manipulate multiple interacting contexts simul- taneously: a theorem may invoke several nested lemmas, a compiler maintains inde- pendent lexical environments, transformer models preserve long-range dependencies across dozens of processing layers, and human reasoning shifts continuously between assumptions, sub-problems, hypothetical scenarios, and meta-level analyses. These ex- amples suggest that semantic preservation cannot be understood solely through indi- vidual admissible regions, but requires a hierarchy of interacting contexts whose rela- tionships are themselves subject to structural constraints. This motivates scope geometry: rather than viewing context as a single admissible set, it models reasoning as a collec- tion of nested bounded regions connected by explicit admissibility-preserving transfor- mations.
Definition 1.11 (Scope geometry, preliminary). A scope geometry is a bounded context together with an explicitly specified collection of subordinate bounded contexts and the structural relations connecting them.
This definition serves only as a conceptual preview; the complete formal treatment, including the algebra of scope transformations, admissibility morphisms, and categori- cal structure, is developed systematically in Chapters 6 through 10. In particular, the full tuple S = โจโฌ, โฐ, ๐ฎchild โฉ is deliberately withheld until Chapter 6, so that it appears there as the culmination of the intervening development rather than as a symbol introduced before its mathematical necessity.
1.6.3 A preview of Structural Semantics The progression established thus far is deliberately incremental: beginning with compu- tational systems, we introduced representations, transformation sequences, boundary constraints, bounded contexts, and finally scope geometries, each concept extending
the previous one while remaining independent of any particular computational archi- tecture. The central hypothesis explored throughout the remainder of this monograph may now be stated informally: semantic capacity is not identified with isolated repre- sentations, but is associated with the preservation of admissible structure across trans- formations acting on hierarchies of bounded contexts.
1.7 The Research Program
The preceding sections established the conceptual vocabulary on which the remainder of this monograph is built, but no substantial mathematical claims have yet been made; we have established only the domain within which such claims may be formulated. The objective of the remainder of this work is to determine whether semantic capacity can be characterized as a structural property of admissible transformations over bounded contexts, and this objective separates into four progressively more specialized research questions, concerning conceptual foundations, mathematical structure, computational implementation, and empirical verification respectively. The first question concerns the object of study itself: can semantic capacity be de- fined independently of phenomenal consciousness, intentional agency, and behavioral performance? The purpose is not to deny the importance of consciousness but to ask whether semantic structure admits an independent mathematical description; an affir- mative answer would imply that semantic capacity constitutes a legitimate object of mathematical investigation irrespective of oneโs preferred philosophical theory of mind. Volumes I and II develop the formal foundations required to address this question. The second question concerns mathematical structure: which mathematical struc- tures are necessary and sufficient to describe semantic preservation under transforma- tion? Several candidate frameworks already exist, including topology, graph theory, rewriting systems, information geometry, category theory, and dynamical systems; rather than replacing these disciplines, the present work attempts to synthesize selected ideas from each into a unified theory of bounded semantic transformations, developed sys- tematically in Chapters 6 through 10. The third question concerns implementation: can structural semantic preservation be extracted from the internal representations of computational systems? Recent ad- vances in mechanistic interpretability have demonstrated that large neural networks ad- mit increasingly sophisticated internal analysis through activation trajectories, sparse feature models, residual stream decompositions, and circuit-level investigations [15, 42, 44], suggesting that semantic structure may itself become an observable computational quantity rather than merely a philosophical abstraction. Volume III develops one possi- ble mathematical framework for such measurements. Finally, every mathematical model intended to describe computational systems should generate consequences capable of empirical evaluation: can structural measures of se- mantic preservation generate falsifiable predictions distinguishing genuine contextual coherence from superficial statistical fluency? The monograph therefore concludes its technical development by constructing explicit diagnostic algorithms together with ex- perimental protocols designed to evaluate the proposed framework, and these protocols are intentionally formulated so that negative results are as informative as positive ones; a framework that cannot in principle be falsified falls outside the scope of the present methodology.
CHAPTER 1. EPISTEMIC FRAMING, DISENTANGLEMENT, AND PRIMARY 10 RESEARCH QUESTIONS
1.8 Statement of Contributions The mathematical developments presented throughout this monograph are organized around five principal contributions. The first is a formal Scope Calculus governing ad- missible transformations over bounded contexts, generated minimally by four primi- tive operations whose algebraic properties are developed in Chapters 7 and 9. The sec- ond is a structural definition of semantic capacity that depends only on admissibility- preserving transformations rather than on any particular computational substrate, per- mitting semantic equivalence to be investigated independently of implementation. The third is a categorical formulation of scope transformations together with a translation functor relating elements of Whiteheadian process philosophy to the Scope Calculus, de- veloped in Chapter 10 as a comparative rather than reductive construction. The fourth is the formulation of the MEM|8 analytical framework for studying activation trajecto- ries within transformer architectures, which treats transformed activation trajectories as objects of signal analysis using established tools from harmonic analysis and digital signal processing rather than proposing a physical wave ontology. The fifth is an em- pirical diagnostic program designed to evaluate the proposed mathematical structures, with explicit falsification criteria, statistical controls, computational complexity analy- ses, and implementation protocols developed in Chapters 14 and 15 and expanded in the appendices.
1.9 Methodological Roadmap The remainder of this monograph is organized into four interconnected volumes. Vol- ume I examines the historical and philosophical development of the concept of under- standing, culminating in a precise characterization of the phenomenological boundary separating structural questions from questions concerning subjective experience. Vol- ume II develops the mathematical foundations of Structural Semantics, beginning with bounded scope geometries and primitive operations and proceeding through category- theoretic constructions, normalization theorems, admissibility preservation, and rep- resentation invariance. Volume III shifts from abstract mathematics to computational application, reviewing established results from mechanistic interpretability before in- troducing activation trajectory analysis, the MEM|8 framework, and empirical diagnos- tic procedures for contemporary neural architectures. Volume IV situates the resulting framework within the broader landscape of philosophy, cognitive science, artificial intel- ligence, and applied mathematics, concluding with explicit limitations, non-claims, and open research questions. The appendices provide complete proofs, notation tables, im- plementation details, and supplementary mathematical derivations referenced through- out the main text.
1.10 Why Structure Rather Than Meaning? Before turning to the historical survey, it is worth pausing on a choice that has been implicit throughout this chapter: the theory developed here does not begin by asking what meaning is. It begins by asking what relations among representations remain in- variant as those representations are transformed. This is a substantive methodological decision, and it is worth making the reasoning behind it explicit rather than leaving it to be inferred from the sequence of definitions.
1.11. CONCLUDING REMARKS 11
A direct definition of meaning inherits every disagreement already present in the literature surveyed in Chapter 2: whether meaning requires reference, mental content, use, verification conditions, or truth conditions is itself a live dispute, and any defini- tion chosen at the outset would import that dispute into the formal development rather than resolving it. Structure, by contrast, is comparatively tractable. Whether a transfor- mation preserves a given boundary, whether two admissible sequences reduce to the same normal form, whether a projection preserves admissibility โ these are questions with determinate answers once the relevant objects are specified, regardless of oneโs antecedent theory of meaning. This is not a claim that structure exhausts meaning, or that a full theory of meaning could be built from structural invariants alone. Chapter 17 states explicitly that Struc- tural Semantics does not attempt such a reduction. The claim is narrower and more defensible: whatever else meaning involves, semantic systems exhibit organization that persists across representational change, and that organization can be studied on its own terms, prior to and independently of settling what meaning ultimately is. Beginning with structure therefore is not a wager that structure is all there is. It is a decision about where a tractable mathematical investigation can honestly start.
1.11 Concluding Remarks The principal objective of this introductory chapter has been methodological rather than technical. Before constructing a formal theory of Structural Semantics, it has been nec- essary to delimit the domain within which the theory is intended to operate. Begin- ning with the observation that the term understanding possesses multiple incompati- ble meanings across philosophy, cognitive science, and artificial intelligence, we argued that many contemporary disagreements arise from treating these distinct meanings as though they referred to a single mathematical object. Rather than attempting to re- solve longstanding debates concerning phenomenal consciousness or intentionality, the present work isolates structural semantic capacity as an independent object of investiga- tion. The resulting framework proceeds from computational systems to representations, from representations to transformations, from transformations to boundary constraints, and finally toward bounded scope geometries; each concept introduces precisely the mathematical structure required for the next while avoiding unnecessary assumptions regarding implementation or physical substrate. Consequently, no substantive claims concerning transformer architectures, neural representations, or signal-processing mod- els have yet been made, since those developments depend on mathematical results not yet established. Maintaining this logical order is an important methodological principle throughout the monograph.
Status of Results: Chapter 1 Established background. This chapter reviewed established literature concerning phi- losophy of mind, artificial intelligence, cognitive science, and machine learning, includ- ing classical discussions of understanding [7, 50, 54], predictive processing [11, 18], modern deep learning [22, 55], and mechanistic interpretability [15, 44].
Formal developments. The chapter introduced the Disentanglement Operator, Bound- ary Constraint, Bounded Context, and preliminary Scope Geometry as original defini-
CHAPTER 1. EPISTEMIC FRAMING, DISENTANGLEMENT, AND PRIMARY 12 RESEARCH QUESTIONS
tions establishing the mathematical domain of Structural Semantics, and stated the pri- mary research questions and overall methodological architecture of the monograph.
Empirical hypotheses. None tested here. The only forward-looking claim is that struc- tural semantic preservation should admit computational measurement independently of behavioral fluency, a proposal developed formally in Volume III and subjected to ex- plicit falsification protocols in Chapter 15.
Chapter 2
Historical Foundations
Chapter Roadmap. This chapter reconstructs the intellectual lineage from which Structural Semantics emerges, from Saussure through mechanistic interpretabil- ity, identifying for each tradition what it contributes and what remains to be for- malized.
The value of a sign results solely from the simultaneous presence of other signs. โ Ferdinand de Saussure, Cours de Linguistique Gรฉnรฉrale
2.1 Introduction The purpose of this chapter is analytical reconstruction rather than mere historical expo- sition. No existing framework discussed here is dismissed as incorrect; rather, each con- tributes an important structural component later incorporated into the Scope Calculus of Volume II. Structural linguistics contributes relational organization, formal seman- tics contributes compositional interpretation, process philosophy contributes dynamic evolution, category theory contributes compositional abstraction, and mechanistic inter- pretability contributes empirical access to internal representations. The central thesis of this chapter is that these traditions become considerably more compatible once semantic organization is treated as a problem of admissible structural transformation rather than static representation. Throughout, we ask of each tradition which mathematical object is taken to be primitive, how semantic organization is represented, and which structural properties remain invariant under transformation.
2.2 The Structuralist Tradition The expression structural semantics has a well-established history within twentieth-century linguistics. Beginning with Ferdinand de Saussure, structuralist approaches empha- sized that the meaning of linguistic units derives not from isolated reference but from their position within systems of relational differences [13], motivating a body of work in- cluding lexical field theory, componential analysis, relational semantics, and later struc- tural linguistics [12, 23, 26, 36]. Despite their differences, these approaches share the methodological commitment that semantic phenomena are explained by analyzing the
13
relational organization of linguistic structures rather than treating individual words as independent semantic atoms. The present monograph adopts this structuralist insight while extending it beyond linguistic objects: rather than studying lexical systems, semantic fields, or componential features, it investigates admissible transformations over bounded computational con- texts, so that the primary mathematical objects become transformations, boundary con- straints, and scope geometries rather than lexical relations. The adjective โstructuralโ therefore retains its original emphasis on relations, but the mathematical domain is fun- damentally different: where classical structural semantics studies relations among signs, Structural Semantics as developed here studies relations among admissible computa- tional transformations.
Classical Structural Semantics Structural Semantics (this work) Lexical units Computational states Semantic fields Bounded scope geometries Sense relations Admissible transformations Componential features Boundary constraints Lexical structure Constraint geometry Meaning relations Structural preservation
Table 2.1: Comparison between classical structural semantics and the framework devel- oped in this monograph.
Definition 2.1 (Classical structural semantics). Classical structural semantics denotes the family of linguistic theories that analyze meaning through systems of structural re- lations among linguistic units, including lexical fields, semantic features, and relational networks [13, 23, 26, 36]. Definition 2.2 (Structural Semantics, this work). Throughout this monograph, Struc- tural Semantics denotes the study of semantic preservation under admissible transfor- mations of bounded computational contexts. Unless stated otherwise, the term refers to this mathematical framework rather than to definition 2.1.
2.3 Ferdinand de Saussure and Relational Structure Modern structural approaches to language originate largely with Saussureโs posthu- mously published Cours de Linguistique Gรฉnรฉrale (1916) [13]. Saussure rejected the view that linguistic meaning is intrinsic to isolated words, arguing instead that linguistic val- ues arise from systems of mutual distinction: a sign consists of a signifier and a sig- nified, neither possessing independent meaning in isolation, with meaning emerging through the network of differences separating one sign from every other sign in the sys- tem. From the perspective developed in this monograph, Saussureโs insight may be read as an early recognition that semantic systems are constraint systems, in which meaning is preserved not because individual symbols remain unchanged but because admissible structural relationships remain stable under transformation. Observation 2.3 (Relational priority). Saussureโs structural linguistics assigns explana- tory priority to relations rather than isolated linguistic entities. The Scope Calculus extends this principle by treating admissible transformations of those relations as the primary mathematical objects of investigation.
2.4. LOUIS HJELMSLEV AND FORMAL STRUCTURAL SYSTEMS 15
Saussureโs framework nevertheless remained largely descriptive: it introduced no explicit algebra governing semantic transformations, no theory of nested contextual or- ganization, and no mathematical account of how structural relationships evolve during reasoning or computation. These omissions lay outside the objectives of early structural linguistics, but they identify precisely the mathematical questions addressed later in this monograph, beginning with the formal development of bounded scope geometries in Chapter 6.
2.4 Louis Hjelmslev and Formal Structural Systems Louis Hjelmslev sought to give structural linguistics a more rigorous formal foundation, attempting in his Prolegomena to a Theory of Language (1943) to construct linguistics as an autonomous discipline based on explicit structural principles rather than psycholog- ical or historical explanation [25]. Central to his account is the distinction between the planes of expression and content, each possessing its own internal organization, with linguistic meaning arising through their systematic correlation; and the distinction be- tween form, the abstract organization governing a system, and substance, the material realization through which that form is implemented, so that the same structural organi- zation may be realized through different physical media without altering the underlying linguistic system. This distinction foreshadows the Representation Invariance Theorem of Chapter 9, which shows in a different mathematical setting that semantic equivalence depends on preserved structural transformations rather than on a particular represen- tational substrate. Hjelmslevโs theory remains primarily descriptive rather than operational: it intro- duces no explicit notion of admissible state transformation, no rewriting calculus gov- erning structural evolution, and no topology of contextual boundaries. The Scope Cal- culus of Volume II may be understood as extending structural linguistics from static organization toward a fully operational theory of constrained semantic dynamics.
2.5 John Lyons and Structural Semantics The expression structural semantics entered modern linguistic usage primarily through the work of John Lyons [35, 36], who shifted attention from the internal organization of linguistic signs toward the systematic organization of lexical meaning itself, arguing that lexical meaning is determined by the network of semantic relations within which each lexical item participates rather than by isolated dictionary definitions. Lexical field theory studies collections of semantically related expressions whose meanings are mu- tually constrained, and Lyons further emphasized systematic sense relations such as synonymy, antonymy, hyponymy, incompatibility, entailment, and meronymy, which organize vocabulary into structured semantic networks and may be read as constraints restricting the admissible configurations of lexical organization. The present monograph agrees with Lyons that semantic organization is relational, but extends this observation in two directions: the primitive objects of the theory are bounded contextual scopes rather than lexical items, and the primary concern is not merely the existence of semantic relations but the admissible transformations that pre- serve those relations during computation. Whereas structural semantics in the linguis- tic tradition describes the organization of semantic systems, the Scope Calculus seeks to describe the dynamics of semantic organization itself.
2.6 The Shift from Static Structures to Dynamic Structures The three traditions surveyed so far โ Saussure, Hjelmslev, and Lyons โ share a fam- ily resemblance worth naming explicitly before the chapter turns to Frege, Peirce, and Whitehead. Each studies a fixed relational system: a language at a given moment, whose signs stand in relations of difference, whose planes of expression and content are inter- nally organized, whose lexical items occupy a stable network of sense relations. The object of analysis is the systemโs synchronic organization, not its evolution. Saussure himself insisted on this separation, distinguishing the synchronic study of a language state from the diachronic study of its historical change, and largely restricting structural analysis to the former. This emphasis was appropriate to the questions classical structuralism asked. It be- comes a limitation only relative to a different question, the one this monograph asks: not how a system of distinctions is organized at a moment, but how that organization be- haves as the system undergoes transformation. A bounded scope geometry, introduced formally in Chapter 6, is in one respect a direct descendant of the Saussurean relational system โ it too specifies a structure of admissible distinctions rather than treating en- tities as meaningful in isolation. What it adds is that the geometry is defined together with the transformations that act on it, so that admissibility becomes a property of tran- sitions rather than a property of a single static configuration. The remaining sections of this chapter โ Fregeโs compositional combination of mean- ings, Peirceโs semiosis as a chain of successive interpretants, and above all Whiteheadโs replacement of substance with process โ can be read as successive steps away from purely synchronic description and toward treating transformation itself as the primary object of study. None of them frames the shift in exactly Saussureโs terms, and none an- ticipates the Scope Calculus specifically. But by the time the chapter reaches Whitehead, the historical trajectory it traces has already moved a considerable distance from the static relational systems with which it began, and Part II inherits that trajectory rather than starting it.
2.7 Gottlob Frege and Compositional Semantics Although Frege predates twentieth-century structural linguistics, his work established many of the foundations of modern formal semantics [17]. His distinction between sense (Sinn) and reference (Bedeutung) observes that two expressions, such as โMorn- ing Starโ and โEvening Star,โ may refer to the same object while presenting it under different conceptual descriptions, so that semantic analysis cannot be reduced solely to reference. Fregeโs Principle of Compositionality further holds that the meaning of a complex expression depends on the meanings of its constituents and the rule by which they are combined, a principle underlying formal semantics, type theory, programming- language semantics, and modern logic. The Scope Calculus accepts that semantic organization is compositional, but general- izes compositionality beyond syntactic construction: rather than asking only how mean- ings combine, it investigates whether the transformations acting on those meanings pre- serve admissible structural constraints, so that composition becomes one instance of a broader mathematical problem concerning the preservation of semantic organization under transformation. Fregeโs framework primarily concerns the interpretation of com- pleted expressions; it provides no explicit theory of evolving contextual states and no algebra governing semantic transformations, gaps that the Scope Calculus is built to
2.8. CHARLES SANDERS PEIRCE AND TRIADIC SEMIOTICS 17
address while building directly on the compositional insight.
2.8 Charles Sanders Peirce and Triadic Semiotics Charles Sanders Peirce developed a broader semiotic framework intended to explain how signs function across reasoning, communication, scientific inquiry, and cognition [46, 51], rejecting the view that signs exist merely as binary relations between words and objects. Every act of signification instead involves a triadic relation of sign, object, and interpretant, and because an interpretant may itself function as a new sign, interpreta- tion may propagate through a chain Peirce called unlimited semiosis: meaning develops through an evolving network of successive interpretations rather than a single isolated semantic event. Peirceโs semiotics anticipates several themes developed later in this monograph, par- ticularly its emphasis on process over static representation, but his interpretant remains a philosophical concept describing semantic mediation rather than an explicit mathe- matical operator. The Scope Calculus instead describes the structural constraints under which interpretation may evolve without producing contextual inconsistency, so that interpretation becomes one possible realization of a more general theory of admissi- ble semantic evolution. Peirceโs emphasis on process provides an important bridge to Whitehead, examined next, whose metaphysics treats process itself as the primitive con- stituent of reality.
2.9 Alfred North Whitehead and Process Philosophy Whiteheadโs Process and Reality (1929) sought to replace traditional substance meta- physics with a comprehensive process ontology [56]. Where classical ontologies treat objects as persisting substances undergoing accidental modification, Whitehead argued that the fundamental constituents of reality are events of becoming, with stability emerg- ing from recurring patterns of organized process rather than from immutable substance โ an inversion from objects to processes that becomes central to the categorical transla- tion developed in Chapter 10. The primitive unit of Whiteheadโs ontology is the actual entity or actual occasion: a transient event that arises, integrates information from prior occasions through prehen- sion, achieves completion (satisfaction), and contributes to future processes. Whitehead distinguishes positive prehensions, which contribute information to the developing en- tity, from negative prehensions, which exclude it, and calls the process by which mul- tiple prehensions integrate into a single completed entity concrescence. Among these components, subjective form โ the manner in which a datum is experienced during pre- hension โ is explicitly phenomenal, and cannot simply be removed without altering the ontology; chapters 4 and 5 ask whether the structural organization of prehensions can be formalized independently of subjective form, and answer affirmatively. Whiteheadโs framework shares with the present work the conviction that transfor- mation is more fundamental than static state and that complex structures emerge from compositions of simpler relations, but it develops a metaphysical ontology intended to describe the general structure of reality, whereas the Scope Calculus develops a mathe- matical theory of admissible transformations acting on bounded contextual structures. The present work therefore neither adopts nor rejects Whiteheadโs metaphysical conclu- sions, extracting instead those structural components capable of rigorous mathematical
formalization, a project undertaken in full in Chapters 4, 5, and 10.
2.10 John Searle and the Chinese Room Among the most influential critiques of computational theories of mind is Searleโs Chi- nese Room argument [50], directed against the claim that the correct execution of a computer program is itself sufficient for understanding. Searle imagines a person who does not understand Chinese, enclosed in a room with a rule book, manipulating Chi- nese symbols according to purely formal instructions and producing output indistin- guishable from that of a fluent speaker; to external observers the room appears to un- derstand Chinese, but Searle argues that no genuine understanding exists, since the individual manipulates symbols syntactically without attaching semantic significance to them. Syntax alone, on this view, is insufficient for semantics. The present monograph neither accepts nor rejects Searleโs conclusions regarding phenomenal consciousness, arguing instead that the Chinese Room addresses a differ- ent question from the one investigated here: Searle asks whether formal symbol manipu- lation is sufficient for subjective understanding, while the Scope Calculus asks whether semantic organization can be mathematically characterized through the preservation of admissible structural transformations. These questions are logically independent, and this distinction motivates the epistemic separation introduced in Chapter 1 and devel- oped formally in Chapter 5. Rather than answering Searleโs question directly, Struc- tural Semantics replaces the binary question of whether a machine understands with the more mathematically tractable question of which structural properties of a compu- tational system remain invariant under admissible semantic transformation.
2.11 Ned Block and the Separation of Consciousness Ned Blockโs distinction between phenomenal consciousness and access consciousness [7] clarified that many philosophical disagreements concerning artificial intelligence arise because these fundamentally different notions of consciousness are routinely conflated. Phenomenal consciousness refers to subjective qualitative experience, what Nagel called โwhat it is likeโ to be a conscious organism [41]1 ; access consciousness, by contrast, con- cerns the availability of information for reasoning, planning, memory, verbal report, and behavioral control, and is defined functionally, admitting computational description far more readily than phenomenal consciousness. Structural Semantics intersects Blockโs framework at a carefully delimited point: the Scope Calculus is not a theory of phenomenal consciousness, nor a complete theory of access consciousness, but investigates a third mathematical object, the geometry of admissible semantic transformations. Information accessibility may facilitate structural semantic organization, but structural semantic organization is defined independently, through the preservation of admissible boundaries (definition 1.6), so that the present framework neither requires nor denies phenomenal consciousness and does not require any particular cognitive architecture for access consciousness. 1 The expression itself predates Nagelโs 1974 essay: it appears in B. A. Farrellโs 1950 paper Experience, and earlier still in Wittgensteinโs remarks of 1946โ47 and Russellโs 1926 Encyclopaedia Britannica entry on psy- chology. Nagelโs essay is credited here, as it standardly is, for making the phrase central to the philosophy of mind and the explanatory gap, not for originating it.
2.12 Predictive Processing and the Free Energy Principle During the early twenty-first century, cognitive science increasingly modeled intelli- gence as a continuous process of prediction, error correction, and dynamical adapta- tion. Predictive Processing, developed by Clark, Hohwy, and others [11, 28], and the closely related Free Energy Principle proposed by Friston [18, 19], both treat cognition as active inference whose objective is the maintenance of coherent internal models un- der uncertain sensory information, describing cognition through probabilistic inference, dynamical systems, and Bayesian optimization rather than explicit logical manipulation. Perception is understood not as passive reception but as continual generation of predic- tions compared against observations, with cognition inferring the posterior ๐ (๐ฅ โฃ ๐ฆ) over latent causes ๐ฅ given observations ๐ฆ. Friston generalizes this into the minimization of an upper bound on Bayesian surprise called variational free energy,
๐น (๐) = ๐ทKL (๐(๐ฅ) โ ๐ (๐ฅ โฃ ๐ฆ)) โ log ๐ (๐ฆ),
where ๐(๐ฅ) is an internal recognition density approximating the true posterior; mini- mizing ๐น approximately minimizes surprise while improving the accuracy of the inter- nal generative model, unifying perception, learning, action, and homeostatic regulation. Active Inference extends the framework so that an organism may also act on its envi- ronment to reduce prediction error, replacing stimulus-response models with a closed feedback loop between organism and environment. Predictive Processing unifies perception, learning, memory, and motor control within a common inferential framework, accommodates uncertainty naturally, scales to hierar- chical architectures, and admits quantitative modeling; but although it explains how internal models evolve under prediction error, it provides comparatively little formal machinery describing the organization of semantic structure itself, since prediction error measures numerical discrepancy rather than the admissibility of contextual transforma- tions or the preservation of semantic boundaries. Structural Semantics shares with Pre- dictive Processing an emphasis on internal organization and stability under continual transformation, but its primary mathematical object differs: Predictive Processing mini- mizes variational free energy over probability distributions, while Structural Semantics studies admissible transformations over bounded scope geometries, with prediction er- ror becoming only one possible signal influencing boundary operations rather than the defining quantity itself. The two approaches are therefore potentially complementary, a relationship revisited systematically in Chapter 16.
2.13 Mechanistic Interpretability The rapid success of transformer architectures introduced a new scientific challenge: while large language models demonstrated remarkable behavioral competence, the in- ternal mechanisms producing this behavior remained largely opaque. Mechanistic in- terpretability emerged as a research program seeking causal explanations grounded di- rectly in the internal organization of the model, rather than treating neural networks as inscrutable black boxes evaluated only on input-output performance. Modern transformer architectures consist of repeated blocks of multi-head attention, feed-forward networks, normalization, and residual connections [55]; if the residual stream at layer ๐ is x๐ โ โ๐model , the forward computation is the iterative transformation x๐+1 = ๐๐ (x๐ ), forming a trajectory through a high-dimensional representation space rather than a sequence of explicitly symbolic operations. Recent work by Olah, Elhage,
Nanda, and collaborators suggests that many such computations can be understood in terms of interacting computational circuits โ induction heads, copy circuits, name- mover circuits, arithmetic circuits, factual retrieval pathways, and long-range attention mechanisms among them [15, 42, 44] โ indicating that distributed neural computation often possesses substantial internal organization despite the absence of explicit symbolic programming. Techniques including activation patching, causal tracing, feature visual- ization, sparse autoencoders, attention intervention, linear probing, and circuit analysis [15, 38, 44] attempt to identify stable computational features whose causal manipulation produces predictable behavioral changes, with the overall objective explanatory rather than merely descriptive. Despite substantial progress, current interpretability techniques remain fragmented: many methods identify individual features, neurons, or circuits without a unified math- ematical description of contextual organization across an entire computational trajec- tory, and no generally accepted algebra currently exists for describing admissible se- mantic transformations across changing internal representations. Structural Semantics builds on the objectives of mechanistic interpretability while proposing a different ab- straction: individual neurons, attention heads, sparse features, or circuits are not them- selves treated as semantic units, but constitute one possible substrate whose state transi- tions may be projected onto bounded scope geometries, so that the primary mathemat- ical object shifts from individual network components to admissible transformations preserving contextual organization. This motivates the Representation Invariance The- orem of Chapter 9, which establishes conditions under which different computational substrates may be considered structurally equivalent within the Scope Calculus.
2.14 Historical Synthesis The historical survey reveals three broad explanatory traditions. The first emphasizes subjective experience as the defining feature of understanding, including phenomenol- ogy, portions of process philosophy, and accounts centered on consciousness or inten- tionality. The second emphasizes functional organization, identifying understanding through behavioral competence or successful interaction with the environment, as in cybernetics, functionalism, and contemporary AI benchmarks. The third emphasizes structural organization, investigating the internal organization of representations and the transformations that preserve them; although elements of this viewpoint appear throughout structural linguistics, information theory, category theory, and mechanistic interpretability, no single unified mathematical framework currently develops this ap- proach systematically, and the present monograph is situated within this third tradition. Existing theories successfully explain many individual aspects of semantic organi- zation โ structural linguistics describes relational organization within languages, cate- gory theory studies structure- preserving mappings, predictive processing models adap- tive inference, and mechanistic interpretability analyzes internal computation โ but none develops a unified algebra describing how semantic contexts themselves trans- form while preserving admissible organizational constraints. This gap motivates the formal developments beginning in Part II. Structural Semantics does not seek to replace the traditions surveyed in this chapter, but to synthesize insights from several of them: the primacy of relational organization from structural linguistics, the study of structure- preserving transformations from category theory, the emphasis on dynamic evolution from process philosophy, the importance of continual adaptation from predictive pro- cessing, and the commitment to analyzing internal computational organization from
mechanistic interpretability.
Status of Results: Chapter 2 Established background. Historical developments in structural linguistics, process philosophy, philosophy of mind, predictive processing, active inference, and mechanis- tic interpretability [7, 11, 13, 15, 18, 36, 42, 50, 53, 55, 56].
Formal developments. A comparative historical synthesis identifying three major ex- planatory traditions and motivating the need for a formal theory of Structural Seman- tics.
Empirical hypotheses. None. This chapter is historical and conceptual, providing the intellectual foundations for the formal developments beginning in Chapter 3.
Chapter 3
Three Definitions of Understanding
Every definition illuminates one aspect of a phenomenon while obscuring another. The scientific task is therefore not merely to define, but to identify which definitions remain appropriate for which explanatory questions.
The preceding chapters established two complementary foundations. Chapter 1 iso- lated the epistemic domain of Structural Semantics by separating semantic capacity from phenomenal consciousness, truth, agency, and syntactic fluency; Chapter 2 sur- veyed the historical development of theories of meaning, cognition, and representation from structural linguistics through mechanistic interpretability. The historical record reveals that disagreements concerning artificial intelligence frequently arise because the word understanding is used to denote fundamentally different properties, so that researchers often disagree not because they evaluate the same property differently, but because they evaluate different properties under a common name. The objective of this chapter is therefore taxonomic rather than polemical: rather than arguing that one defi- nition is correct and the others mistaken, we distinguish three conceptually independent notions of understanding that have emerged throughout the literature โ phenomeno- logical, functional, and structural โ each capturing a genuine explanatory question, each admitting different methodologies, and each possessing different mathematical re- quirements.
3.1 Motivating the Taxonomy Throughout the history surveyed in Chapter 2, philosophers have asked whether ma- chines understand language, whether symbolic manipulation is sufficient for intelli- gence, whether consciousness is necessary for meaning, and whether behavioral com- petence implies semantic knowledge. Although these questions appear closely related, they address distinct levels of explanation: some concern subjective experience, others observable behavior, still others the internal organization of representations. Asking whether an artificial system โunderstandsโ without specifying the intended definition therefore produces an ill-posed scientific question, and a precise taxonomy becomes a prerequisite for rigorous discussion. The classification introduced below is not an ex- haustive ontology of cognition, but identifies three broad explanatory frameworks that recur throughout contemporary philosophy of mind, cognitive science, and artificial in- telligence.
23
Definition 3.1 (Understanding, neutral form). Let ๐ denote a system operating over a domain ๐. An understanding relation is a mapping ๐ โถ ๐ ร ๐ โ ๐ซ, where ๐ซ denotes a collection of explanatory properties whose interpretation depends on the theoretical framework under consideration.
The codomain ๐ซ is intentionally left abstract: for phenomenological theories it con- cerns subjective experience, for functional theories behavioral competence, and for Struc- tural Semantics the admissible preservation of contextual organization. This neutrality allows the three theories to be compared without presupposing the correctness of any one of them. The three definitions that follow represent distinct research programs rather than competing philosophical dogmas, and are not mutually exclusive: a biological organism may satisfy all three, an artificial system may satisfy one or two, and the objective here is clarification of explanatory questions rather than classification of particular systems. This taxonomy functions as the conceptual interface between the historical literature of Chapter 2 and the mathematical framework of Chapters 6 through 10; without it, later formal definitions of Scope Geometry and structural semantic capacity would remain disconnected from the broader philosophical literature.
3.2 Phenomenological Understanding The oldest and perhaps most intuitive conception of understanding identifies it with con- scious experience itself: not merely the successful manipulation of symbols or the pro- duction of appropriate behavioral responses, but something inseparable from the exis- tence of a first-person perspective through which meaning is directly experienced. This perspective appears throughout Husserlโs phenomenology, Heideggerโs existential anal- ysis, Merleau-Pontyโs embodied cognition, Whiteheadโs process philosophy, Nagelโs analysis of subjective experience, Searleโs critique of computational intelligence, Chalmersโ formulation of the hard problem, and Blockโs distinction between phenomenal and ac- cess consciousness [7, 10, 24, 30, 39, 41, 50, 56]; despite differences in terminology and metaphysical commitment, these theories converge on the claim that understanding is inseparable from phenomenal experience.
Definition 3.2 (Phenomenological understanding). Let ๐ be a cognitive system oper- ating over a semantic domain ๐. The system possesses phenomenological understanding over ๐ if and only if successful semantic activity is accompanied by subjective phenom- enal experience, formally ๐phen โถ ๐ ร ๐ โ ฮฆ, where ฮฆ denotes the domain of phenom- enal conscious experience. The precise ontology of ฮฆ is left unspecified, as different phenomenological traditions provide different accounts of subjective experience.
Phenomenological theories generally hold that semantic understanding is intrinsi- cally first-person, that meaning is not exhausted by behavioral competence, and that subjective experience is explanatorily fundamental rather than derivative; consequently they distinguish appearing intelligent from genuinely understanding, and their best- known contemporary defense is Searleโs Chinese Room, discussed already in Chapter 2. Within Whiteheadโs process philosophy, phenomenology enters the formal structure through Subjective Form, ๐ด, the manner in which an occasion experiences its own be- coming, formalized fully in Chapter 4. These theories capture an undeniable feature of ordinary experience and raise ques- tions that remain largely unresolved by current neuroscience and artificial intelligence,
3.3. FUNCTIONAL UNDERSTANDING 25
but they encounter a substantial methodological difficulty when applied to computa- tional systems: because phenomenal experience is intrinsically first-person, no generally accepted procedure exists for measuring it objectively, so that if subjective experience is adopted as part of the definition of understanding, any system lacking independently established phenomenology is excluded by definition rather than by empirical investiga- tion. This observation does not demonstrate that artificial systems lack phenomenology; it indicates only that phenomenological theories alone provide no operational criterion by which the question can be resolved, an observation developed into a general argu- ment in Chapter 5. The present monograph neither accepts nor rejects phenomenolog- ical accounts, isolating them instead as one legitimate explanatory framework among several, and asks in the chapters that follow whether semantic organization can also be characterized independently through formal structural properties.
3.3 Functional Understanding A second tradition identifies understanding not with subjective experience but with what a system can reliably do, treating the internal substrate responsible for cognition as secondary to whether the system exhibits appropriate causal organization, behav- ioral competence, adaptive capacity, or information-processing ability. This perspective has profoundly influenced cybernetics, cognitive science, artificial intelligence, robotics, and computational neuroscience, forming the dominant methodological framework for modern AI research [16, 43, 48, 54].
Definition 3.3 (Functional understanding). Let ๐ be a system operating over a seman- tic domain ๐. The system possesses functional understanding if its observable behavior satisfies the operational requirements associated with successful performance over ๐, formally ๐func โถ ๐ ร ๐ โ โฌ, where โฌ denotes an appropriate space of behavioral or computational performance measures. Unlike phenomenological understanding, no as- sumptions are made concerning subjective experience or phenomenal awareness.
Turing proposed replacing metaphysical questions concerning thought with oper- ational tests based on conversational behavior [54]; cybernetics emphasized feedback, control, and adaptive behavior [3, 57]; and Putnam and Fodor developed forms of func- tionalism identifying mental states with their causal roles rather than their physical re- alization [16, 48], naturally supporting the possibility of multiple realizability, in which distinct physical systems instantiate the same cognitive organization. Most contempo- rary evaluations of large language models remain largely functionalist, measuring ques- tion answering, mathematical reasoning, code generation, planning, factual recall, and conversational competence through observable outputs rather than subjective experi- ence, a methodology that has enabled rapid empirical progress because behavioral met- rics admit objective measurement and direct comparison across systems. Functional theories provide operational definitions suitable for empirical investiga- tion, remain largely independent of metaphysical assumptions concerning conscious- ness, naturally accommodate substrate independence, and rely on criteria that are re- producible and experimentally testable. However, behavior alone does not uniquely de- termine internal organization: distinct computational mechanisms may produce nearly identical external behavior while employing fundamentally different internal represen- tational structures, so that behavioral equivalence does not necessarily imply structural equivalence. Consider, for instance, two theorem-proving systems, one using symbolic logical deduction and the other a transformer trained on formal proof corpora, which
produce identical formal proofs for every theorem in a benchmark suite; from a func- tional perspective the two exhibit equivalent understanding, yet their internal organiza- tions differ substantially, and functional evaluation alone does not determine whether they preserve semantic context through comparable structural mechanisms. Structural Semantics accepts the functionalist insight that semantic organization need not depend on any particular physical substrate, but proposes that behavioral compe- tence alone is insufficient for a complete mathematical theory of understanding, and instead investigates whether successive internal state transitions preserve bounded se- mantic organization throughout a computation, treating behavior as evidence for un- derstanding rather than as its complete definition.
3.4 Structural Understanding The third conception developed in this monograph begins from a different premise than either the phenomenological or the functional: rather than identifying understanding with subjective experience or observable behavior, it treats understanding as a property of the internal organization of representations and, specifically, of the transformations that preserve them. This does not deny the importance of consciousness or reject behav- ioral evaluation; it isolates a third object of scientific investigation whose mathematical properties can be studied independently of both. The central question is therefore not what a system experiences, nor how successfully it behaves, but which structural rela- tions remain invariant as it transforms its internal representations โ a shift that moves the study of semantic organization from psychology toward mathematics.
Definition 3.4 (Structural understanding). Let ๐ be a computational or cognitive sys- tem operating over a semantic domain ๐. The system possesses structural understanding if there exists a family of admissible transformations ๐ฏ = {๐๐ } acting on bounded scope geometries such that every transformation preserves the admissibility constraints gov- erning contextual organization; equivalently, ๐struct โถ ๐ ร ๐ โ Scope, where semantic capacity is evaluated through admissible morphisms in the category Scope rather than through phenomenal experience or behavioral observation.
The defining property of structural understanding is invariance: although internal representations may change continuously throughout computation, certain organiza- tional relationships remain preserved, and these invariant relationships constitute the semantic content of the representation, precisely characterized in Parts II and III. Struc- tural Semantics proposes that semantic organization is fundamentally a problem of scope preservation โ of maintaining coherent contextual organization while new in- formation is introduced, previous assumptions are revised, and nested contexts evolve โ which motivates the bounded scope geometries of Chapter 6 and the primitive op- erations of Chapter 7. A defining feature of the framework is its independence from particular representational substrates: whether information is encoded symbolically, neurally, graphically, or through another computational medium is secondary to the induced transformation over bounded scope geometries, a principle formalized later as the Representation Invariance Theorem (Theorem 9.7). Consider again two reasoning systems proving the same mathematical theorem, one by symbolic deduction and the other by continuous optimization over neural activation vectors: although their internal representations differ radically, both may maintain iden- tical assumptions, preserve identical logical dependencies, introduce hypotheses in the
same admissible order, and discharge those assumptions according to equivalent struc- tural rules. From the perspective of Structural Semantics, both systems instantiate the same semantic organization despite their differing computational substrates, since their semantic equivalence depends on preserved contextual structure rather than on identi- cal internal encodings.
3.5 Comparing the Three Conceptions The three conceptions of understanding introduced in this chapter may now be com- pared directly. Phenomenological understanding asks whether semantic activity is ac- companied by subjective experience; functional understanding asks whether seman- tic activity produces appropriate observable behavior; structural understanding asks whether semantic activity preserves the organization of contextual constraints through- out computation. These questions are logically independent: a biological organism may satisfy all three, a theorem prover may satisfy the latter two while remaining silent with respect to phenomenology, and a hypothetical conscious system suffering severe cog- nitive impairment might possess phenomenology while exhibiting poor functional or structural organization. The taxonomy therefore separates explanatory questions rather than ranking competing theories.
Framework Primary object Evaluation Representative literature Phenomenological Subjective experi- First-person Husserl, Heidegger, ence awareness Whitehead, Nagel, Chalmers Functional Behavioral com- Observable per- Turing, Putnam, Fodor, petence formance Newell, Simon Structural Contextual orga- Admissible trans- Scope Calculus, category nization formations theory, mechanistic inter- pretability
Table 3.1: Three complementary conceptions of understanding.
3.6 Chapter Summary This chapter established a conceptual taxonomy separating three distinct research pro- grams concerning understanding: phenomenological theories identify understanding with subjective experience, functional theories with successful behavioral organization, and Structural Semantics introduces a third conception in which semantic capacity is identified with the preservation of admissible contextual organization under transfor- mation. The remainder of this monograph develops this third perspective systemati- cally, beginning in Chapter 4 with a mathematical reconstruction of Whiteheadโs process ontology that later permits comparison with the Scope Calculus through the translation functor of Chapter 10.
Status of Results: Chapter 3 Established background. Phenomenological, functional, and structural traditions in philosophy of mind, cognitive science, and artificial intelligence [7, 50, 54, 56].
Formal developments. Definitions of Phenomenological Understanding, Functional Understanding, and Structural Understanding, together with their comparative taxon- omy.
Empirical hypotheses. Structural semantic capacity constitutes an independently mea- surable property that is neither reducible to subjective phenomenology nor fully cap- tured by behavioral performance alone. Subsequent chapters develop the formal math- ematical machinery necessary to evaluate this hypothesis.
Chapter 4
The Whiteheadian Framework
The many become one, and are increased by one. โ Alfred North Whitehead, Process and Reality
The preceding chapter established that discussions of understanding separate into phenomenological, functional, and structural frameworks. This chapter returns to one of the most influential systems within the phenomenological tradition: Whiteheadโs process philosophy. Although the present monograph ultimately develops a different formal framework, Whiteheadโs emphasis on dynamic organization, relational struc- ture, and process makes his work unusually compatible with modern mathematical ap- proaches to computation, and this chapter is accordingly presented not as a critique but as a careful mathematical reconstruction. Its objective is twofold: to provide precise formal definitions of Whiteheadโs principal concepts using contemporary notation, and to identify exactly which components admit structural formalization and which remain explicitly phenomenological, preparing the translation functor developed in Chapter 10.
4.1 Process Rather Than Substance Classical metaphysics generally treats enduring substances as the primary objects of reality, with processes and changes regarded as properties or modifications of an un- derlying entity. Whitehead reversed this relationship: reality consists fundamentally of events, and objects are abstractions constructed from sufficiently stable patterns within those events. If processes are primitive, transformation becomes more fundamental than state and relations more fundamental than isolated entities, an orientation shared by category theory, dynamical systems, network theory, and process algebra, though Whitehead developed his system decades before these fields matured.
4.2 Actual Occasions The primitive object of Whiteheadโs ontology is the actual occasion: not a material object but a single completed event of becoming, from which all larger structures โ organisms, mountains, societies, even physical particles โ are understood as persistent patterns generated by vast collections of interconnected occasions. Definition 4.1 (Actual occasion). An actual occasion is represented by an ordered tuple ๐ธ = โจ๐ท, ๐ , ๐โฉ, where ๐ท denotes the collection of antecedent data inherited from prior
29
occasions, ๐ denotes the collection of prehensions performed on that data, and ๐ denotes the completed satisfaction state produced after concrescence.
This definition is intentionally schematic; the remaining sections formalize each com- ponent individually.
4.3 Prehension Whitehead replaces classical perception with the broader concept of prehension, the fun- damental relational operation through which one occasion incorporates aspects of pre- vious occasions into its own becoming. Unlike ordinary perception, prehension is not restricted to conscious organisms: every actual occasion prehends antecedent reality, so that prehension functions as the primitive relational operator of Whiteheadโs ontology.
Definition 4.2 (Prehension). A prehension is represented by the ordered triple ๐ = โจ๐ท, ๐ฃ, ๐ดโฉ, where ๐ท denotes the datum being prehended, ๐ฃ โ {+, โ} denotes the valu- ation polarity, with positive valuation indicating inclusion and negative valuation indi- cating exclusion, and ๐ด denotes the Subjective Form associated with the prehension.
Whitehead distinguishes positive prehensions, which contribute directly to the for- mation of the new occasion, from negative prehensions, which exclude data from fur- ther participation. Formally, if ๐ = {๐1 , โฆ , ๐๐ } denotes the available antecedent data, the process of becoming partitions this collection into disjoint subsets ๐ = ๐ + โช ๐ โ with ๐ + โฉ ๐ โ = โ : positive prehensions contribute to the evolving occasion, negative prehensions remove information from subsequent construction. This distinction will later admit a natural comparison with the โฌ๐ and โ operations of the Scope Calculus, although no such identification is assumed at this stage.
4.4 Concrescence The collection of individual prehensions does not remain an unordered set; instead it undergoes a process of integration Whitehead calls concrescence, transforming many in- herited influences into one completed occasion, symbolically {๐1 , โฆ , ๐๐ } โถ ๐, sum- marized in Whiteheadโs own phrase: โthe many become one, and are increased by one.โ From a mathematical perspective, concrescence resembles an aggregation operator act- ing on a structured family of relational inputs; precisely which algebraic properties such an operator possesses is a question examined later through the Scope Calculus.
4.5 Subjective Form Among the components of Whiteheadโs ontology, Subjective Form occupies a special position. Whereas antecedent data and valuation polarity describe observable struc- tural relations, Subjective Form describes the manner in which the occasion experiences its own process of becoming, and unlike positive and negative prehension it is not in- troduced as an externally observable property but concerns the intrinsic character ac- companying each prehension. Consequently Subjective Form functions as the explicitly phenomenological component of Whiteheadโs ontology; the present monograph seeks to determine which components of his system admit purely structural reconstruction and which remain irreducibly phenomenological, preserving Subjective Form explicitly
rather than eliminating it, while carefully distinguishing it from the structural opera- tions that surround it.
4.6 Chapter Summary This chapter reconstructed the fundamental concepts of Whiteheadโs process ontology using contemporary mathematical notation: actual occasions as primitive events of be- coming; prehensions as ordered triples of antecedent data, valuation polarity, and Sub- jective Form; positive and negative prehensions as an explicit set-theoretic decomposi- tion; and concrescence as the aggregation process transforming many prehensions into one completed satisfaction state. Most importantly, Subjective Form was isolated as the explicitly phenomenological component of the framework. The following chapter exam- ines this distinction in detail, showing that Subjective Form functions as an independent axiom within Whiteheadโs ontology and forms the critical bridge between phenomeno- logical theories of understanding and the structural framework developed throughout the remainder of the monograph.
Status of Results: Chapter 4 Established background. Whiteheadโs process ontology, actual occasions, prehensions, concrescence, and Subjective Form [56].
Formal developments. Mathematical reconstruction of actual occasions, prehensions, valuation polarity, and concrescence using explicit definitions suitable for later categor- ical translation.
Empirical hypotheses. None. This chapter reconstructs an existing philosophical frame- work in preparation for the structural analysis developed in Chapters 5โ10.
Chapter 5
The Phenomenological Boundary
Everything is vague to a degree you do not realize till you have tried to make it precise. โ Bertrand Russell
The preceding chapter reconstructed Whiteheadโs process ontology without either criticism or reinterpretation, identifying its primitive components in their own terms and observing that the architecture naturally separates into structural entities โ ac- tual occasions, prehensions, valuation polarity, and concrescence, which admit explicit mathematical representation โ and Subjective Form, introduced not as an externally observable relation but as the intrinsic manner in which an occasion experiences its own becoming. The present chapter examines the consequences of this distinction. The question is not whether Subjective Form exists, nor whether Whiteheadโs metaphysics is correct, but methodological: if Subjective Form is treated as an indispensable prereq- uisite for semantic capacity, the evaluation of artificial systems becomes independent of any mathematical or empirical investigation, and the conclusion that such systems lack understanding follows directly from the initial definition rather than from subse- quent analysis. The purpose of this chapter is to isolate the precise logical boundary separating structural questions from phenomenological axioms, so that the mathemat- ical framework developed in subsequent chapters can be evaluated independently of broader metaphysical commitments.
5.1 Structural and Phenomenological Components The reconstruction of Chapter 4 naturally partitions Whiteheadโs ontology into two classes. The first consists of objects whose behavior may be described through explicit structural relations: ๐ท, ๐, ๐ฃ, and ๐ each admit mathematical description, whether as sets, graphs, tensors, categories, discrete operators, or aggregation functionals, and each possesses an externally describable operational role. The second class contains only Subjective Form, ๐ด: unlike every other component, it is not introduced through observable trans- formation rules but denotes the intrinsic character accompanying each prehension, so that its inclusion within the theory is fundamentally different from every other primitive introduced thus far. Definition 5.1 (Structural component). A primitive object is structural if its contribution to the behavior of a system is completely specified by externally observable relations and transformation laws.
33
Definition 5.2 (Phenomenological component). A primitive object is phenomenological if its defining properties concern intrinsic experience rather than externally observable structural relations.
Under these definitions ๐ท, ๐, ๐ฃ, and ๐ are structural components, while ๐ด is phe- nomenological. The distinction is purely methodological: it does not deny the existence of phenomenological properties, but identifies which primitives participate directly in the structural mathematics developed throughout the remainder of the monograph.
5.2 Axiomatic Dependence The central logical observation of this chapter concerns the role played by Subjective Form within theories of understanding. Suppose semantic capacity is defined by the predicate Understand(๐ ). If one adopts the axiom ๐ด โ โ โ Understand(๐ ) together with its converse ๐ด = โ โ ยฌ Understand(๐ ), then the semantic status of every system is determined immediately by the presence or absence of Subjective Form, and no sub- sequent behavioral experiment, mathematical analysis, or empirical measurement can alter this conclusion. Observation 5.3. If Subjective Form is introduced as a necessary condition for semantic capacity, then every system lacking Subjective Form is excluded by definition rather than by empirical investigation. This observation concerns logical dependence rather than philosophical correctness; one may certainly choose such an axiom, but the resulting theory belongs to a different methodological category from one whose conclusions depend on mathematical or em- pirical evidence.
5.3 Definitions and Empirical Criteria Scientific theories generally distinguish between definitions, which specify the concepts under consideration, and measurements, which determine whether particular systems satisfy those definitions; when these two roles are conflated, empirical investigation becomes impossible. If one defines intelligence as human intelligence, no experiment comparing humans and machines could demonstrate machine intelligence, since the conclusion follows directly from the definition; similarly, if semantic capacity is de- fined as requiring Subjective Form, every non-conscious computational architecture is excluded before its internal organization is examined. Structural Semantics adopts a different strategy, defining semantic capacity through structural preservation and treat- ing whether phenomenology accompanies such preservation as a distinct philosophical question.
5.4 The Structural Research Program Separating structural organization from phenomenology produces a well-defined math- ematical research program whose central question is which structural invariants must be preserved for semantic organization to remain stable. Unlike questions concerning subjective experience, this question admits explicit mathematical treatment: one may de- fine admissible transformations, characterize boundary operators, prove preservation theorems, derive computational diagnostics, and compare different representational
architectures through common structural invariants. Consequently the mathematical framework developed throughout Volumes II and III does not depend on resolving the philosophical debate concerning consciousness, studying instead the preservation of structured relational organization. This methodological separation is compatible with, while differing from, several ex- isting traditions: classical functionalism evaluates systems primarily through behavioral performance [16, 48], phenomenological theories through subjective experience or in- trinsic intentionality [41, 50], and mechanistic interpretability through internal compu- tational organization [15, 44]. Structural Semantics occupies a different position, iden- tifying semantic capacity with neither behavior nor phenomenology but investigating whether semantic organization can be characterized by mathematically definable struc- tural invariants, an orientation that motivates the algebraic development beginning in Chapter 6.
5.5 Chapter Summary This chapter identified the precise methodological boundary separating phenomeno- logical and structural theories of semantic capacity, distinguishing structural primitives from phenomenological primitives and showing that Subjective Form occupies a unique position within Whiteheadโs ontology: theories requiring Subjective Form as an axiom classify machine understanding by definition rather than by empirical investigation. Structural Semantics adopts a complementary research program, developing a mathe- matical theory of structural organization whose predictions may be evaluated indepen- dently of phenomenological commitments. This completes Volume I. Beginning with Chapter 6, the monograph transitions from philosophical analysis to the formal mathe- matical development of Scope Geometry.
Status of Results: Chapter 5 Established background. Whiteheadโs distinction between structural process and Sub- jective Form, together with classical debates concerning consciousness and intentional content [41, 50, 56].
Formal developments. Definitions of structural and phenomenological components, together with the logical separation between axiomatic phenomenological theories and structural mathematical theories.
Empirical hypotheses. None. This chapter establishes methodological distinctions that motivate the formal development of Scope Geometry in Volume II.
Part II
Scope Geometry, Algebra, and Categories
37
Chapter 6
Scope Geometry
The art of reasoning consists in getting hold of the subject at the right end, of seizing the few general ideas that illuminate the whole, and of persistently organizing all subsidiary facts around them. โ Alfred North Whitehead
Volume I established the philosophical and epistemic foundations of Structural Se- mantics. Its principal conclusion was not that phenomenology is false or unnecessary, but that the mathematical analysis of semantic organization can proceed independently of questions concerning subjective experience. Having isolated the structural domain, we now begin constructing the mathematical objects that will carry the remainder of the theory. The central claim of this volume is that semantic organization is fundamentally geo- metric. Every computational system, biological or artificial, operates within collections of constraints that determine which transformations preserve coherence and which pro- duce contradiction or collapse. These collections of constraints are neither arbitrary sets nor merely symbolic contexts: they possess internal organization, hierarchical nesting, admissible boundaries, and lawful mechanisms for interaction. This chapter introduces the mathematical object that captures these properties, the bounded scope geometry, which provides the primitive space on which the Scope Calculus of subsequent chapters oper- ates. Primitive operations such as Pop, Refuse, Bind, and Collapse will later be defined as transformations acting on these geometries (Chapter 7), while Chapters 8โ10 estab- lish their categorical, algebraic, and functorial properties. Throughout this chapter we deliberately avoid introducing operational rules, defining first the spaces on which op- erations act before studying the operations themselves โ mirroring the development of modern topology and differential geometry, where spaces are characterized indepen- dently of the mappings subsequently defined between them. Remark 6.1 (Terminological note). The formalism developed in this and the following four chapters was originally introduced and published under the name Spherepop. The present exposition is self-contained and does not presuppose that earlier work; it is re- named and reorganized here as the Scope Calculus to foreground its role within Struc- tural Semantics rather than its original programming-language framing. Readers al- ready familiar with Spherepop will recognize ๐ซ, โ, โฌ๐ , and ๐๐ as the same four primi- tives under new emphasis; readers meeting the material for the first time can treat the two names as synonymous.
39
6.1 Motivation Nearly every formal treatment of semantics assumes, explicitly or implicitly, the ex- istence of context. Natural language interpretation, programming languages, logical deduction, theorem proving, biological regulation, and distributed computation all de- pend on determining which objects belong to a given context and which transformations preserve that membership; yet context is rarely treated as a mathematical object in its own right. In linguistics it is often represented informally through discourse structure or lexical fields [13, 36]; in formal logic it appears as assumptions within derivations [20]; programming-language semantics typically represents it through environments, symbol tables, or typing judgments [47, 58]; and category theory studies morphisms between objects while generally leaving contextual admissibility to external structure [37]. These approaches have been extraordinarily successful within their respective do- mains, but none was developed specifically to study semantic stability under arbitrary transformations across multiple representational substrates. Structural Semantics therefore introduces an explicit geometry of contexts. Rather than viewing context merely as auxiliary information attached to a computation, we regard context itself as the primary mathematical object: semantic transformations be- come transformations of context geometries, and semantic preservation becomes the preservation of admissible geometric structure. This shift is analogous to the historical transition from Euclidean coordinate calculations to differential geometry, in which ge- ometry became primary while coordinates became representations; likewise, scope ge- ometry treats context as primary while individual representations become coordinates within that context.
6.2 Primitive Intuition Before introducing formal definitions, it is useful to develop geometric intuition. Con- sider a mathematical proof: at any stage there exists a collection of assumptions, def- initions, variables, and established propositions that remain valid, forming a coherent environment in which introducing an undefined symbol, violating an assumption, or ap- plying a theorem outside its domain produces inconsistency. Consider similarly lexical interpretation: a discussion of planetary motion establishes a context in which the word โorbitโ possesses one meaning, while a discussion of computing establishes a different context in which the same lexical item acquires an entirely different interpretation, and the semantic validity of subsequent expressions depends on remaining inside the appro- priate context unless an explicit contextual transition occurs. Programming languages exhibit the same phenomenon: variables exist only within declared scopes, references outside these scopes become invalid, and nested functions inherit some contextual in- formation while introducing new local constraints. Although these examples originate in different disciplines, they share a common structural pattern: each consists of a bounded region of admissible entities, explicit constraints governing admissibility, hierarchical nesting, and lawful transitions between nested regions. Scope Geometry abstracts exactly these common structural features.
6.3 Why Boundaries Are Fundamental The formal definitions beginning in the next section take a boundary, not a symbol, a vec- tor, or an entity, as the primitive notion from which everything else is built. This choice
6.4. STATE DOMAINS 41
is worth motivating before the definitions arrive, since it is not the only one available: a theory could equally well begin with a space of representations and add constraints afterward, treating admissibility as a derived property rather than a starting point. The reason for beginning with boundaries instead is that the examples of the preced- ing section share nothing at the level of representation. A proofโs assumptions, a wordโs discourse context, and a variableโs lexical scope have no common representational for- mat โ one is a set of propositions, another a topic under discussion, the third a region of program text. What they share is only the pattern of admissibility itself: a bounded re- gion within which certain moves are licensed and others are not, independently of what occupies that region. A theory that started from representations would need a separate account of admissibility for every representational format it encountered. A theory that starts from the boundary can be stated once and instantiated by whatever representa- tion a given application supplies, which is exactly the substrate independence pursued formally in Chapter 9. This also explains why the definitions that follow introduce โฌ before โฐ rather than the reverse. The interior of a scope, Definition 6.5 below, is defined as the states satisfying the boundary, so that the boundary is prior in the order of definition as well as in the order of motivation: there is no admissible interior to speak of until the boundary that determines it has already been specified.
6.4 State Domains Every computational or cognitive system evolves through collections of states. We there- fore begin with the underlying state domain. Definition 6.2 (State domain). A state domain is a nonempty set ๐ whose elements rep- resent admissible configurations of a computational, logical, or cognitive system. The internal structure imposed on ๐ depends on the application: for symbolic rea- soning it may consist of logical derivation states, for transformer architectures ๐ โ โ๐ may represent activation vectors, and for programming languages it may denote envi- ronments consisting of variable bindings and execution frames. The theory developed here does not depend on the specific realization of ๐; only the existence of distinguish- able states is assumed.
6.5 Boundaries State domains alone are insufficient. Semantic organization requires constraints deter- mining which transitions remain valid; these constraints are represented by boundaries. Definition 6.3 (Boundary). Let ๐ be a state domain. A boundary โฌ is a collection of admissibility predicates โฌ = {๐ฝ๐ }๐โ๐ผ , where each ๐ฝ๐ โถ ๐ โ {0, 1} determines whether a given state satisfies a particular structural constraint. Boundaries need not be purely logical: they may represent semantic consistency, typing constraints, physical conservation laws, security permissions, resource limits, or probabilistic admissibility criteria, and the abstraction intentionally encompasses all such cases. A boundary in this sense also admits a topological reading, in which it sepa- rates admissible from inadmissible regions of a semantic space in exactly the sense that a topological boundary separates a region from its complement; this reading is developed rigorously in Appendix E.
6.6 Interior Spaces Having introduced boundaries, we define the region they enclose.
Definition 6.4 (Interior space). Given a state domain ๐ and boundary โฌ, the associated interior space is โฐ = {๐ฅ โ ๐ โถ ๐ฝ(๐ฅ) = 1 for every ๐ฝ โ โฌ}.
Thus โฐ contains precisely those states satisfying every active admissibility constraint; states outside โฐ may still exist within the global domain ๐, but they are not locally admissible within the current scope.
6.7 Nested Scope Geometries Contexts rarely exist in isolation. Most systems construct hierarchies: programming languages contain nested blocks, logical proofs contain nested assumptions, human dis- course continually opens and closes subordinate contexts, and transformer attention similarly constructs localized contextual substructures. To model this hierarchy we in- troduce nested scope geometries.
Definition 6.5 (Bounded scope geometry). A bounded scope geometry is the ordered tuple S = โจโฌ, โฐ, ๐ฎchild โฉ, where โฌ is a boundary, โฐ is the associated admissible interior, and ๐ฎchild is a finite or countable collection of child scope geometries.
The recursive appearance of ๐ฎchild permits arbitrarily deep hierarchical structures while preserving a uniform mathematical description, and this recursive definition will later permit induction over scope depth and the derivation of decomposition theorems in Chapter 9.
6.8 Hierarchy, Containment, and Inheritance The recursive definition of bounded scope geometries naturally induces a hierarchical organization: rather than existing as isolated objects, scopes form partially ordered con- tainment structures in which local contexts inherit constraints from larger environments while introducing additional local restrictions. This hierarchical organization appears across numerous disciplines: block-structured programming languages inherit variable environments from enclosing functions while introducing local bindings [47], formal proof systems inherit assumptions from outer derivations while temporarily extending them with auxiliary hypotheses [20], and linguistic discourse similarly maintains a per- sistent conversational context while permitting nested quotations, counterfactuals, and hypothetical reasoning [13, 36]. The purpose of Scope Geometry is to provide a com- mon mathematical framework for each of these situations.
Definition 6.6 (Parent and child scope). Let S = โจโฌ, โฐ, ๐ฎchild โฉ be a bounded scope ge- ometry. If Sโฒ โ ๐ฎchild , then S is the parent scope of Sโฒ and Sโฒ is a child scope of S.
Parenthood defines only an immediate containment relation; ancestor and descen- dant relationships are obtained by repeated application of the parent relation.
Definition 6.7 (Scope containment). Let S1 = โจโฌ1 , โฐ1 , ๐ฎchild 1 โฉ and S2 = โจโฌ2 , โฐ2 , ๐ฎchild 2 โฉ. We write S2 โชฏ S1 if and only if S2 is contained within S1 .
6.8. HIERARCHY, CONTAINMENT, AND INHERITANCE 43
The relation โชฏ is interpreted as geometric containment rather than set inclusion: two scopes may contain overlapping semantic information while remaining distinct objects within the hierarchy, so containment captures organizational structure rather than mem- bership alone. A central property of hierarchical contexts is inheritance: child scopes do not arise in- dependently, but inherit structural constraints established by their parents unless those constraints are explicitly modified through admissible transformations.
Definition 6.8 (Boundary inheritance). Let S2 โชฏ S1 . The boundary of the child scope inherits from the parent if โฌ1 โ โฌ2 ; additional local constraints may be introduced, but inherited constraints remain active unless explicitly discharged.
Boundary inheritance guarantees that entering a child scope cannot erase the struc- tural commitments established by its ancestors; instead, deeper scopes become progres- sively more constrained. This monotonic refinement of admissibility will later prove essential for the decomposition theorems of Chapter 9.
6.8.1 Local and Global Admissibility The distinction between local and global validity is fundamental: a state may be perfectly admissible within one scope while violating the constraints of another. A temporary variable inside a programming block is valid locally but ceases to exist after leaving the block, and an assumption introduced during an indirect proof is admissible only within the subproof in which it was declared. This motivates two complementary notions.
Definition 6.9 (Local admissibility). A state ๐ฅ โ ๐ is locally admissible with respect to a scope S if ๐ฅ โ โฐ.
Definition 6.10 (Global admissibility). A state is globally admissible if it satisfies the boundary constraints of every ancestor scope containing the current scope.
Every globally admissible state is locally admissible, but the converse need not hold.
6.8.2 Boundary Violations Boundary constraints divide state transitions into two classes: some preserve admissi- bility, others produce structural inconsistency.
Definition 6.11 (Boundary violation). Let ๐ฅ โ โฐ and let ๐ โถ ๐ โ ๐ be a state transfor- mation. The transformation is a boundary violation if ๐ (๐ฅ) โ โฐ.
Boundary violations need not correspond to logical contradictions; depending on the application they may represent type errors, inconsistent assumptions, invalid seman- tic references, resource exhaustion, security violations, unstable activation trajectories, or loss of contextual coherence. The Scope Calculus introduced in the next chapter exists precisely to characterize the permissible responses to such violations.
6.8.3 The Scope Hierarchy as a Partially Ordered Structure The containment relation naturally equips the collection of scopes with a partial order.
Proposition 6.12. The relation โชฏ over bounded scope geometries is reflexive, antisymmetric, and transitive.
Proof. Reflexivity follows immediately since every scope contains itself. Antisymmetry follows because if S1 โชฏ S2 and S2 โชฏ S1 , then both scopes possess identical contain- ment relationships and are identified as the same scope geometry. Transitivity follows because containment through an intermediate scope implies containment within every ancestor. Hence (Scope, โชฏ) forms a partially ordered set.
This order will later support induction over scope depth, recursive repair proce- dures, and the categorical constructions of Chapters 8โ10.
6.9 Discussion The geometry introduced thus far deliberately remains static: we have defined the spaces on which semantic organization is represented, but not yet described how those spaces evolve. This distinction mirrors the historical development of geometry itself โ classical topology first characterizes spaces and only subsequently studies continuous mappings between them, and differential geometry first defines manifolds before introducing vec- tor fields and flows. Scope Geometry follows the same methodological order: Chap- ters 6 and 7 separate geometric structure from dynamical evolution, so that once the spaces have been rigorously defined, transformations may be introduced without ambi- guity. In particular, the four primitive operations of the Scope Calculus will be shown to generate every admissible structural transformation required by the framework, pro- viding the dynamical counterpart to the static geometries defined here.
6.10 Chapter Summary This chapter introduced the fundamental geometric objects underlying Structural Se- mantics. Beginning with state domains and boundary constraints, we constructed bounded scope geometries as recursively nested contexts possessing admissible interiors and hi- erarchical organization, defined parent-child relationships, boundary inheritance, local and global admissibility, and established that the collection of scope geometries forms a partially ordered hierarchy under containment. These constructions intentionally re- main independent of any operational mechanism, defining the spaces on which seman- tic organization is represented without prescribing how those spaces evolve. The next chapter introduces the minimal operational language required to manipulate these ge- ometries: the four primitive operations Pop, Refuse, Bind, and Collapse.
Status of Results: Chapter 6 Established background. Hierarchical contexts in formal logic, programming-language semantics, topology, and lexical semantics [20, 36, 37, 47].
Formal developments. Definitions of parent and child scopes, scope containment, bound- ary inheritance, local and global admissibility, boundary violations, and the partial or- dering of scope hierarchies.
Empirical hypotheses. None. This chapter establishes the static geometric framework on which the operational Scope Calculus will be developed in Chapter 7.
Chapter 7
Primitive Operations and the Scope Calculus
Mathematics possesses not only truth, but supreme beauty โ a beauty cold and austere, like that of sculpture. โ Bertrand Russell
Chapter 6 introduced the static geometry underlying Structural Semantics: bounded scope geometries, boundary constraints, hierarchical containment, and admissibility re- lations, without specifying how these objects evolve. The present chapter supplies that missing component by introducing the primitive operations that generate all admissible transformations within Scope Geometry. The guiding principle is one of structural minimality: rather than constructing a large collection of specialized operators, we seek the smallest operational alphabet ca- pable of expressing every admissible transformation required by semantic organization, paralleling how group theory seeks minimal generating sets, formal language theory studies finite alphabets generating infinite languages, and universal algebra investigates operations sufficient to construct entire classes of algebraic structures [9, 29, 37]. The claim advanced in this chapter is that four primitive operations, ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }, are sufficient to describe the evolution of bounded scope geometries. These primitives are defined independently of any particular computational substrate: they are not programming- language instructions, neural-network operations, or logical inference rules, but abstract transformations of context itself. Subsequent chapters show that this primitive alphabet admits a rich algebraic structure, supports canonical normal forms, and provides the foundation for categorical and topological analysis.
7.1 Motivation Any sufficiently expressive computational system must introduce new contextual in- formation, reject information incompatible with existing constraints, establish relation- ships between previously independent entities, and eventually resolve temporary struc- tures into stable results. Programming languages accomplish these tasks through scope creation, exception handling, variable binding, and function evaluation; mathematical proofs introduce assumptions, reject contradictions, construct dependencies, and dis- charge temporary hypotheses; conversation continually opens new topics, rejects inap-
45
propriate interpretations, links ideas, and eventually resolves discussion into shared conclusions. Although the implementations differ dramatically, the underlying struc- tural transformations remain remarkably similar, and the Scope Calculus abstracts these transformations into four primitive operations acting directly on bounded scope geome- tries.
7.2 Primitive Alphabet Definition 7.1 (Primitive alphabet). The primitive operational alphabet of the Scope Calculus is ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }. Each primitive denotes a partial transformation acting on bounded scope geometries.
The primitives are interpreted operationally rather than symbolically: they describe transformations of context rather than transformations of syntax, and every admissible computation considered throughout this monograph is represented as a finite composi- tion of these primitive operations. The remainder of this chapter examines each primi- tive individually before constructing their algebraic interactions.
7.2.1 The Pop Operator The first primitive governs contextual escalation. Many computations require informa- tion generated within a local scope to become available within an enclosing scope โ returning a function value, proving an intermediate lemma, exporting a variable, or promoting a local conclusion into a global argument.
Definition 7.2 (Pop). Let S๐ โชฏ S๐ be a child scope contained within a parent scope. The Pop operator ๐ซ โถ S๐ โ S๐ promotes an admissible entity from the child scope into its parent while preserving all inherited boundary constraints.
Pop does not destroy the child scope; it merely transfers designated structural infor- mation across the boundary between nested contexts. The admissibility of such promo- tion depends entirely on the boundary constraints of Chapter 6: if promotion violates an inherited boundary, the operation is not admissible. The interaction between Pop and Refuse is examined later in this chapter.
7.2.2 The Refuse Operator The second primitive governs boundary enforcement. Every coherent context must dis- tinguish admissible transformations from inadmissible ones, without which semantic organization degenerates into unrestricted accumulation of incompatible information.
Definition 7.3 (Refuse). Let ๐ฅ โ ๐ be a proposed state transition. The Refuse operator โ rejects the transition whenever ๐ฅ โ โฐ; equivalently, โ acts whenever the proposed transformation violates one or more active boundary predicates.
Refuse is not an exceptional condition added to the calculus, but a primitive struc- tural mechanism preserving admissibility. In programming languages it corresponds to type violations or access restrictions, in logical deduction to invalid inference, in con- versational reasoning to rejecting incompatible interpretations, and in the transformer diagnostics of Chapters 13โ15 it will be interpreted as the detection of structural context collapse rather than low statistical confidence.
7.3. COMPOSITION OF PRIMITIVE OPERATIONS 47
7.2.3 The Bind Operator Whereas Pop transfers existing entities between contexts, Bind creates new structural relationships within a context.
Definition 7.4 (Bind). Let ๐, ๐ โ โฐ. The Bind operator โฌ๐ creates an admissible depen- dency relation ๐ โ ๐ within the current scope geometry.
The precise mathematical nature of this dependency depends on the application: within programming languages it may represent variable association, within logical sys- tems dependence between hypotheses and conclusions, and within neural architectures it may later be interpreted as persistent activation coupling. The calculus intentionally abstracts over these realizations, concerning itself with the creation of lawful structural dependence rather than its physical implementation. Readers who know the ๐-calculus may already suspect a connection between โฌ๐ and variable binding; that connection is made precise, rather than left as an analogy, in Appendix A.
7.2.4 The Collapse Operator The final primitive governs contextual resolution. Temporary structures cannot remain indefinitely unresolved: at some stage subordinate computations terminate and their results become available to surrounding contexts.
Definition 7.5 (Collapse). Let S = โจโฌ, โฐ, ๐ฎchild โฉ. The Collapse operator ๐๐ maps the current scope geometry onto a canonical admissible result while discharging temporary internal structure.
Collapse should not be interpreted as destruction but as evaluation: temporary inter- nal organization is summarized by a stable outcome that can subsequently participate in larger computations, such as evaluation of an expression, completion of a proof, ter- mination of recursion, resolution of a dialogue, or completion of an optimization step.
7.2.5 Operational Completeness Theorem 7.6 (Operational completeness). Every admissible transformation of bounded scope geometries admits a finite presentation as a composition of the primitive operations {๐ซ, โ, โฌ๐ , ๐๐ }.
Proof. Proceed by structural induction over admissible context transformations. Every transformation either introduces contextual information, rejects inadmissible informa- tion, creates admissible dependency, or resolves temporary structure; complex transfor- mations decompose recursively into finite compositions of these elementary cases. The complete inductive argument is deferred to Appendix I.
7.3 Composition of Primitive Operations The primitive operators acquire their mathematical significance only through composi- tion: individual operations describe isolated contextual events, while computation, rea- soning, and semantic interpretation arise from finite sequences of such events. Unlike arithmetic operations, the primitive transformations of the Scope Calculus are inher- ently contextual, so the algebra they generate is generally non-commutative.
Definition 7.7 (Primitive composition). Let ๐ผ, ๐ฝ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }. The composition ๐ฝ โ ๐ผ denotes sequential execution of ๐ผ followed by ๐ฝ, provided the intermediate scope geometry remains admissible.
Whenever admissibility fails after the first operation, the composed transformation is undefined, so the Scope Calculus naturally forms a partial algebra rather than a total one. A finite computation over a bounded scope geometry S0 is represented by an ordered sequence ๐ = ๐ผ๐ โ โฏ โ ๐ผ1 with ๐ผ๐ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }, executed right to left and generating an evolution S0 โ S1 โ โฏ โ S๐ in which each transition satisfies the admissibility conditions of Chapter 6.
Definition 7.8 (Admissible computation). A finite primitive sequence ๐ = ๐ผ๐ โ โฏ โ ๐ผ1 is an admissible computation if every intermediate transition S๐ โ S๐+1 preserves local boundary admissibility.
Admissibility is therefore a property of the entire computation rather than merely its initial and final states.
Proposition 7.9 (Associativity). Let ๐ผ, ๐ฝ, ๐พ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }. Whenever both sides are defined, (๐พ โ ๐ฝ) โ ๐ผ = ๐พ โ (๐ฝ โ ๐ผ).
Proof. Primitive transformations are partial functions acting on bounded scope geome- tries, so associativity follows directly from associativity of functional composition, the only restriction being that every intermediate transformation remains admissible.
Associativity permits omission of parentheses in finite primitive sequences; hence- forth ๐ผ1 โ ๐ผ2 โ โฏ โ ๐ผ๐ denotes the unique associative composition. Although composition is associative, it is generally not commutative: the order in which contextual transformations occur affects the resulting scope geometry, since bind- ing an entity before it exists differs fundamentally from binding it after introduction.
Proposition 7.10 (Non-commutativity). There exist ๐ผ, ๐ฝ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } such that ๐ผ โ ๐ฝ โ ๐ฝ โ ๐ผ.
Proof. Consider a local entity ๐ฅ โ โฐ. Applying โฌ๐ followed by ๐ซ first establishes a dependency before promoting the resulting structure; reversing the order promotes the entity prior to creating the dependency. Unless the parent scope already contains the corresponding relational structure, the resulting scope geometries differ, so ๐ซ โ โฌ๐ โ โฌ๐ โ ๐ซ.
Non-commutativity is therefore an intrinsic consequence of contextual organization rather than an implementation artifact.
Definition 7.11 (Identity operation). The identity operation id is the unique primitive sequence of length zero satisfying id(S) = S for every bounded scope geometry.
Consequently ๐ โ id = ๐ = id โ ๐; this identity transformation later becomes the identity morphism of the category Scope introduced in Chapter 8. The four primitives interact in characteristic ways: some compositions are always ad- missible, others require boundary verification, and still others are undefined whenever structural constraints are violated. Table 7.1 summarizes these interactions informally; the precise rewrite rules follow in the next section.
7.4. THE PRIMITIVE REWRITING SYSTEM 49
First Second Typical interpretation โฌ๐ ๐ซ Promote established relation ๐ซ โฌ๐ Bind after promotion โ ๐ซ Promotion denied ๐ซ โ Boundary validation after promotion ๐๐ ๐ซ Promote evaluated result ๐๐ โฌ๐ Bind canonical representative โฌ๐ ๐๐ Evaluate constructed dependency
Table 7.1: Typical interactions among primitive operators. Formal rewrite rules follow in the next section.
7.4 The Primitive Rewriting System We now equip the algebra with a rewriting system that identifies operationally equiv- alent computations. Its purpose is twofold: first, to eliminate redundant or vacuous transformations, yielding shorter and more canonical descriptions of computations; sec- ond, to provide the foundation for the normal-form results of Chapter 9. The rewriting system is deliberately conservative โ it does not alter the semantic content of a computa- tion, but replaces one admissible presentation with another that is structurally simpler while preserving boundary constraints and contextual relationships. The elementary treatment given here is sufficient for the results of this chapter; the general theory of abstract reduction systems that this construction is an instance of, including the full ter- mination and confluence arguments, is developed independently in Appendix B.
Definition 7.12 (Primitive word). A primitive word is a finite sequence ๐ค = ๐ผ1 ๐ผ2 โฏ ๐ผ๐ with ๐ผ๐ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }. The empty word is denoted ๐.
The collection of all primitive words forms the free monoid ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ under concatenation, containing every syntactically valid computation, including many that are operationally redundant or semantically equivalent; the purpose of rewriting is to identify these equivalent representations.
Definition 7.13 (Rewrite relation). A rewrite relation โถ is a binary relation on ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ such that ๐ข โถ ๐ฃ whenever the primitive word ๐ข may be replaced by the simpler word ๐ฃ without altering its admissible structural action on bounded scope geometries.
Repeated application of rewrite rules produces a reduction sequence ๐ค โถ ๐ค1 โถ ๐ค2 โถ โฏ; a word is irreducible if no rewrite rule applies. The primitive algebra admits four elementary reductions, motivated by operational redundancy rather than arbitrary symbolic simplification: identity elimination, ๐ค๐ โถ ๐ค and ๐๐ค โถ ๐ค (R1); consecutive refusal, โโ โถ โ (R2), since repeated refusals without an intervening admissible trans- formation are equivalent to a single refusal; consecutive collapse, ๐๐ ๐๐ โถ ๐๐ (R3), since further collapse operations after a scope has already been collapsed into its canonical representative produce no additional effect; and empty promotion, ๐ซ๐ โถ ๐ (R4), since promotion of an empty computation is redundant. These elementary rules intentionally avoid introducing semantic assumptions regarding application-specific behavior; more specialized rewrite rules may be added in concrete implementations provided they pre- serve admissibility.
Definition 7.14 (Operational equivalence). Two primitive words ๐ข, ๐ฃ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ are operationally equivalent, written ๐ข โก ๐ฃ, if one can be transformed into the other through a finite sequence of rewrite rules and inverse rewrite rules.
Operational equivalence partitions the free monoid into equivalence classes whose members describe identical admissible computations, so the Scope Calculus distinguishes between the syntactic form of a computation and its structural effect.
Lemma 7.15 (Termination of elementary rewriting). Every finite sequence of elementary rewrite rules terminates after a finite number of reductions.
Proof. Each elementary rewrite rule strictly decreases either the length of the primitive word or the number of redundant operators it contains. Since every primitive word has finite length, no infinite descending chain exists, so every reduction sequence terminates.
Termination alone is insufficient: different rewrite orders must also lead to compati- ble results.
Definition 7.16 (Local confluence). A rewriting system is locally confluent if whenever ๐ค โถ ๐ข and ๐ค โถ ๐ฃ, there exists a primitive word ๐ง such that ๐ข โถโ ๐ง and ๐ฃ โถโ ๐ง.
For the elementary rewriting rules above, overlapping rewrite patterns occur only in a finite number of cases, each resolving consistently; the detailed verification ap- pears in Appendix I. Termination and local confluence together motivate the search for unique canonical representatives of operational equivalence classes; rather than prov- ing uniqueness immediately, we postpone the complete argument to Chapter 9, where the Primitive Normal Form Theorem is established after the algebraic and categorical foundations have been completed. For the remainder of this chapter a primitive word is called reduced if no elementary rewrite rule applies; every admissible computation considered in later chapters may be represented by one or more primitive words, each reducible to a common reduced representative.
7.5 Algebraic Properties of the Scope Calculus Having introduced the primitive alphabet and its rewriting system, we now examine the algebraic structure it generates. Unlike many familiar algebraic systems, the Scope Calculus is neither a group nor a ring: primitive operations are generally irreversible, context-dependent, and only partially defined. Nevertheless they exhibit a rich collec- tion of structural regularities sufficient to support canonical reduction, categorical inter- pretation, and representation independence.
Definition 7.17 (Scope monoid). The pair (ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ , โ) consisting of all finite primitive words together with concatenation is called the Scope Monoid.
Proposition 7.18. The Scope Monoid is closed under concatenation, associative, and possesses the identity element ๐.
Proof. Closure follows because concatenation of two finite primitive words produces an- other finite primitive word; associativity follows from concatenation of finite sequences; and ๐ acts as both left and right identity.
This monoid is intentionally syntactic; operational semantics arise only after admis- sibility constraints and rewriting relations are introduced.
Definition 7.19 (Operational quotient). Let โก denote operational equivalence under the rewrite system. The quotient algebra ๐ฌ = ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ / โก is the Scope Quotient Algebra.
Elements of ๐ฌ are equivalence classes of primitive computations rather than individ- ual primitive words, so the quotient algebra captures computational behavior instead of syntactic representation.
Definition 7.20 (Partial composition). Let ๐ข, ๐ฃ โ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ . The composi- tion ๐ฃ โ ๐ข is defined only if execution of ๐ข terminates in a scope geometry satisfying the admissibility conditions required for the initial application of ๐ฃ.
Thus the Scope Calculus forms a partial algebra over admissible computations; this partiality is not an inconvenience but one of the defining features of the theory, since invalid compositions correspond precisely to structural violations of context.
Proposition 7.21. Under the elementary rewrite rules, โ2 = โ and ๐2๐ = ๐๐ .
Proof. Both identities follow immediately from rewrite rules (R2) and (R3): repeated refusal introduces no additional structural restriction beyond the first refusal, and once a scope has been collapsed into its canonical representative, further collapse operations leave the resulting geometry unchanged.
These identities illustrate that certain operations are naturally stabilizing, with re- peated application eventually reaching a fixed point.
Proposition 7.22. The Scope Calculus is not, in general, a group.
Proof. Suppose ๐๐ possessed an inverse; then every collapsed computation could be re- constructed uniquely into its complete internal history. However, Collapse intentionally replaces temporary internal structure by a canonical representative, so information re- garding intermediate computation is generally lost, and ๐๐ admits no universal inverse. An identical argument applies to โ, whose rejection of inadmissible transitions does not uniquely determine the rejected proposal. Hence the primitive algebra cannot sat- isfy the group axioms.
This non-invertibility distinguishes Scope Geometry from reversible computation while aligning naturally with evaluation, abstraction, and proof construction.
Definition 7.23 (Operational fixed point). A bounded scope geometry S is a fixed point of a primitive transformation ๐ผ if ๐ผ(S) = S.
Examples include a fully evaluated computation under ๐๐ , a boundary already en- forcing every relevant admissibility constraint under โ, and an established dependency under repeated โฌ๐ where no additional relations are introduced. Fixed points provide the algebraic foundation for convergence arguments developed in Chapter 9 and later become central to the diagnostic theory of stable semantic trajectories in Chapters 13 and 15.
7.6 Discussion The algebra introduced in this chapter deliberately occupies a middle ground between abstract algebra and operational semantics. On one hand, the primitive operators gener- ate an algebraic structure supporting composition, quotient construction, and canonical reduction; on the other, the algebra remains closely connected to concrete computational behavior through admissibility constraints and contextual boundaries. This dual char- acter distinguishes the Scope Calculus from traditional rewriting systems: rather than manipulating symbols in isolation, every rewrite reflects a structural transformation of bounded semantic contexts. The following chapter abstracts these operational ideas one level further, examining the category whose objects are bounded scope geometries and whose morphisms are admissible context-preserving transformations; the algebra developed here becomes the operational substrate for the categorical constructions of Chapter 8.
7.7 Chapter Summary This chapter completed the construction of the Scope Calculus. Beginning with the four primitive operations Pop, Refuse, Bind, and Collapse, we developed their composition laws, established the elementary rewriting system, introduced operational equivalence, and constructed the quotient algebra governing admissible computations, showing that the primitive alphabet generates a partial monoid, identifying idempotent operators, proving the absence of global inverses, and introducing operational fixed points. These results provide the complete algebraic machinery required for the categorical develop- ment that follows, in which Chapter 8 elevates these operational constructions into the category Scope.
Status of Results: Chapter 7 Established background. Free monoids, rewriting systems, quotient algebras, partial algebras, and operational semantics [5, 29, 37].
Formal developments. Definition of the Scope Monoid, Scope Quotient Algebra, par- tial composition, operational equivalence, idempotence, non-invertibility, and opera- tional fixed points.
Empirical hypotheses. None. Chapter 7 develops the algebraic machinery underlying the Scope Calculus; empirical application begins only in Volume III after the categorical and topological foundations have been completed.
Chapter 8
Category-Theoretic Foundations of Scope Geometry
Mathematics is the art of giving the same name to different things. โ Henri Poincarรฉ
Category theory was introduced to mathematics by Eilenberg and Mac Lane as a language for describing mathematical structure independently of any particular real- ization [14, 37], and during the twentieth century it became one of the central orga- nizing frameworks of modern algebra, topology, geometry, logic, and computer science. The motivation for introducing it into Structural Semantics is considerably more modest. The previous chapters established two complementary viewpoints: bounded scopes as mathematical objects possessing boundaries, interiors, and nested child scopes, and ad- missible computations as finite compositions of primitive operations governed by the Scope Calculus. These developments naturally suggest a higher level of abstraction, in which we study the transformations between scope geometries rather than reasoning di- rectly about primitive words or individual scope objects, and category theory provides precisely the mathematical language required for this transition. Throughout this chapter we adopt a principle of structural minimalism: only those categorical constructions required by subsequent chapters are introduced. We do not assume that the resulting category is monoidal, cartesian closed, complete, cocomplete, or possesses universal limits; such structures may eventually arise as derived properties, but they are never assumed a priori.
8.1 Motivation Many mathematical theories begin by studying individual objects before discovering that the important information lies not within isolated objects but in the relationships between them: linear algebra studies vector spaces together with linear maps, topol- ogy studies spaces together with continuous maps, and group theory studies groups to- gether with homomorphisms. Likewise, Structural Semantics ultimately studies scope geometries together with admissible context-preserving transformations. This shift of emphasis is essential, since a semantic system rarely remains static โ programs execute, proofs evolve, conversations accumulate context, reasoning systems revise hypotheses, and large language models continuously transform latent representations from one layer
53
to the next โ so semantic organization should be understood primarily as a family of transformations rather than as isolated states. The purpose of category theory is there- fore not to replace the Scope Calculus, but to provide an abstract language describing the algebra constructed in Chapter 7.
Definition 8.1 (Structural minimalism principle). A categorical construction shall be introduced only if every required axiom is explicitly verified, the construction is neces- sary for subsequent mathematical development, and no stronger categorical structure is assumed than is required.
This principle ensures that the categorical framework grows naturally from the alge- bra rather than being imposed externally.
8.2 Objects The objects of the category are precisely the bounded scope geometries introduced ear- lier, reinterpreted from the categorical perspective.
Definition 8.2 (Objects of Scope). The objects of Scope are all well-formed bounded scope geometries S = โจโฌ, โฐ, ๐ฎchild โฉ satisfying the admissibility conditions of Chapter 6.
No assumptions are made concerning the internal structure of the state space: the interior may consist of symbolic expressions, graphs, neural activations, logical formu- las, computational processes, or any other representation satisfying the boundary con- ditions, so the objects of Scope are intentionally substrate independent.
8.3 Morphisms The morphisms of the category represent admissible semantic transformations. Op- erationally these correspond to computations generated by the primitive alphabet; ab- stractly they are structure-preserving maps between bounded scope geometries.
Definition 8.3 (Scope morphism). Let S1 , S2 โ Ob(Scope). A morphism ๐ โถ S1 โ S2 is an admissible transformation preserving the boundary relations of the source scope throughout execution.
This definition is deliberately abstract: a morphism is characterized by what it pre- serves, not by how it is presented syntactically. The connection to the primitive alphabet of Chapter 7 is then a fact to be proved rather than a feature built into the definition, which keeps the category free to accommodate other presentations of the same transfor- mations later.
Proposition 8.4 (Primitive presentation). Every scope morphism ๐ โ Hom(S1 , S2 ) admits a presentation as a finite sequence of primitive operations drawn from ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }.
Proof. By Definition 6.5, the structural difference between S1 and S2 is confined to finite modifications of โฌ, โฐ, and ๐ฎchild . Theorem 7.6 established that every admissible transfor- mation of a bounded scope geometry decomposes into a finite composition of ๐ซ, โ, โฌ๐ , and ๐๐ ; applying this to each of the finitely many modifications yields a finite primitive presentation of ๐.
Morphisms need not preserve every internal detail: temporary computational states may change dramatically, and only admissibility constraints are required to remain valid throughout the transformation. This observation foreshadows the Representation In- variance Theorem of Chapter 9. The categorical structure introduced in this chapter is kept deliberately minimal; the stronger question of whether Scope supports the full Cartesian closed structure needed to interpret typed ๐-terms is addressed, under ex- plicit hypotheses, in Appendix C. Definition 8.5 (Identity morphism). For every bounded scope geometry S there exists an identity morphism idS โถ S โ S defined by idS (๐ฅ) = ๐ฅ for every admissible state ๐ฅ โ โฐ. Operationally, the identity morphism corresponds to the empty primitive computa- tion of Chapter 7: no boundaries change, no scope relationships are modified, and no computational history is generated. Identity therefore represents structural persistence rather than computational inactivity. Definition 8.6 (Composition). Suppose ๐1 โถ S1 โ S2 and ๐2 โถ S2 โ S3 . Their compo- sition ๐2 โ ๐1 โถ S1 โ S3 is defined whenever the codomain of ๐1 equals the domain of ๐2 . This definition mirrors ordinary functional composition; operationally it represents the execution of one admissible context transformation immediately after another while preserving the boundary constraints inherited from intermediate scopes. Theorem 8.7. The pair (Ob(Scope), Hom(Scope)) forms a category. Proof. Identity morphisms exist by construction, and composition is defined whenever codomain and domain agree. Associativity follows immediately from associativity of se- quential composition established for the Scope Calculus in Chapter 7. Finally, idS2 โ๐ = ๐ = ๐ โ idS1 for every admissible morphism ๐ โถ S1 โ S2 . Therefore all category axioms are satisfied.
The importance of this theorem is conceptual rather than computational: it estab- lishes that semantic transformations possess exactly the structural regularity required for categorical reasoning.
8.4 Interpretation The category Scope should not be interpreted as describing cognition, consciousness, or intelligence; instead it provides an abstract mathematical language for discussing families of admissible transformations independently of their concrete implementation. Whether a computation is carried out by symbolic rewriting, biological neurons, silicon processors, or distributed agents is irrelevant at the categorical level; only the preser- vation of admissible scope structure remains visible. This abstraction is precisely what later permits comparison between Whiteheadian process philosophy and Scope Geome- try through a translation functor developed in Chapter 10. Readers wondering whether the single notion of composition used throughout this chapter is the only one Scope Geometry supports may wish to know that it is not: a second, independent notion of composition, arising from refining a scope rather than executing it, is developed as a con- ditional extension in Appendix D, though nothing in the chapters that follow depends on it.
8.5 Chapter Summary This chapter introduced the category Scope using the principle of structural minimalism. Objects were defined as bounded scope geometries, morphisms as admissible context- preserving transformations, and sequential execution was formalized through categor- ical composition. Verification of the category axioms established that Scope Geometry possesses a mathematically well-defined categorical structure while avoiding unneces- sary assumptions concerning monoidal or higher categorical properties. The next chap- ter develops the principal structural theorems governing this category: closure, decom- position, canonical normal forms, admissibility preservation, reconstruction, and the Representation Invariance Theorem.
Status of Results: Chapter 8 Established background. Elementary category theory: objects, morphisms, identity morphisms, and composition [4, 14, 37].
Formal developments. Definition of the category Scope, the Structural Minimalism Principle, categorical interpretation of bounded scope geometries, admissible morphisms, and verification of the category axioms.
Empirical hypotheses. None. This chapter develops only the abstract mathematical framework required for subsequent structural theorems.
Chapter 9
Foundational Theorems of Structural Semantics
The essence of mathematics lies in its freedom. โ Georg Cantor
The preceding chapters introduced the fundamental components of the Scope Calcu- lus: bounded scope geometries as the primary semantic objects, primitive operations as generators of admissible computations, and Chapter 8 demonstrated that these objects and transformations naturally form the category Scope. The present chapter instead establishes the principal structural results on which the remainder of the monograph depends, providing the internal consistency of the Scope Calculus and justifying its in- terpretation as a framework for describing semantic organization independently of any particular computational substrate. Several proofs are necessarily lengthy; to preserve continuity of the main exposition, complete technical proofs are deferred to Appendix I, and only the principal arguments are presented here.
9.1 Motivation Every mathematical theory requires internal guarantees: linear algebra requires that matrix multiplication be associative, topology requires continuity to behave well under composition, and group theory requires closure under multiplication. Structural Seman- tics requires analogous guarantees, without which the primitive operations of Chapter 7 would remain merely descriptive rather than forming a coherent mathematical system. The goals of this chapter are therefore to demonstrate that admissible computations re- main within the space of admissible scope geometries; that every admissible computa- tion possesses a canonical decomposition; that the rewriting system converges to unique normal forms; that semantic preservation can be characterized in terms of admissibil- ity; and finally to establish the Representation Invariance Theorem, which provides the mathematical basis for substrate independence.
9.2 Preliminary Assumptions Throughout this chapter we assume the framework established previously: bounded scope geometries as in Definition 6.5, primitive operations satisfying the operational
57
rules of Chapter 7, admissible morphisms forming the category Scope of Chapter 8, and finiteness of every primitive computation. The finiteness assumption is important: with- out it, uniqueness of normal forms and termination of rewriting would generally require additional hypotheses concerning infinitary rewriting systems.
9.3 Closure of Scope Geometry Theorem 9.1 (Closure theorem). Let S โ Ob(Scope) and let ๐ โถ S โ Sโฒ be an admissible morphism. Then Sโฒ โ Ob(Scope). Proof. Every admissible morphism preserves the boundary relations defining the target scope. Because admissibility requires every primitive operation to satisfy the boundary tensor โฌ, the resulting object satisfies the same well-formedness conditions required by Definition 6.5, so the codomain remains a bounded scope geometry. A complete inductive proof appears in Appendix I. The Closure Theorem guarantees that semantic computation cannot escape the math- ematical universe of discourse, analogous to closure properties throughout algebra.
9.4 Canonical Primitive Decomposition Theorem 9.2 (Scope decomposition). Every admissible morphism ๐ โถ S1 โ S2 admits a finite decomposition into primitive operations drawn from ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }. Proof. Proceed by structural induction on the length of admissible computations. The base case consists of identity morphisms. The inductive step follows because every ad- missible transformation changes only one of โฌ, โฐ, or ๐ฎchild , and Chapter 7 established that each such modification is generated by one of the primitive operations. Hence every admissible computation possesses a primitive decomposition. The significance of this theorem extends beyond implementation: it states that the four primitive operations are operationally complete with respect to admissible seman- tic transformations.
9.5 Canonical Normal Forms Primitive decompositions are generally not unique, since different computational histo- ries may represent the same semantic transformation. The next theorem resolves this ambiguity. Theorem 9.3 (Primitive normal form). Every admissible primitive sequence reduces, through the rewriting rules of Chapter 7, to a unique canonical normal form ๐โ . Proof. Termination follows because every rewrite decreases a well-founded complexity measure on primitive expressions. Local confluence follows from pairwise verification of overlapping rewrite rules. Since the rewriting system is terminating and locally con- fluent, Newmanโs Lemma implies global confluence, so every admissible computation possesses a unique normal form. The complete proof occupies Appendix I. Without canonical normal forms, semantic comparison between different compu- tations would remain fundamentally ambiguous; normal forms therefore provide the canonical representatives of semantic equivalence classes.
9.6. ADMISSIBILITY PRESERVATION 59
9.6 Admissibility Preservation Theorem 9.4 (Admissibility preservation). A computational system possesses structural se- mantic capacity precisely when every intermediate transformation preserves admissibility.
Proof. Suppose admissibility fails at some intermediate computation; then at least one boundary relation becomes invalid, and the resulting computation no longer represents a legal transformation within Scope. Conversely, if every intermediate transformation preserves admissibility, every composed morphism remains well-defined by the Clo- sure Theorem, so semantic organization is preserved throughout the computation.
This theorem formalizes one of the central philosophical claims of Chapter 1: se- mantic capacity is characterized by preservation of admissible structure rather than by phenomenological or behavioral criteria.
Theorem 9.5 (Composition integrity). Suppose ๐1 โถ S1 โ S2 and ๐2 โถ S2 โ S3 are admissible. Then ๐2 โ ๐1 is likewise admissible.
Proof. Because both morphisms individually preserve the boundary tensor โฌ, every in- termediate scope remains admissible. Associativity of composition established in Chap- ter 8 then implies that their sequential execution also preserves admissibility.
This theorem allows arbitrarily long semantic computations to be constructed from locally admissible components, providing the compositional principle underlying sub- sequent chapters.
Theorem 9.6 (State reconstruction). Let S be a bounded scope geometry satisfying the admis- sibility conditions of Chapter 6. If every collapse operation records its boundary history, then the parent scope can be reconstructed uniquely.
Proof. Each collapse records precisely those admissible boundary modifications required to reconstruct the parent scope. Because admissibility excludes ambiguous boundary transitions, the reconstruction algorithm is deterministic. Full details are presented in Appendix I.
This theorem plays an important role later in the development of diagnostic algo- rithms for transformer models (Chapter 14), where observable trajectories are used to infer latent scope evolution. The underlying pattern โ assembling a global object from a family of locally consistent pieces โ recurs throughout mathematics under the name of gluing; Appendix F develops this reading of reconstruction explicitly, in terms gen- eral enough to state precisely when local reconstruction data fails to determine a unique global scope.
9.7 Representation Invariance The preceding theorems remain entirely internal: they establish that the calculus is mathematically coherent, but do not yet address its principal philosophical motivation. Structural Semantics was introduced in Chapter 1 as a substrate-independent account of semantic organization. To justify this claim mathematically, we require a precise crite- rion under which two computational systems possessing different internal architectures may nevertheless instantiate the same semantic computation. Unlike many discussions
of representation in philosophy of mind, the theorem below makes no claims concern- ing consciousness, subjective experience, intentionality, or ontology; it concerns only the preservation of admissible transformation structure within Scope.
Definition 9.7 (Admissibility-preserving projection). Let ๐ be a computational state space. A mapping โ โถ ๐ โ Ob(Scope) is an admissibility-preserving projection if it is surjective; for every admissible transition ๐ฅ โ ๐ฆ in ๐, the induced scope transition โ(๐ฅ) โ โ(๐ฆ) is admissible in Scope; and boundary relations are preserved, โฌ(๐ฅ) = โฌ(๐ฆ) โ โฌ(โ(๐ฅ)) = โฌ(โ(๐ฆ)).
The definition deliberately does not require injectivity: many computational states may project onto the same scope geometry, since different implementations frequently possess redundant internal degrees of freedom that play no role in semantic organiza- tion. The projection should not be interpreted as a decoder recovering hidden semantic content, but as an abstraction analogous to quotient mappings in algebra or projection operators in topology, identifying the structural information retained by the Scope Cal- culus while ignoring implementation-specific variation. Remark 9.8. The admissibility clause of Definition 9.7 is a compatibility condition, not a free choice: it requires ๐ฅ โผadm ๐ฆ in ๐ to imply โ(๐ฅ) โผadm โ(๐ฆ) in Scope for every pair of states connected by an admissible transition, not merely for some convenient subset of them. Theorem 9.10 below depends on this condition holding for the specific โ1 , โ2 under consideration; the theorem is conditional on the existence of such projections; it does not assert that admissibility-preserving projections exist for every pair of systems one might wish to compare.
Definition 9.9 (Semantic equivalence). Let ๐1 , ๐2 be admissible morphisms in Scope. They are semantically equivalent, ๐1 โผsem ๐2 , whenever their canonical normal forms coincide: ๐1โ = ๐2โ .
Because canonical normal forms are unique (Theorem 9.3), this relation is well de- fined, and it is straightforward to check that โผsem is an equivalence relation on the class of admissible morphisms; the admissible morphisms therefore partition into equiva- lence classes, each represented uniquely by its canonical normal form.
Theorem 9.10 (Representation invariance). Let โณ1 and โณ2 be computational systems with state spaces ๐1 , ๐2 . Suppose there exist admissibility-preserving projections โ1 โถ ๐1 โ Ob(Scope) and โ2 โถ ๐2 โ Ob(Scope) such that every admissible computation induces a morphism of Scope; corresponding induced morphisms possess identical canonical normal forms; and corre- sponding boundary tensors are preserved under โ1 and โ2 . Then โณ1 and โณ2 are semantically equivalent relative to the Scope Calculus.
Proof. Let ๐1 = (๐ฅ0 , โฆ , ๐ฅ๐ ) be an admissible computation inside โณ1 , and let ๐2 = (๐ฆ0 , โฆ , ๐ฆ๐ ) be the corresponding computation inside โณ2 . Applying the projections yields admissible morphisms ๐1 = โ1 (๐1 ) and ๐2 = โ2 (๐2 ). By Definition 9.7, both morphisms preserve admissibility. By Theorem 9.3, each possesses a unique canonical normal form, ๐1โ , ๐2โ . The second hypothesis states ๐1โ = ๐2โ , hence ๐1 โผsem ๐2 by Def- inition 9.9. Preservation of boundary tensors guarantees that the equivalence extends beyond the terminal states to every admissible intermediate transition. Therefore both systems determine precisely the same semantic computation inside Scope, and โณ1 , โณ2 are semantically equivalent relative to the Scope Calculus.
9.7. REPRESENTATION INVARIANCE 61
The theorem deliberately avoids stronger claims: it does not assert that two compu- tational systems are physically equivalent, nor that they have equal computational com- plexity, energy consumption, learning dynamics, or phenomenological equivalence. It establishes only that the admissible semantic structure recognized by the Scope Calcu- lus is invariant under the stated projection hypotheses.
Corollary 9.11. Structural semantic capacity depends only on admissible transformation struc- ture and not on the physical substrate implementing that structure.
Proof. Immediate from Theorem 9.10.
The Representation Invariance Theorem serves as the principal mathematical bridge between the abstract theory of Volumes I and II and the empirical investigations of trans- former representations developed in Volume III: rather than searching for semantic orga- nization directly within vectors, tensors, symbolic graphs, or biological neurons, subse- quent chapters investigate whether these systems induce equivalent admissible transfor- mation structures after projection into the common categorical framework established by Scope.
9.7.1 Interpretation of Representation Invariance The theorem should not be understood as claiming that all computational architectures are equivalent, but as specifying a precise criterion under which equivalence may be established โ one based not on physical realization, implementation language, com- putational complexity, or biological plausibility, but solely on the preservation of ad- missible transformation structure within Scope. This mirrors analogous developments throughout mathematics: the same abstract group may be represented by matrices, per- mutations, or symmetries of geometric objects, and although these realizations differ substantially in implementation, they are regarded as equivalent because they preserve the same algebraic structure. Structural Semantics adopts an analogous perspective, treating the organization of admissible transformations rather than the implementation or coordinate system as the object of interest.
Definition 9.12 (Substrate independence). A semantic property is substrate independent whenever its validity depends only on admissible structural transformations within Scope and not on the physical realization of those transformations.
This definition intentionally avoids stronger metaphysical claims: it does not imply that all substrates are equally efficient, expressive, or capable of supporting arbitrary computations, only that once admissible transformations have been projected into the Scope Calculus, semantic evaluation depends on those transformations rather than on their physical implementation. Consider, for instance, a compiler translating a high- level programming language into two different machine architectures: although the re- sulting executable instructions differ substantially, both preserve identical control flow, scope hierarchy, and variable-binding relationships, and within the Scope Calculus these executions correspond to the same canonical admissible morphism despite differing bi- nary representations. Similarly, if a symbolic reasoning system built on graph rewriting and a transformer neural network induce identical admissible primitive transformations after projection into Scope, Theorem 9.10 establishes their semantic equivalence relative to the Scope Calculus, without implying identical learning mechanisms, computational cost, training dynamics, or phenomenological properties.
Representation Invariance occupies an intermediate position between several estab- lished traditions: denotational equivalence in formal semantics identifies programs whose observable behaviors coincide despite differing implementations [49, 58], quotient con- structions in algebra identify objects modulo equivalence relations, and naturally iso- morphic constructions in category theory are routinely regarded as equivalent despite differing internal presentations [4, 37]. The present theorem shares this structural phi- losophy while preserving a different invariant: instead of numerical values, logical for- mulas, or observable program outputs, it preserves admissible context transformations. This distinction becomes particularly important in Chapters 13โ15, where the empir- ical question investigated is not whether latent vectors resemble human concepts di- rectly, but whether they preserve admissible semantic organization under projection into Scope Geometry. Remark 9.13 (A parallel in philosophy of physics). A structurally similar move appears in the philosophy of physics. Ismael and van Fraassen [32] argue that a symmetry trans- formation identifies surplus theoretical structure precisely when it leaves every observ- able prediction unchanged: two models related by such a transformation, ๐ โผ ๐ โฒ , should be treated as one physical state rather than two, so that the physically meaningful object is the quotient โณ/ โผ rather than the original space of models. The admissibility- preserving projections of Definition 9.7 play an analogous role here: two computational systems related by such a projection are treated, relative to the Scope Calculus, as one semantic computation rather than two, regardless of how their underlying representa- tions differ. The two constructions are not the same โ theirs identifies structure that is dynamically inert, ours identifies structure that is admissibility-irrelevant โ but both replace a space of concrete representations with a coarser space of what those repre- sentations are for. Separately, Ismael [31] argues that a deterministic process cannot introduce a symmetry-breaking distinction into its effect that was not already present, in some form, in its cause: an asymmetric outcome requires an asymmetric antecedent. The Closure Theorem (Theorem 9.1) and Admissibility Preservation Theorem (Theo- rem 9.4) make a comparable claim about admissible transformations specifically: no admissible sequence of primitive operations can produce a scope geometry violating constraints that were not already violable given the boundary structure it started from. Neither parallel is used to derive anything in this monograph; both are noted because the same structural intuition โ that transformation can reveal or reorganize distinctions but not manufacture them from nothing โ recurs across fields that otherwise share little vocabulary.
9.7.2 Representations as Coordinates Rather Than Meaning It is worth stating plainly what Theorem 9.10 implies about the status of representations themselves, since the implication is easy to state informally but easy to misread formally. The theorem does not say that vectors, symbols, neural activations, and graphs are all secretly the same kind of thing. It says that, once an admissibility-preserving projection into Scope has been specified, the differences among them become differences in coordi- nates on a common invariant object rather than differences in the object itself โ in the same sense that a point in the plane has different numerical descriptions in Cartesian and polar coordinates without being two different points. This reframing has a consequence for how the question โwhat does this representa- tion mean?โ should be posed within this framework. Asked of a coordinate system di- rectly, the question is comparatively hard: a 4096-dimensional activation vector, consid-
9.8. LEVELS OF STRUCTURAL EQUIVALENCE 63
ered as raw numbers, carries no more intrinsic meaning than a pair of polar coordinates carries intrinsic direction independent of the convention that assigns it one. Asked of the admissible transformation the vector participates in, the question becomes tractable, because admissibility is exactly the structure the projection is built to preserve. The shift in emphasis from Chapters 11 through 13, from cataloguing what activation space looks like to tracking what transformations it admits, is this reframing applied to a specific representational format. The reframing has a limit worth stating alongside it. A choice of coordinates is never entirely arbitrary in practice: some coordinate systems make a given computation ef- ficient and others make it intractable, exactly as Cartesian coordinates simplify some geometric problems that polar coordinates complicate. Representation Invariance estab- lishes that semantic content, as this monograph defines it, does not depend on the choice; it does not establish that the choice is inconsequential for anything else, including learn- ability, computational cost, or the ease with which a projection satisfying Definition 9.7 can be found or verified in the first place.
9.7.3 Limitations of the Present Results The Representation Invariance Theorem establishes only a conditional equivalence: its conclusions depend on the hypotheses stated explicitly in Theorem 9.10, and if admissibility- preserving projections do not exist, or if canonical normal forms fail to coincide, no se- mantic equivalence follows. Nor does the theorem provide a constructive algorithm for discovering such projections; the existence of admissibility-preserving mappings is a mathematical assumption whose verification depends on the computational architec- ture under investigation, and developing practical procedures for constructing these projections is one of the principal objectives of Volume III. The distinction between ex- istential theorems and constructive algorithms is fundamental: Chapter 9 proves that semantic equivalence follows if suitable projections exist, while Chapter 14 investigates algorithmic procedures for approximating such projections within transformer activa- tion spaces. The mathematical theorem should therefore not be interpreted as an im- plemented diagnostic method, but as the theoretical target toward which the diagnostic algorithms are directed.
9.8 Levels of Structural Equivalence The equivalence established by Theorem 9.10 is exact: two systems are semantically equivalent relative to the Scope Calculus only if their projected morphisms reduce to the identical canonical normal form. Part III will repeatedly ask a looser question โ whether two activation trajectories are similar enough, in some measurable sense, to in- dicate comparable semantic organization โ and it is worth distinguishing, before that material begins, several notions of equivalence that exact identity does not exhaust. Con- flating them is a natural source of overclaiming in exactly the kind of empirical work Chapters 13 through 15 undertake. Exact equivalence is the relation of Definition 9.9: two morphisms are equivalent only if ๐1โ = ๐2โ , with no tolerance. This is the only notion of equivalence any theorem in this monograph establishes. Approximate equivalence weakens exact equivalence by a distance threshold: two mor- phisms might be considered equivalent if their canonical normal forms differ by less than some ๐ under a suitable metric on ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ . Nothing in Chapters 6โ9
constructs such a metric or proves that any candidate metric interacts well with compo- sition, so approximate equivalence in this sense remains an open extension rather than an established consequence of the present theorems. Probabilistic equivalence replaces a fixed morphism with a distribution over morphisms, as arises naturally once retrieval or routing in a real architecture is stochastic rather than deterministic; two systems might then be called equivalent if their induced distributions over canonical normal forms coincide or are close in some divergence. Appendix Gโs measure-theoretic constructions provide some of the vocabulary such a notion would need, but the monograph does not develop it into a theorem. Functional equivalence is weaker still, and closest to what Chapters 14 and 15 can ac- tually test: two systems are functionally equivalent if they produce statistically indistin- guishable diagnostic behavior โ comparable Phase Locking Values, comparable admis- sibility estimates โ without any claim that their underlying admissible transformations coincide, approximately or otherwise. This is the notion implicitly in play whenever an experiment in Chapter 15 compares model architectures rather than proving a theorem about them. The four levels form a strict hierarchy: exact equivalence implies every weaker no- tion, but none of the implications reverse. An empirical finding of functional equiv- alence between two systems is evidence bearing on, but not a proof of, probabilistic, approximate, or exact equivalence in the senses above. Keeping this hierarchy in view is part of what the Methodological Statement of the front matter asks of every claim in this monograph: an empirical result belongs to the empirical-hypothesis category re- gardless of how strongly it is worded, and dressing a functional-equivalence finding in the language of Theorem 9.10 would misstate what has actually been shown.
9.9 Chapter Summary
This chapter established the principal mathematical foundations of Structural Seman- tics. Beginning with the Closure Theorem, we demonstrated that admissible computa- tions remain within the category Scope. The Scope Decomposition and Primitive Nor- mal Form Theorems showed that every admissible computation possesses a canonical structural representation. The Admissibility Preservation and Composition Integrity Theorems characterized semantic organization as the preservation of boundary con- straints under sequential transformation, while the State Reconstruction Theorem estab- lished that admissible computational histories can be recovered from recorded bound- ary events. These internal consistency results culminated in the Representation Invari- ance Theorem, which demonstrated that semantic equivalence depends on admissible transformation structure rather than the underlying computational substrate, provid- ing the mathematical foundation for treating symbolic systems, neural architectures, graph-based reasoners, and other computational models within a common structural framework. The next chapter builds on these results by introducing a translation func- tor from Whiteheadian process ontology into Scope Geometry.
Status of Results: Chapter 9
Established background. Canonical rewriting systems, equivalence relations, quotient constructions, denotational semantics, and elementary category theory [4, 37, 49, 58].
Formal developments. The Closure, Scope Decomposition, Primitive Normal Form, Admissibility Preservation, Composition Integrity, and State Reconstruction Theorems, together with the Representation Invariance Theorem and its corollary establishing substrate- independent structural semantic capacity.
Empirical hypotheses. None. All results established in this chapter are purely mathe- matical, conditional on the stated hypotheses. The construction of practical admissibility- preserving projections is deferred to the computational framework developed in Vol- ume III.
Chapter 10
Comparative Ontology and the Translation Functor
The many become one, and are increased by one. โ Alfred North Whitehead, Process and Reality
The preceding chapter established the internal mathematical foundations of Struc- tural Semantics: closure, canonical decomposition, admissibility preservation, and rep- resentation invariance demonstrated that the Scope Calculus forms a coherent math- ematical system independent of any particular computational substrate. The present chapter serves a different purpose. Rather than proving additional properties of the cal- culus itself, we construct a formal correspondence between the Scope Calculus and an existing philosophical framework, asking whether the structural relationships described by Whiteheadโs process ontology admit a mathematically precise translation into the language of bounded scope geometries. The answer developed here is affirmative, al- though carefully qualified: we define a functor ๐น โถ Process โ Scope that preserves the operational structure of process interactions while explicitly identifying the point at which the two theories diverge, namely Whiteheadโs notion of Subjective Form.
10.1 Motivation A recurring theme of twentieth-century philosophy is that process precedes substance: rather than viewing reality as composed of static objects possessing accidental prop- erties, process philosophies understand stable objects as relatively persistent organiza- tions of ongoing events. Although Whitehead developed this view for metaphysical purposes, remarkably similar ideas appear throughout modern theoretical computer science โ programming languages describe evolving computational states, distributed systems describe interacting processes, category theory emphasizes morphisms before objects, dynamical systems focus on trajectories rather than isolated points, and large language models themselves perform successive transformations of latent representa- tions rather than manipulating static symbolic objects. These parallels motivate a care- ful mathematical comparison. The purpose of this chapter is therefore not to defend Whiteheadian metaphysics, nor to derive the Scope Calculus from process philosophy, but to investigate the extent to which the operational structure of one framework may be translated into the other.
67
10.2 The Category of Processes Throughout this chapter we use Process to denote an abstract category whose objects represent Whiteheadian actual occasions and whose morphisms represent admissible process transitions. This construction is not a complete formalization of Whiteheadโs philosophy, but a mathematical abstraction capturing only those structures required for comparison with Scope Geometry.
Definition 10.1 (Category of processes). The category Process consists of objects rep- resenting actual occasions, morphisms representing admissible transitions between oc- casions, ordinary categorical composition, and identity morphisms corresponding to persistence of an occasion.
This definition intentionally avoids introducing unnecessary metaphysical structure; concepts such as eternal objects, creativity, and the extensive continuum remain outside the scope of the present construction.
Definition 10.2 (Actual occasion, categorical form). An actual occasion is represented by the ordered triple ๐ = โจ๐ท, ๐ฃ, ๐ดโฉ, where ๐ท denotes the inherited data, ๐ฃ the polarity of prehension, and ๐ด the Subjective Form associated with the occasion.
This representation follows the operational interpretation of Chapter 4 and serves as the domain of the translation functor; no assumptions are yet made concerning the interpretation of ๐ด, whose treatment emerges naturally from the construction of the functor.
10.3 Operational Correspondence Inspection of Chapters 4 through 9 reveals a striking operational parallel between the two theories: positive prehensions accumulate structural relations, negative prehen- sions exclude incompatible relations, transduction propagates contextual information, and concrescence produces completed organizations, while the Scope Calculus pos- sesses analogous operations in binding, refusal, promotion, and canonical evaluation. This observation motivates the correspondence summarized in Table 10.1.
Whiteheadian process Scope Calculus Positive prehension โฌ๐ Negative prehension โ Transduction ๐ซ Concrescence ๐๐
Table 10.1: Operational correspondence between process ontology and Scope Geometry.
The purpose of the remainder of the chapter is to determine whether this informal correspondence extends to a genuine functor.
Definition 10.3 (Translation functor). The translation functor ๐น โถ Process โ Scope acts on objects according to ๐น (๐ ) = S, where S = โจโฌ, โฐ, ๐ฎchild โฉ is the corresponding bounded scope geometry, and on morphisms maps process transitions to admissible scope morphisms preserving boundary structure.
10.4. FUNCTORIAL PRESERVATION 69
The operational component of the translation is determined by the assignments ๐น (๐ + ) = โฌ๐ , ๐น (๐ โ ) = โ, ๐น (๐ ) = ๐ซ, and ๐น (๐ถ) = ๐๐ , motivated by operational rather than philo- sophical similarity: each pair performs analogous structural functions despite arising from different theoretical traditions, and the interpretation should be understood as structural rather than metaphysical. The remaining question concerns Subjective Form, which unlike the preceding components has no direct analogue inside the Scope Calcu- lus and requires separate consideration.
Example 10.4 (The functor applied to a single occasion). Let ๐ = โจ๐ท, ๐ฃ, ๐ดโฉ describe an occasion prehending two antecedent data, ๐1 positively and ๐2 negatively. Under ๐น, the positive prehension of ๐1 becomes an instance of โฌ๐ , establishing an admissible dependency between ๐1 and the developing occasion within the corresponding scope; the negative prehension of ๐2 becomes an instance of โ, excluding ๐2 from the interior of that scope. If the occasionโs data derive from an antecedent occasion in a different scope, the transduction carrying that data forward becomes an instance of ๐ซ; the occa- sionโs concrescence into a completed satisfaction becomes an instance of ๐๐ , the scopeโs resolution into a canonical result. Every component of ๐ except ๐ด is accounted for by one of these four assignments; nothing in the translation invents a fifth operator or maps prehension onto a notion โ such as a labeled โinclusion vectorโ or โexclusion predicateโ โ outside the primitive alphabet ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } already fixed in Chapter 7.
10.4 Functorial Preservation Having defined the action of the translation functor on both objects and primitive oper- ations, we must verify that it satisfies the defining axioms of a functor: preservation of identity morphisms and of composition. Recall from Chapter 8 that a functor between two categories preserves the categorical structure, so the validity of the present con- struction depends on ordinary categorical reasoning [4, 37] rather than philosophical interpretation. This is the second functor constructed in this monograph rather than the first: Appendix C constructs one between the simply typed ๐-calculus and Scope under a Cartesian closure hypothesis, and the categorical vocabulary used to verify ๐น here is the same vocabulary developed rigorously there.
Proposition 10.5 (Identity preservation). For every object ๐ โ Process, the translation functor satisfies ๐น (id๐ ) = id๐น (๐ ) .
Proof. Let ๐ = ๐ be an arbitrary process object. The identity morphism id๐ produces no modification of inherited data, process polarity, or admissible transition structure. Under the action of the translation functor, ๐ โฆ S; since no structural modification occurs, the corresponding scope geometry also remains unchanged, so ๐น (id๐ ) = idS = id๐น (๐ ) .
Theorem 10.6 (Composition preservation). Let ๐ โถ ๐1 โ ๐2 and ๐ โถ ๐2 โ ๐3 be mor- phisms in Process. Then ๐น (๐ โ ๐) = ๐น (๐) โ ๐น (๐).
Proof. Suppose ๐ induces the primitive sequence (๐ผ1 , โฆ , ๐ผ๐ ) while ๐ induces (๐ฝ1 , โฆ , ๐ฝ๐ ). Composition in Process corresponds to concatenation of these transition sequences. The translation functor maps each primitive process operation to its corresponding primitive scope operation according to the assignments above, so ๐น (๐ โ ๐) produces exactly the concatenated admissible primitive sequence ๐น (๐ฝ1 ), โฆ , ๐น (๐ฝ๐ ), ๐น (๐ผ1 ), โฆ , ๐น (๐ผ๐ ), identi- cal to first applying ๐น (๐) followed by ๐น (๐). Hence ๐น (๐ โ ๐) = ๐น (๐) โ ๐น (๐).
10.5 The Image of Subjective Form
The previous sections established a nearly complete structural correspondence between process ontology and Scope Geometry. One component, Subjective Form (๐ด), has re- mained untouched, and this absence is not accidental: it marks the precise mathematical location at which the two frameworks diverge.
Definition 10.7 (Image of the functor). The image of ๐น โถ Process โ Scope consists of every object and morphism generated solely through admissible scope operations.
Theorem 10.8 (Phenomenological boundary). Subjective Form ๐ด lies outside the image of the translation functor: ๐ด โ Im(๐น ).
Proof. The Scope Calculus is generated entirely by the primitive operational alphabet {๐ซ, โ, โฌ๐ , ๐๐ }. Every object and morphism in Scope is constructed by finite compositions of these operations together with their induced boundary relations. Subjective Form ๐ด does not participate in any primitive operational rule: it neither modifies admissibility conditions nor induces additional scope transformations, so no admissible sequence of primitive operations generates an object corresponding to it. Therefore ๐ด does not belong to the image of the functor.
Theorem 10.8 should not be interpreted as a refutation of Whiteheadโs philosophy, but as identifying the precise boundary between two theories: the translation functor preserves every operational component of process organization represented within the Scope Calculus, while Subjective Form remains external because it is introduced as an additional primitive of the philosophical theory rather than as an operational compo- nent of the structural calculus. This distinction mirrors the methodological boundary established in Chapter 5 between formal mathematical constructions and phenomeno- logical interpretations.
10.6 Interpretive Consequences
The existence of the translation functor demonstrates that substantial portions of White- headโs process ontology admit an entirely structural interpretation: positive and neg- ative prehensions become admissible binding and refusal operations, concrescence be- comes canonical collapse, and propagation of process becomes scope transduction. The resulting correspondence is remarkably extensive, but not complete, and its incomplete- ness is mathematically informative rather than philosophically problematic. Instead of attempting to eliminate Subjective Form or reinterpret it as a hidden computational vari- able, the present framework identifies the limits of the translation: the image of the functor consists precisely of those aspects of process ontology that admit operational re- alization within the Scope Calculus. Structural Semantics therefore neither affirms nor denies the existence of phenomenological experience; it demonstrates only that the struc- tural organization of semantic transformations can be developed without requiring phe- nomenological primitives, while whether additional phenomenological structure exists remains an independent philosophical question lying outside the mathematical scope of the present theory.
10.7 Chapter Summary This chapter constructed the translation functor ๐น โถ Process โ Scope, verified that it preserves identity morphisms and composition, and established that Subjective Form lies outside its image, giving a formal mathematical expression of the phenomenological boundary established in Chapter 5. The next chapter leaves philosophy behind and turns toward contemporary machine learning, examining the mathematical structure of transformer architectures before introducing the novel dynamical and spectral analyses developed later in Volume III.
Status of Results: Chapter 10 Established background. Elementary category theory, functors, and Whiteheadโs pro- cess ontology as reconstructed in Chapters 4โ5 [4, 37, 56].
Formal developments. The category Process, the translation functor ๐น โถ Process โ Scope, proofs of identity and composition preservation, and the Phenomenological Bound- ary Theorem establishing that Subjective Form lies outside the image of ๐น.
Empirical hypotheses. None. This chapter is a purely mathematical comparative con- struction.
Part III
Computational Semantics: Transformer Diagnostics
73
Chapter 11
The Mathematical Structure of Transformer Architectures
Attention is all you need. โ Ashish Vaswani et al., 2017
The previous volume developed the abstract mathematical framework of Structural Semantics independently of any particular computational substrate: Scope Geometry, admissibility, categorical structure, and the Representation Invariance Theorem were formulated without reference to neural networks, symbolic systems, or biological cog- nition. Having established this substrate-independent foundation, we now turn to one important computational realization, modern transformer architectures. This chapter differs fundamentally from those preceding it: no new mathematical framework is intro- duced, and no novel definitions of Structural Semantics are proposed. It instead serves as a survey of established mathematical results concerning transformer architectures, establishing a common notation and mathematical vocabulary for the remainder of Vol- ume III. Maintaining this separation matters methodologically: the constructions intro- duced later, most notably the activation trajectory framework of Chapter 12 and the resonant analytic signal model of Chapter 13, are intended as original proposals that should be evaluated against an accurate presentation of the current state of transformer theory rather than replacing or reinterpreting it.
11.1 Historical Development The transformer architecture was introduced by Vaswani et al. [55] as an alternative to recurrent and convolutional architectures for sequence modeling. Rather than process- ing sequences strictly in temporal order, transformers employ attention mechanisms that permit every token to interact directly with every other token within the available con- text window. This architectural change increased computational parallelism, since token representations could be processed simultaneously rather than recursively; made long- range dependencies easier to model, since every token could attend directly to distant contextual information; and proved highly scalable, with increasing parameter count, model depth, and training corpus size producing remarkably predictable improvements across a wide variety of language tasks [27, 33]. These developments established trans- formers as the dominant architecture for contemporary large language models.
75
CHAPTER 11. THE MATHEMATICAL STRUCTURE OF TRANSFORMER 76 ARCHITECTURES
11.2 Residual Stream Representation Mechanistic interpretability has converged on the residual stream as one of the principal mathematical objects of transformer analysis [15, 42, 44]. Throughout this monograph we adopt the standard notation X โ โ(๐ฟ+1)ร๐ ร๐model , where ๐ฟ denotes the number of transformer layers, ๐ the sequence length, and ๐model the embedding dimension; each element x๐,๐ก โ โ๐model represents the residual vector corresponding to token ๐ก after com- pletion of layer ๐. The residual stream functions as a persistent information channel through which every attention block and multilayer perceptron contributes incremental modifications, providing a natural mathematical object for analysing evolving semantic organization.
11.3 Attention Mechanisms Self-attention constitutes the defining computational mechanism of the transformer ar- chitecture. Given an input matrix ๐ โ โ๐ ร๐ , linear projections generate query, key, and value matrices ๐ = ๐๐๐ , ๐พ = ๐๐๐พ , ๐ = ๐๐๐ , and scaled dot-product attention is computed as ๐๐พ โค Attention(๐, ๐พ, ๐ ) = softmax( )๐, โ๐๐ where ๐๐ denotes the key dimension. Multi-head attention repeats this computation across several independent projection spaces before concatenating the resulting repre- sentations. From the perspective of Structural Semantics, attention itself should be re- garded simply as an established computational mechanism; no structural interpretation is imposed at this stage, and such interpretations are deferred until the activation trajec- tory framework of Chapter 12.
11.4 Feed-Forward Networks Each transformer layer also contains a position-wise multilayer perceptron, typically represented as MLP(๐ฅ) = ๐2 ๐(๐1 ๐ฅ + ๐1 ) + ๐2 , where ๐ denotes a nonlinear activation function such as GELU or SwiGLU. Although attention is often viewed as the defining component of transformers, mechanistic interpretability studies increasingly recognize MLP layers as major repositories of learned features [21], so the residual stream reflects the combined effects of attention and feed-forward computation rather than either mech- anism individually.
11.5 Activation Manifolds Each residual vector x๐,๐ก may be regarded as a point within the high-dimensional space โ๐model , and as successive transformer layers are evaluated these vectors trace continuous trajectories through activation space. The collection of all such vectors defines a high- dimensional activation manifold whose geometry has become a major topic of mech- anistic interpretability research, investigated through cosine similarity, Euclidean dis- tance, principal-component structure, singular-vector decompositions, sparse feature representations, clustering, activation patching, and causal mediation. These methods have substantially improved understanding of transformer representations, but remain
primarily geometric descriptions of isolated layers or local neighborhoods. The follow- ing chapter introduces a complementary viewpoint in which activation vectors are in- terpreted as continuous trajectories evolving through depth rather than as independent static points.
11.6 Current Directions in Mechanistic Interpretability Recent work in mechanistic interpretability has shifted from identifying individual neu- rons toward understanding distributed computational circuits, including induction heads, composition of attention circuits, feature superposition, sparse autoencoders, causal tracing, and activation patching [2, 15, 52]. These studies share the common objective of seeking explicit mathematical explanations for the computations performed by inter- nal representations rather than treating neural networks as opaque statistical devices. Structural Semantics aligns naturally with this broader objective, though the framework developed in the present monograph differs in emphasis: whereas mechanistic inter- pretability often studies the functional role of particular neurons or circuits, subsequent chapters focus on admissibility-preserving trajectories through activation space, shift- ing the object of study from isolated computational units to the global organization of transformations.
11.7 Relationship to Scope Geometry At this point no mathematical identification is made between transformer representa- tions and Scope Geometry; this restraint is deliberate. Chapter 9 established that seman- tic equivalence requires explicit admissibility-preserving projections into Scope, and the existence of such projections cannot simply be assumed for neural activation spaces. In- stead the remainder of Volume III develops these constructions gradually: Chapter 12 in- troduces activation trajectories as continuous dynamical systems, Chapter 13 constructs analytic signal representations of those trajectories, Chapter 14 develops computational algorithms for estimating admissibility within latent representations, and Chapter 15 proposes empirical experiments capable of evaluating these constructions.
11.8 Chapter Summary This chapter reviewed the established mathematical foundations of transformer archi- tectures, introducing the residual stream tensor, attention mechanisms, feed-forward networks, activation manifolds, and current approaches in mechanistic interpretability while adopting the standard notation used throughout the contemporary literature. Im- portantly, no novel theoretical claims were introduced; the purpose has been exclusively expository. The next chapter begins the original developments of Volume III by replac- ing the conventional view of activation vectors as isolated points with a dynamical in- terpretation in which semantic organization is studied through continuous trajectories evolving across transformer depth.
Status of Results: Chapter 11 Established background. Transformer architecture, self-attention, residual streams, feed-forward networks, activation manifolds, scaling laws, and mechanistic interpretabil-
CHAPTER 11. THE MATHEMATICAL STRUCTURE OF TRANSFORMER 78 ARCHITECTURES
ity [15, 27, 33, 42, 44, 55].
Formal developments. None. This chapter establishes notation, terminology, and math- ematical background only.
Empirical hypotheses. None. Novel empirical hypotheses begin in Chapter 12 with the dynamical analysis of activation trajectories.
Chapter 12
Dynamic Geometry of Activation Trajectories
The mathematics of change is not merely the mathematics of states, but of the paths connecting them. โ inspired by the dynamical systems tradition
The previous chapter introduced the standard mathematical formalism used to de- scribe transformer architectures, establishing a common vocabulary without introduc- ing new theory. This chapter begins the original developments of Volume III. Rather than studying activation vectors as isolated points in a high-dimensional space, we ex- amine the trajectories generated as those vectors evolve throughout the forward com- putation of a transformer, a shift from static to dynamical geometry paralleling a well- established transition throughout mathematics and physics โ differential geometry stud- ies curves rather than isolated coordinates, dynamical systems investigate flows rather than individual states, and control theory analyzes trajectories rather than instantaneous configurations. The constructions developed in this chapter remain entirely within con- ventional mathematics, employing only linear algebra, multivariable calculus, and clas- sical dynamical systems theory; the analytic signal formulation of Chapter 13 is deliber- ately deferred until this ordinary dynamical framework has been established.
12.1 Activation Trajectories Consider a transformer of ๐ฟ layers. For a fixed input token ๐ก, the residual stream pro- duces a sequence of vectors x0,๐ก , x1,๐ก , โฆ , x๐ฟ,๐ก , with x๐,๐ก โ โ๐model . Rather than treating these vectors independently, we interpret them as a trajectory through activation space.
Definition 12.1 (Activation trajectory). For a fixed token ๐ก, the activation trajectory is the ordered sequence ๐พ๐ก = (x0,๐ก , x1,๐ก , โฆ , x๐ฟ,๐ก ); equivalently, ๐พ๐ก โถ {0, โฆ , ๐ฟ} โ โ๐model is a discrete curve parameterized by transformer depth.
This trajectory represents the progressive refinement of the latent representation as- sociated with a single token as information propagates through the network. Where Chapter 11 emphasized the geometry of activation space itself, the present chapter em- phasizes the geometry of motion within that space.
79
12.2 Discrete Dynamical Systems Interpretation Transformer evaluation naturally induces a discrete dynamical system: each layer de- fines a transition operator ๐น๐ โถ โ๐model โ โ๐model such that x๐+1 = ๐น๐ (x๐ ), and the complete forward computation is the composition ๐น = ๐น๐ฟโ1 โ โฏ โ ๐น1 โ ๐น0 . Unlike autonomous dy- namical systems, transformer dynamics are generally layer-dependent, since each layer possesses distinct learned parameters, so ๐น๐ โ ๐น๐ for ๐ โ ๐ and the resulting system is non-autonomous. Nevertheless, many classical concepts from dynamical systems the- ory remain applicable.
12.3 Layerwise Jacobians and Local Linearization Definition 12.2 (Layerwise Jacobian). For each layer ๐น๐ , the layerwise Jacobian is ๐ฝ๐ = ๐x๐+1 /๐x๐ . The Jacobian describes the local sensitivity of subsequent activations to small per- turbations of the current activation vector; large singular values correspond to locally expansive directions, small singular values to locally contractive directions, and the eigenspectrum of ๐ฝ๐ therefore provides valuable information concerning trajectory sta- bility. Classical dynamical systems theory studies nonlinear trajectories through their local linear approximations: for a sufficiently small perturbation ๐ฟx๐ , a first-order Taylor expansion yields ๐ฟx๐+1 = ๐ฝ๐ ๐ฟx๐ + ๐(โ๐ฟx๐ โ2 ), so that to first order ๐ฝ๐ completely deter- mines the local evolution of perturbations and local trajectory stability can be analyzed independently of the global nonlinear dynamics.
12.4 Curvature and Trajectory Length Activation trajectories rarely follow straight paths through latent space; instead, succes- sive layers continually redirect the representation, motivating the introduction of curva- ture. Definition 12.3 (Discrete curvature). Let ฮ๐ = x๐+1 โ x๐ . The discrete curvature at layer ๐ is ๐ ๐ = โฮ๐+1 โ ฮ๐ โ. Curvature measures the rate at which the direction of the activation trajectory changes between successive layers: large curvature indicates rapid representational restructur- ing, small curvature relatively stable semantic evolution. Unlike the spectral quantities of the following chapter, curvature depends only on elementary Euclidean geometry. Definition 12.4 (Trajectory length). The length of an activation trajectory is โ(๐พ) = โ๐ฟโ1 ๐=0 โx๐+1 โ x๐ โ. Trajectory length measures the total representational displacement experienced dur- ing forward computation. Different transformer architectures frequently exhibit sub- stantially different trajectory lengths despite achieving similar downstream task perfor- mance, consequently providing a useful architecture-independent geometric statistic.
12.5 Local Stability and Lyapunov Analysis One of the central concepts of dynamical systems theory is the Lyapunov exponent. Al- though transformers are finite discrete systems rather than continuous flows, an analo- gous local quantity may still be defined.
Definition 12.5 (Local layerwise Lyapunov exponent). For each layer ๐, define ๐๐ = log ๐max (๐ฝ๐ ), where ๐max (๐ฝ๐ ) denotes the largest singular value of the Jacobian.
Positive values indicate local expansion, negative values contraction, and values near zero approximately neutral dynamics; these quantities describe only local behavior and should not be confused with global Lyapunov exponents defined for autonomous dy- namical systems. The singular value decomposition ๐ฝ๐ = ๐๐ ฮฃ๐ ๐๐โค provides an orthogo- nal decomposition of the local tangent space, in which the right singular vectors describe perturbation directions entering the layer, the left singular vectors the corresponding output directions, and the singular values the local amplification, permitting activation trajectories to be analyzed in terms of dominant expansion and contraction modes. Later chapters reinterpret these modes using analytic signal representations.
12.6 Relationship to Scope Geometry The mathematical constructions introduced thus far remain entirely within ordinary dynamical systems theory; no assumptions have yet been made regarding admissibil- ity, boundary-preserving projections, or Scope Geometry. Nevertheless, the motivation for introducing activation trajectories should now be apparent. The Representation In- variance Theorem (Theorem 9.10) established that semantic equivalence depends on admissible structural transformations rather than raw coordinate values, and activation trajectories provide the continuous objects from which such transformations may ulti- mately be extracted. Chapter 13 develops the mathematical machinery required for this step by constructing analytic signal representations of these trajectories, allowing phase relationships and synchronization phenomena to be studied directly.
12.7 Chapter Summary This chapter reformulated transformer computation as a discrete dynamical system evolv- ing through activation space. We introduced activation trajectories, layerwise Jacobians, local linearization, trajectory curvature, cumulative path length, local Lyapunov expo- nents, and principal tangent directions, each derived directly from standard linear al- gebra and dynamical systems theory. No spectral or wave-based interpretation has yet been introduced; the purpose here has been to establish a rigorous dynamical founda- tion on which the subsequent analytic signal framework may be constructed.
Status of Results: Chapter 12 Established background. Discrete dynamical systems, Jacobian matrices, singular value decomposition, Lyapunov analysis, differential geometry, and local linearization.
Formal developments. Activation trajectories, discrete trajectory curvature, trajectory length, local layerwise Lyapunov exponents, and their application to transformer resid- ual streams.
Empirical hypotheses. Semantic organization within transformer models is reflected more faithfully by the geometry of activation trajectories than by isolated activation vec- tors. This hypothesis motivates the spectral analysis developed in Chapter 13.
Chapter 13
Resonant Analysis of Activation Trajectories
Every signal contains both amplitude and phase. Understanding often lies not merely in what changes, but in how changes remain synchronized. โ authorโs formulation
The previous chapter reformulated transformer computation as a discrete dynami- cal system whose principal mathematical object is the activation trajectory. The present chapter extends that analysis by constructing an analytic signal representation of acti- vation trajectories using the discrete Hilbert transform, a construction well established throughout signal processing and communications theory, where analytic signals sep- arate amplitude and phase components of a real signal. The novelty proposed here is not the mathematics of analytic signals themselves, but their application to trans- former activation trajectories. Throughout this chapter, every reference to oscillation, resonance, standing waves, synchronization, or phase should be understood as an an- alytic description of latent trajectory geometry after transformation into analytic signal space; no claim is made that physical waves exist inside neural networks.
13.1 Motivation Many existing interpretability techniques analyze activation magnitude, feature direc- tion, neuron attribution, or attention weights, generally treating successive layer acti- vations as independent observations or static geometric objects. Activation trajectories suggest a different viewpoint: once activations are interpreted as ordered signals evolv- ing through layer depth, additional mathematical quantities become available, includ- ing instantaneous phase, phase synchronization, local frequency, amplitude envelopes, coherence, and spectral stability. These quantities are standard throughout signal anal- ysis but have been only minimally explored within transformer representations.
13.2 Analytic Signal Representation Consider an activation trajectory ๐พ = (x0 , x1 , โฆ , x๐ฟ ). Each coordinate defines a discrete real-valued signal indexed by layer depth; applying the discrete Hilbert transform pro- duces a corresponding analytic representation.
83
Definition 13.1 (Analytic activation signal). Let x๐ โ โ๐model denote the residual activa- tion at layer ๐. The associated analytic signal is z๐ = x๐ + ๐ โ(x๐ ), where โ denotes the discrete Hilbert transform applied coordinate-wise.
The analytic signal contains the original activation as its real part and its Hilbert transform as the imaginary part, together defining a complex-valued trajectory z0 , โฆ , z๐ฟ that will serve as the principal object of analysis throughout the remainder of this chap- ter.
13.3 Amplitude, Instantaneous Phase, and Synchronization The analytic representation permits every activation to be decomposed into polar coordi- nates: the instantaneous amplitude ๐ด๐ = |z๐ | and the instantaneous phase ๐๐ = arg(z๐ ). Amplitude measures the local magnitude of the activation trajectory, while phase de- scribes its relative orientation within analytic signal space; neither quantity individu- ally characterizes semantic organization, but meaningful structure is hypothesized to emerge through the interaction of phase relationships across layers and tokens. Signal processing frequently studies synchronization between multiple analytic sig- nals, a construction we adapt to activation trajectories. Given two trajectories ๐พ๐ and ๐พ๐ , (๐) (๐) their instantaneous phase difference is ฮ๐๐ = ๐๐ โ ๐๐ ; perfect synchronization corre- sponds to constant phase difference, while rapidly varying phase indicates desynchro- nization. The principal synchronization statistic employed throughout this monograph is the Phase Locking Value (PLV), widely used in neuroscience and signal processing.
1 Definition 13.2 (Phase Locking Value). Given ๐ paired observations, PLV = โฃ ๐ โ๐ ๐=1 ๐ ๐ฮ๐๐ โฃ.
The Phase Locking Value satisfies 0 โค PLV โค 1: values approaching 1 indicate strong phase coherence, values near 0 phase dispersion, and within the present framework PLV is interpreted as a quantitative measure of synchronization between evolving activation trajectories. The derivative of instantaneous phase defines the local frequency; since transformer layers are discrete we employ finite differences, ๐๐ = ๐๐+1 โ ๐๐ . Large values correspond to rapid phase evolution, small values to relatively stable semantic progression, and ๐๐ will later contribute to the detection of abrupt context transitions.
13.4 Semantic Coherence Hypothesis Having introduced the principal spectral quantities, we state the central hypothesis mo- tivating the remainder of Volume III.
Hypothesis 13.3 (Semantic coherence). Activation trajectories associated with success- ful reasoning tasks exhibit higher phase coherence than trajectories associated with hal- lucination, context collapse, or inconsistent reasoning.
This hypothesis is intentionally empirical: unlike the theorems of Chapters 8โ10, it is not proved mathematically, but motivates the experimental protocols of Chapter 15.
13.5. THE MARINE OPERATOR 85
13.5 The Marine Operator The previous quantities characterize local spectral properties. We now introduce a de- rived operator intended to summarize the overall spectral organization of an activation trajectory. Unlike the amplitude, phase, and PLV quantities above, which are standard signal-processing constructions applied to a new object, the Marine Operator is an orig- inal proposal of this monograph: a specific way of combining those quantities into a single filtering functional, motivated by acoustic and radar salience filtering. Definition 13.4 (Marine Operator, abstract form). Let ๐พ be an activation trajectory. The Marine Operator โณ(๐พ) is any functional combining instantaneous amplitude, phase evo- lution, local synchronization, and trajectory stability into a single diagnostic operator โณ โถ ฮ โ โ๐ , where ฮ denotes the space of activation trajectories. We now give one concrete instance of this abstract form, built from the quantities already introduced. Nothing in the remainder of this chapter depends on this being the only possible instance; it is presented as a candidate realization, to be evaluated experimentally in Chapter 15 alongside whatever alternatives later work proposes.
13.5.1 Doppler Velocity The instantaneous frequency ๐๐ = ๐๐+1 โ ๐๐ introduced above measures local phase change but says nothing about how that rate of change is itself evolving across depth. Borrowing the Doppler analogy from acoustic and radar signal processing, we track this second-order quantity directly. Definition 13.5 (Doppler velocity). Let ๐๐ be the unwrapped instantaneous phase at layer ๐, with wrap(๐) = mod(๐ + ๐, 2๐) โ ๐ applied to keep successive differences in (โ๐, ๐]. The Doppler velocity is the discrete layerwise derivative of the unwrapped phase trajectory, ๐ฃ๐ = wrap(๐๐+1 โ ๐๐ ) โ wrap(๐๐ โ ๐๐โ1 ). Large |๐ฃ๐ | indicates that the local frequency ๐๐ is itself changing rapidly โ a trajectory undergoing acceleration through phase space rather than settling into a stable rate โ which we take as a candidate indicator of representational restructuring beyond what ๐๐ alone captures.
13.5.2 The Salience Gate We combine phase coherence and Doppler velocity into a single gate that attenuates layers exhibiting either low coherence or high Doppler drift. Definition 13.6 (Marine salience gate). Let PLV๐ denote the Phase Locking Value com- puted in a window around layer ๐, and let ๐ฃ๐ be the Doppler velocity there. The Marine salience gate is ๐บ๐ = ๐gate (๐ผ PLV๐ โ๐ฝ |๐ฃ๐ | โ ๐พ), where ๐gate (๐ฅ) = 1/(1 + ๐โ๐ฅ ) is the logistic sigmoid, ๐ผ, ๐ฝ > 0 weight the two contribu- tions, and ๐พ is a static attenuation threshold. The gate satisfies ๐บ๐ โ (0, 1): layers with high coherence and low Doppler drift receive ๐บ๐ near 1 and are passed through largely unchanged, while layers with low co- herence or high drift receive ๐บ๐ near 0 and are attenuated. Applying the gate to the original activation by xฬ๐ = ๐บ๐ โ x๐ , and then rescaling to restore the original norm, yields a filtered trajectory in which incoherent, rapidly drifting layers contribute less to down- stream diagnostic statistics than stable, phase-locked ones.
13.5.3 Energy Conservation Filtering by ๐บ๐ changes the magnitude of x๐ by construction. The following observation records that this magnitude change can always be undone by rescaling, so that the gate reshapes direction and relative emphasis across layers without discarding activation en- ergy outright.
Proposition 13.7 (Energy conservation under rescaling). Let ๐ธ๐ = โx๐ โ22 and let xฬ๐ = ๐ ๐ ๐บ๐ x๐ with ๐ ๐ = โx๐ โ2 /(โ๐บ๐ x๐ โ2 + ๐) for small ๐ > 0. Then โxฬ๐ โ2 โ โx๐ โ2 as ๐ โ 0, so the rescaled filtered activation recovers the original energy ๐ธ๐ exactly in the limit.
Proof. Since ๐บ๐ is a positive scalar, โ๐บ๐ x๐ โ2 = ๐บ๐ โx๐ โ2 . Substituting ๐ ๐ gives โxฬ๐ โ2 = โx๐ โ2 ๐ ๐ ๐บ๐ โx๐ โ2 = โx โ , which tends to โx๐ โ2 as ๐ โ 0. โx๐ โ2 + ๐/๐บ๐ ๐ 2 This is a normalization convenience, not a deep result: it guarantees that applying the Marine filter does not, by itself, cause activation magnitudes to decay or explode across layers, which matters for using ๐บ๐ as a diagnostic weight rather than a replace- ment for the original representation. We state it as a proposition rather than a theorem because its content is a routine renormalization identity, not a structural claim about admissibility. The precise choice of ๐ผ, ๐ฝ, ๐พ, and the decision to treat โณ(๐พ) as (๐บ0 , โฆ , ๐บ๐ฟ ) or as some summary statistic of it, are left as implementation choices for Chapter 14, which realizes this operator as part of the diagnostic pipeline.
13.6 Interpretive Remarks Before proceeding, three observations bear on how these quantities should be read. None of the spectral quantities introduced here require the existence of physical oscil- lations inside transformer networks, since the Hilbert transform constructs an analytic representation of arbitrary signals regardless of their physical origin. Resonance should likewise be interpreted mathematically rather than metaphysically: within this mono- graph it denotes persistent synchronization patterns observed in transformed activation trajectories, a property of the analytic representation rather than a physical wave prop- agating through hardware. Finally, the quantities introduced here remain independent of Scope Geometry; only in Chapter 14 will admissibility-preserving projections connect these spectral measurements to the structural framework developed earlier in the book. Where the Representation Invariance Theorem established that semantic equivalence depends on admissible transformations rather than internal coordinates, the present chapter has instead introduced mathematical quantities capable of describing how those internal coordinates evolve in the first place; these spectral quantities are therefore not themselves semantic invariants, but provide measurable observables from which admis- sibility may be inferred. Structural Semantics remains the theoretical framework, while analytic signal analysis supplies one possible family of empirical measurement opera- tors.
13.7 Chapter Summary This chapter introduced the analytic signal representation of activation trajectories. Us- ing the discrete Hilbert transform, residual stream vectors were extended into complex-
valued trajectories possessing instantaneous amplitude, phase, local frequency, and syn- chronization properties, and these were combined into a concrete instance of the Marine Operator via the Doppler velocity and the salience gate ๐บ๐ . The principal empirical proposal is that semantic coherence may be reflected by stable phase relationships be- tween evolving activation trajectories; this proposal remains explicitly hypothetical and is subjected to experimental evaluation in Chapter 15. Before those experiments can be formulated, the spectral quantities developed here must be transformed into practical diagnostic algorithms, the objective of the following chapter.
Status of Results: Chapter 13 Established background. Analytic signals, Hilbert transforms, instantaneous phase, amplitude envelopes, Phase Locking Values, and classical signal-processing methods.
Formal developments. The analytic activation signal, discrete instantaneous frequency, application of Phase Locking Values to activation trajectories, the abstract definition of the Marine Operator together with a concrete instance built from the Doppler velocity and salience gate, and the energy-conservation renormalization proposition.
Empirical hypotheses. Semantic coherence is associated with increased phase syn- chronization between activation trajectories, and the salience gateโs attenuation pattern is associated with regions of representational instability. These hypotheses are tested in the experimental framework developed in Chapters 14 and 15.
Chapter 14
Transformer Diagnostics and Algorithmic Architecture
A mathematical theory becomes scientifically useful only when its quantities can be computed. โ adapted from Lord Kelvin
The preceding chapter introduced the spectral quantities forming the empirical core of the proposed analytic framework, but did not specify how these quantities should be extracted from real transformer models. The objective of the present chapter is there- fore computational rather than theoretical: we construct an algorithmic architecture for estimating the quantities of Chapter 13 directly from transformer activation streams. Throughout this chapter we remain deliberately implementation-neutral; the algorithms described here are intended to be realizable in PyTorch, JAX, TensorFlow, or equivalent numerical environments; reference implementations are left for future companion ma- terial rather than included in this monograph.
14.1 Diagnostic Pipeline The proposed diagnostic framework consists of six sequential stages: extraction of resid- ual stream activations; construction of activation trajectories; analytic signal transforma- tion; spectral feature computation; admissibility estimation; and diagnostic reporting. Each stage operates independently of any particular language model, consequently the framework may be applied to open-weight transformer architectures regardless of pa- rameter count or tokenizer design. For a transformer model ๐ consisting of ๐ฟ layers and an input sequence ๐ = (๐ฅ1 , โฆ , ๐ฅ๐ ), forward evaluation produces the residual tensor R โ โ(๐ฟ+1)ร๐ ร๐model , whose elements r๐,๐ก record the residual activation associated with token ๐ก after completion of layer ๐; this ten- sor forms the raw observational data for all subsequent diagnostic computations. For each token ๐ก we define the activation trajectory ๐พ๐ก = (r0,๐ก , r1,๐ก , โฆ , r๐ฟ,๐ก ), and the collec- tion ฮ = {๐พ1 , โฆ , ๐พ๐ } contains one trajectory for every token within the context window; subsequent computations operate exclusively on these trajectories rather than directly on the residual tensor. Each activation trajectory undergoes analytic transformation using the construction of Chapter 13. For every layer ๐, the amplitude envelope ๐ด๐ , instantaneous phase ๐๐ ,
89
CHAPTER 14. TRANSFORMER DIAGNOSTICS AND ALGORITHMIC 90 ARCHITECTURE
instantaneous frequency ๐๐ , local phase difference ฮ๐๐ , Phase Locking Value, and tra- jectory curvature ๐ ๐ are computed, collectively defining a spectral descriptor vector s๐ก for every token trajectory.
Definition 14.1 (Diagnostic state vector). For every activation trajectory ๐พ๐ก , define d๐ก = (๐ด, ๐, ๐, PLV, ๐ , ๐), where ๐ denotes the local Lyapunov statistic of Chapter 12.
The diagnostic state vector summarizes the local dynamical behavior of an activation trajectory; subsequent statistical procedures operate on d๐ก rather than on individual ob- servables.
14.2 Admissibility Estimation and Context Collapse The central objective of the diagnostic framework is not merely to compute spectral quan- tities but to estimate semantic admissibility.
Definition 14.2 (Admissibility estimator). An admissibility estimator is a mapping ๐ โถ d โ [0, 1] assigning each diagnostic state vector an estimated probability of admissible semantic evolution.
The precise form of ๐ is intentionally left unspecified. Different implementations may employ probabilistic classifiers, Bayesian estimators, neural predictors, or determin- istic threshold rules, allowing the mathematical framework developed in earlier chap- ters to remain independent of particular machine-learning techniques. The salience gate ๐บ๐ of Chapter 13 is one such candidate: setting ๐(d๐ก ) to the layerwise average of ๐บ๐ over a tokenโs trajectory yields a fully specified, if simple, admissibility estimator built entirely from quantities already defined; nothing in the remainder of this chapter de- pends on that particular choice, and the experiments of Chapter 15 evaluate it against the alternatives listed there. Whichever estimator is chosen, treating ๐(d๐ก ) as a probabil- ity of admissible evolution presupposes a formal measure of admissibility over semantic space in the first place; Appendix G constructs exactly such a measure, together with the entropy and mutual-information quantities that a more statistically grounded estimator than the salience gate might use in its place.
Definition 14.3 (Context collapse). A context collapse event occurs whenever ๐(d๐ก ) < ๐ for some admissibility threshold ๐ โ (0, 1).
The threshold ๐ is a hyperparameter whose optimal value depends on the applica- tion domain. Rather than identifying isolated incorrect tokens, this definition identifies regions of semantic instability within the evolving activation trajectory.
14.3 Computational Complexity The proposed diagnostic framework introduces relatively modest additional computa- tional cost. Residual stream extraction requires no additional forward passes, trajectory construction is linear in both sequence length and transformer depth, and analytic sig- nal computation using the Fast Fourier Transform requires ๐(๐ฟ log ๐ฟ) operations per activation coordinate. The overall complexity of the diagnostic pipeline is therefore ๐(๐ฟ๐ ๐model ) up to logarithmic factors associated with Hilbert transform implementa- tion, scaling linearly with the size of the activation tensor.
14.4 Case Study: Multi-Head Attention Residuals The preceding section compared the diagnostic architecture of this chapter with existing interpretability techniques at the level of purpose. It is also useful to examine a single recent architectural proposal in enough detail to see the Scope Calculus vocabulary of Chapters 6โ9 applied to something concrete. This section reads Luo et al. [34] through that vocabulary. The reading is offered as an interpretation, not as a formal result: noth- ing here constructs an admissibility-preserving projection in the sense of Definition 9.7 and verifies its hypotheses, so no claim is made that the architecture instantiates Scope in the way Theorem 9.10 would require. The value of the exercise is illustrative โ it shows what the Scope Calculus vocabulary buys a reader confronted with an ordinary architecture paper, and where that vocabulary runs out.
14.4.1 The Architecture An ordinary transformer residual stream accumulates layer outputs by addition, โโ = โโโ1 + ๐โ (โโโ1 ), so that an earlier contribution ๐ ๐ remains causally present in โโ but is no longer separately addressable: it has been folded into a running sum rather than retained as a distinguishable object. Attention Residuals [34] relax this by keeping the sequence of prior layer outputs ๐โ = (๐ 0 , โฆ , ๐ โ ) available and retrieving from it via a learned softmax query, so that each layer reads a weighted combination โ๐ ๐ผ๐ ๐ ๐ of its full depth history rather than only its immediate predecessor. Multi-Head Attention Residuals (MHAR) then observe that this retrieval query is shared across the entire representational width: every feature subspace of โโ is forced to read the same weighted combination ๐ผ of the history, even when different subspaces would prefer to weight the same history differently. MHAR reshapes the single query into ๐ป per-subspace queries, giving each subspace its own softmax over the depth history, at essentially no additional parameter or compute cost; the ordinary Attention Residuals architecture is recovered exactly as the special case ๐ป = 1. Trained from scratch across several model scales, MHAR improves validation loss over both the plain residual stream and single-query Attention Residuals, with the improvement reported to grow with model width.
14.4.2 Reading the Architecture as Scope Retrieval The depth history ๐โ = (๐ 0 , โฆ , ๐ โ ) available at retrieval time reads naturally as an ad- missible source family in the sense of Chapter 6: at each layer, the retrieval operation must select from among several available prior states rather than being handed a single determined predecessor, much as a child scopeโs Interior consists of several admissible states rather than one. The retrieval itself, a weighted combination followed by use in the next sublayer, matches the two-step pattern of โฌ๐ followed by ๐๐ already used in Appendix C to interpret function application: the history is first bound into a candidate combination, and that combination is then collapsed into the single vector the next layer consumes. Read this way, ordinary Attention Residuals apply one ๐๐ operation across the en- tire width of โโ : a single admissible combination is selected, and every subspace of the output is constructed from that same combination. MHAR instead applies ๐ป separate collapse operations, one per subspace, each free to select a different admissible combi- nation of the same history. Chapter 7 already distinguishes a single global admissibility check from a factorized one; MHAR is a concrete instance of the latter, applied not to the primitive alphabet itself but to the routing mechanism built on top of it.
CHAPTER 14. TRANSFORMER DIAGNOSTICS AND ALGORITHMIC 92 ARCHITECTURE
14.4.3 What the Reading Clarifies, and What It Does Not
The reading clarifies why the single-query architecture pays an increasing cost as width grows: forcing every subspace through one ๐๐ is, in the vocabulary of this monograph, an admissibility restriction that becomes more binding as the number of genuinely dif- ferent subspaces increases, since more of them are made to accept a combination that was not built for them specifically. It also clarifies why ๐ป = 1 recovering the original ar- chitecture exactly is the right design choice from this monographโs perspective: it is the discipline, familiar from the zero-initialized gates discussed in Chapter 1โs distinction between admissible and inadmissible transitions, of introducing a structural change as an identity transformation before allowing it to diverge. The reading does not establish that MHARโs retrieval operators satisfy the boundary- preservation conditions of Definition 9.7, nor that the Representation Invariance Theo- rem applies to a comparison between the single-query and multi-query architectures. Doing so would require specifying what boundary constraints the depth history is re- quired to preserve and verifying that both retrieval mechanisms respect them, which is an empirical and architectural question this monograph does not address. The case study should therefore be read as showing that an independently developed, empiri- cally motivated architecture exhibits the same qualitative shape โ a single shared col- lapse operation restricting what a heterogeneous system can express, relaxed by factor- izing that collapse per subspace โ that Chapters 6โ9 develop abstractly, not as evidence for any theorem in this monograph.
14.5 Relationship to Existing Interpretability Methods
The proposed diagnostic architecture complements rather than replaces existing inter- pretability techniques: attention visualization identifies communication pathways be- tween tokens, activation patching identifies causal computational circuits, sparse au- toencoders identify approximately monosemantic features, and logit attribution esti- mates token-level prediction contributions, whereas the present framework analyzes the global dynamical evolution of representations. These approaches are therefore largely orthogonal and may be combined within a unified interpretability workflow.
14.6 Limitations
Several limitations should be emphasized. The diagnostic state vector provides descrip- tive rather than causal information โ high phase synchronization does not by itself im- ply successful reasoning. Admissibility estimation depends on the quality of the chosen estimator ๐, and poor statistical estimation may produce unreliable diagnostic results even if the underlying mathematical framework is correct. Different transformer archi- tectures may also exhibit substantially different spectral statistics despite similar down- stream behavior, so normalization procedures will likely be required before cross-model comparisons become meaningful. These limitations motivate the controlled experimen- tal framework developed in the following chapter.
14.7. CHAPTER SUMMARY 93
14.7 Chapter Summary This chapter translated the mathematical quantities of Chapter 13 into a practical com- putational pipeline. Beginning with residual stream extraction, activation trajectories were transformed into analytic signals, summarized through diagnostic state vectors, and mapped to estimated admissibility scores capable of identifying potential regions of semantic instability. None of these procedures establishes the empirical validity of the proposed framework; they specify only how the relevant measurements may be com- puted. The scientific evaluation of these measurements requires carefully controlled experiments, statistical hypothesis testing, and explicit falsification criteria, the subject of the next chapter.
Status of Results: Chapter 14 Established background. Residual stream extraction, computational complexity anal- ysis, signal processing algorithms, statistical feature extraction, and the Attention Resid- uals and Multi-Head Attention Residuals architectures of Luo et al. [34].
Formal developments. The diagnostic pipeline, diagnostic state vector, admissibility estimator, and formal definition of context collapse as an admissibility-threshold event.
Interpretive discussion. ยง14.4 reads Multi-Head Attention Residuals through the Scope Calculus vocabulary of Chapters 6โ9, as an illustration rather than a theorem: no admissibility- preserving projection is constructed for it, and no claim of the monograph depends on the reading.
Empirical hypotheses. Diagnostic quantities derived from activation trajectories can identify regions of semantic instability prior to observable output failure. The validity of this hypothesis is evaluated experimentally in Chapter 15.
CHAPTER 14. TRANSFORMER DIAGNOSTICS AND ALGORITHMIC 94 ARCHITECTURE
Chapter 15
Experimental Predictions, Controls, and Falsification Protocols
The value of a scientific theory lies not merely in what it explains, but in what observations could demonstrate it to be incorrect. โ Karl Popper (adapted)
The preceding chapters introduced the mathematical and computational framework of Structural Semantics for transformer analysis. The present chapter addresses the cen- tral scientific question: can these proposed quantities be shown to possess explanatory power beyond existing interpretability methods? Unlike previous chapters, which pri- marily developed mathematical structures, this chapter develops an empirical research program in which every hypothesis is explicitly falsifiable. Failure of the stated predic- tions should be interpreted as evidence against the proposed framework rather than evidence requiring ad hoc modification.
15.1 Scientific Methodology The empirical program developed in this monograph follows four guiding principles: every proposed diagnostic quantity must be measurable using existing transformer ar- chitectures; every hypothesis must admit explicit statistical rejection; comparisons must be performed against established baseline methods; and negative results are regarded as scientifically valuable because they delimit the scope of applicability of the proposed framework. These principles reflect the methodological distinction introduced in Chap- ter 1 between formal mathematics and empirical science.
15.2 General Experimental Design Let โณ = {๐1 , โฆ , ๐๐พ } denote a collection of transformer models. Experiments are conducted over a benchmark collection ๐ = {๐ท1 , โฆ , ๐ท๐ } containing reasoning tasks, mathematical proofs, code generation, multi-document question answering, and long- context inference problems. For every prompt ๐ฅ โ ๐, each model produces output tokens, residual stream activations, activation trajectories, diagnostic state vectors, and admissibility estimates, forming the dataset used throughout the remainder of the chap- ter.
95
CHAPTER 15. EXPERIMENTAL PREDICTIONS, CONTROLS, AND 96 FALSIFICATION PROTOCOLS
15.3 Hypothesis I: Phase Coherence and Semantic Stability Hypothesis 15.1 (Phase coherence hypothesis). Successful reasoning trajectories ex- hibit larger average Phase Locking Values than trajectories containing semantic incon- sistency or hallucination: PLVcorrect > PLVincorrect .
A benchmark of at least ๐ โฅ 1000 paired reasoning traces should be constructed, each independently classified by human annotators as correct, partially correct, hallu- cinated, or internally inconsistent, and average Phase Locking Values then computed for each category. The null hypothesis is ๐ป0 โถ ๐correct = ๐incorrect against the alternative ๐ป1 โถ ๐correct > ๐incorrect ; depending on the empirical distribution, Studentโs ๐ก-test, Welchโs test, or the MannโWhitney ๐ test may be employed, with statistical significance assessed throughout this monograph at ๐ผ = 0.01. The hypothesis is rejected whenever ๐ > 0.01, or when the observed effect size is practically negligible despite statistical significance.
15.4 Hypothesis II: Frequency Drift and Context Collapse Hypothesis 15.2. Abrupt semantic inconsistency is accompanied by statistically detectable changes in instantaneous frequency.
Let ๐๐ denote the instantaneous frequency of Chapter 13; large discontinuities |๐๐+1 โ ๐๐ | are hypothesized to precede observable failures of reasoning. Long-context reason- ing tasks are modified through controlled insertion of contradictory premises, and the resulting activation trajectories are analyzed before, during, and after the injected con- tradiction and compared with ordinary successful reasoning traces. If no statistically significant frequency differences are observed between successful and contradictory rea- soning trajectories, the hypothesis is rejected.
15.5 Hypothesis III: Admissibility and Error Prediction Hypothesis 15.3. Declining admissibility estimates precede observable output errors.
Rather than identifying hallucinations after incorrect tokens have already appeared, admissibility estimation is hypothesized to detect semantic instability during internal computation. For every generated sequence, the first observable output error is recorded, diagnostic quantities computed before that token are analyzed, and receiver operating characteristic curves and precisionโrecall statistics are calculated to determine predic- tive performance, summarized using the area under the ROC curve, precision, recall, the F1 score, and the Matthews Correlation Coefficient.
Example 15.4 (What this hypothesis does and does not claim). Suppose the experiment above returns a strong result: admissibility estimates decline detectably several layers before an incorrect token is sampled, with an area under the ROC curve well above chance. This would establish, for the specific models and tasks tested, that the diag- nostic quantities of Chapter 14 carry predictive information about upcoming output errors โ an empirical finding, reported with the statistical qualifications Chapter 15 requires throughout. It would not, by itself, establish a working intervention: nothing in this chapter specifies what a system should do upon detecting declining admissibility, whether halting, revising, or continuing is preferable in a given context, or what compu- tational overhead an intervention would add beyond the detection step itself. A system
that halts and backtracks on a declining-admissibility signal is a natural extension to propose, but it is a distinct research question โ one requiring its own experimental vali- dation and its own comparison against existing decoding strategies โ not a consequence already established by Hypothesis III.
15.6 Control Experiments Perhaps the greatest danger facing any new diagnostic quantity is that it merely repro- duces information already contained in simpler statistics, so every experiment above must be accompanied by control analyses. Phase coherence must be distinguished from token confidence: softmax entropy is computed alongside PLV, and regression analysis determines whether Phase Locking Values retain predictive power after controlling for entropy. Large activation magnitudes might artificially influence spectral statistics, so activation vectors are normalized before analytic transformation and equivalent exper- iments are repeated using raw, normalized, and randomized activations. To determine whether synchronization reflects genuine structure rather than statistical coincidence, phase values are randomly permuted while preserving amplitude and the diagnostic framework re-evaluated, with significant performance degradation supporting the in- terpretation that phase relationships contain meaningful information. Finally, experi- ments should be repeated across transformer families differing in parameter count, to- kenizer, training corpus, attention implementation, and instruction tuning, since agree- ment across architectures would strengthen the claim that the observed phenomena reflect general computational organization rather than architecture-specific artefacts.
15.7 Baseline Comparisons The proposed framework should not be evaluated in isolation, but compared against established diagnostic approaches including attention entropy, logit confidence, activa- tion norm, perplexity, sparse autoencoder feature activation, activation patching meth- ods, and uncertainty estimation. A proposed diagnostic quantity possesses scientific value only if it provides predictive information beyond these existing baselines.
15.8 Failure Modes Several outcomes would argue against the proposed framework: Phase Locking Values showing no predictive relationship with semantic correctness; instantaneous frequency exhibiting no systematic relationship with context collapse; admissibility estimates fail- ing to predict errors before observable output failure; existing baseline methods consis- tently outperforming every proposed spectral quantity; or performance failing to gener- alize across transformer architectures. Observation of any of these outcomes should be interpreted as evidence requiring revision or rejection of the corresponding hypotheses.
15.9 Discussion The purpose of the present chapter is not to demonstrate that the proposed framework succeeds, but to specify the conditions under which success or failure may be deter- mined objectively. This distinction is essential: the mathematical framework developed
CHAPTER 15. EXPERIMENTAL PREDICTIONS, CONTROLS, AND 98 FALSIFICATION PROTOCOLS
in Volumes I and II remains valid independently of these experiments, whereas the em- pirical usefulness of the diagnostic operators of Volume III depends entirely on experi- mental evidence. The separation between formal mathematics and empirical evaluation therefore remains explicit throughout the monograph.
15.10 Chapter Summary This chapter established a complete experimental program for evaluating the diagnos- tic framework proposed throughout Volume III. Three principal hypotheses concerning phase coherence, frequency drift, and admissibility prediction were formulated together with statistical procedures, control experiments, baseline comparisons, and explicit fal- sification criteria. The completion of this empirical program concludes the technical development of Structural Semantics; the remaining volume returns to a broader per- spective, comparing the framework with neighboring fields, identifying its limits, and discussing future directions.
Status of Results: Chapter 15 Established background. Experimental design, statistical hypothesis testing, effect sizes, receiver operating characteristic analysis, and benchmark evaluation.
Formal developments. Explicit experimental protocols, statistical controls, baseline comparisons, and falsification criteria for evaluating the proposed diagnostic frame- work.
Empirical hypotheses. Phase synchronization, instantaneous frequency, and admissi- bility estimation provide measurable indicators of semantic coherence whose predictive value exceeds existing baseline diagnostics. These hypotheses remain subject to experi- mental confirmation or rejection.
Part IV
Synthesis
99
Chapter 16
Cross-Disciplinary Integration and Comparative Landscape
Science progresses not merely by constructing new theories, but by understanding precisely how those theories relate to existing ones. โ authorโs formulation
The preceding three volumes developed Structural Semantics from first principles. The purpose of the present chapter is fundamentally different: rather than introducing additional mathematical structures, we situate Structural Semantics within the broader scientific landscape. Throughout this chapter our objective is comparison rather than competition; Structural Semantics is not presented as replacing existing frameworks, but as addressing a different class of questions through a different mathematical language. To facilitate comparison, we evaluate each framework according to four questions: what constitutes a semantic object, how is semantic change represented, how is correctness evaluated, and what mathematical structures are fundamental.
16.1 Predictive Processing and Active Inference Predictive Processing and Active Inference describe cognition as the continuous mini- mization of prediction error under hierarchical generative models [11, 18]. Structural Semantics shares with this tradition an emphasis on dynamics over static representation and on hierarchical organization, but its mathematical objectives differ: Predictive Pro- cessing centers on probabilistic inference and free-energy optimization, while Structural Semantics studies admissibility-preserving transformations independent of probabilis- tic interpretation, with prediction error playing no primitive role in the Scope Calculus. Consequently the two approaches should be regarded as complementary, one empha- sizing optimization and the other structural constraint.
16.2 Mechanistic Interpretability Mechanistic interpretability attempts to explain neural computation by identifying inter- nal circuits responsible for observable model behavior [15, 42, 44]. Structural Semantics aligns closely with this program in seeking internal mathematical explanations rather
101
CHAPTER 16. CROSS-DISCIPLINARY INTEGRATION AND COMPARATIVE 102 LANDSCAPE
than behavioral descriptions, but the two frameworks operate at different levels of de- scription: circuit analysis identifies computational components such as neurons, atten- tion heads, and sparse features, while Scope Geometry analyzes the global organization of admissible trajectories connecting those components.
16.3 Information Geometry Information geometry studies statistical models using differential geometric methods, introducing Fisher information, geodesics, divergence functions, and statistical mani- folds [1]. Both frameworks employ manifold language and investigate geometric orga- nization, but information geometry derives its structure from probability distributions while Scope Geometry derives its structure from admissibility constraints, so the corre- sponding metrics encode fundamentally different objects โ statistical distinguishability in one case, preservation of contextual organization in the other.
16.4 Category Theory Category theory provides one of the principal mathematical languages of this mono- graph, though Structural Semantics employs only a deliberately restricted subset of cat- egorical machinery: objects represent bounded scope geometries and morphisms rep- resent admissible context-preserving transformations, with no assumptions regarding limits, exponentials, adjunctions, enriched categories, or higher categories unless explic- itly introduced. This structural minimalism distinguishes the present framework from category-theoretic approaches seeking maximal abstraction; category theory functions here primarily as an organizational language for admissible transformations.
16.5 Formal Language Theory Formal language theory studies symbolic computation through grammars, automata, rewrite systems, and logical calculi. Structural Semantics inherits several ideas from this tradition: the primitive alphabet ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } functions analogously to a rewriting alphabet, canonical normal forms resemble rewriting theory, and admissible transformations resemble typed derivations. The principal distinction concerns seman- tics: formal language theory primarily studies syntactic derivability, while Structural Semantics studies preservation of contextual admissibility, so that syntax becomes one particular instance of a broader structural framework.
16.6 Classical Structural Semantics The terminology adopted in this monograph inevitably overlaps with the classical lin- guistic tradition known as Structural Semantics [12, 36], discussed at length in Chapter 2. Both approaches emphasize relationships rather than isolated entities and reject purely referential theories of meaning, but the mathematical scope differs substantially: classi- cal structural semantics studies relationships between linguistic units, while the present work studies admissible transformations within arbitrary computational systems, so the relationship should be understood as one of historical continuity rather than mathemat- ical identity.
16.7 Process Philosophy Chapter 10 established a formal translation functor between Whiteheadian process on- tology and Scope Geometry, demonstrating that many operational aspects of process philosophy admit structural reformulation. Structural Semantics neither adopts nor re- jects Whiteheadโs metaphysical commitments, identifying instead those components ca- pable of operational formalization; Subjective Form remained outside the image of the translation functor, marking the precise boundary between phenomenological philoso- phy and structural mathematics.
16.8 Large Language Models Modern large language models constitute the primary computational domain investi- gated throughout Volume III, but Structural Semantics should not be regarded as a theory of transformers alone. The Representation Invariance Theorem established that semantic organization depends on admissible transformations rather than specific ar- chitectures, so nothing in principle restricts the framework to transformers; recurrent neural networks, graph neural networks, state-space models, symbolic systems, and hybrid neuro-symbolic architectures may equally be analyzed whenever appropriate admissibility-preserving projections can be constructed.
16.9 Synthesis Comparison across these diverse frameworks reveals a common pattern: every approach identifies an important aspect of semantic organization โ Predictive Processing empha- sizes optimization, mechanistic interpretability internal computation, information ge- ometry statistical structure, formal language theory symbolic derivation, classical struc- tural semantics lexical relations, and process philosophy becoming. Structural Seman- tics contributes a different perspective, whose central object is neither probability, syn- tax, prediction, consciousness, nor optimization, but the preservation of admissible struc- tural transformations under changing representations, occupying a distinct position within the broader landscape of semantic theory.
16.10 Chapter Summary This chapter situated Structural Semantics within the wider scientific and philosoph- ical literature. Rather than competing with existing approaches, the framework was shown to address a complementary class of questions centered on admissible structural transformation. The next chapter explicitly enumerates the boundary conditions, scope limits, and non-claims of Structural Semantics, clarifying exactly what the framework does, and does not, attempt to explain.
Status of Results: Chapter 16 Established background. Predictive Processing, Active Inference, mechanistic inter- pretability, information geometry, formal language theory, classical structural semantics, category theory, and process philosophy.
CHAPTER 16. CROSS-DISCIPLINARY INTEGRATION AND COMPARATIVE 104 LANDSCAPE
Formal developments. Comparative positioning of Structural Semantics relative to ex- isting frameworks and clarification of its distinctive mathematical focus.
Empirical hypotheses. None. This chapter is comparative and interpretive rather than mathematical or experimental.
Chapter 17
Scope Limits, Boundary Conditions, and Non-Claims
The strength of a mathematical theory is measured not only by what it explains, but equally by what it deliberately leaves unexplained. โ authorโs formulation
Every mathematical theory possesses boundaries, and one of the distinguishing char- acteristics of mature scientific frameworks is the explicit recognition of those questions lying beyond their intended domain of applicability. The purpose of this chapter is therefore not to weaken the preceding results but to strengthen their interpretation by specifying exactly what Structural Semantics does and does not claim. Throughout, the distinction introduced in Chapter 1 between formal mathematics, empirical hypotheses, and philosophical interpretation remains strictly enforced.
17.1 The Scope of Structural Semantics Structural Semantics is fundamentally a theory of admissible transformations. Its pri- mary mathematical object is neither consciousness, intelligence, meaning in the ordi- nary linguistic sense, nor biological cognition, but the preservation of structural organi- zation under transformations constrained by explicit boundary relations. Accordingly, every theorem proved throughout Volumes IโIII concerns admissibility, context preser- vation, structural stability, categorical composition, representation invariance, activa- tion trajectory geometry, or diagnostic observables; no theorem proved in this mono- graph establishes any claim beyond these domains.
17.2 Consciousness, Truth, Agency, and Ethics Structural Semantics makes no claim that systems possessing high semantic capacity are conscious, nor that conscious systems necessarily possess maximal structural se- mantic organization; phenomenal consciousness concerns subjective experience, while Structural Semantics concerns admissible transformations, and whether these domains ultimately intersect remains an open philosophical question outside the mathematical scope of the present framework.
105
Proposition 17.1 (Non-claim: consciousness). None of the theorems established in Chap- ters 6โ15 implies the existence, absence, measurement, simulation, or explanation of phenomenal consciousness.
Proof. Immediate. Neither phenomenal predicates nor subjective experience appear among the primitive symbols of the Scope Calculus, so no theorem concerning these concepts can be derived within the formal system.
The Scope Calculus likewise evaluates structural coherence rather than factual accu- racy: a perfectly coherent fictional narrative may satisfy every admissibility constraint while describing events that never occurred, while a scientifically correct explanation may violate contextual constraints if contradictory assumptions are introduced during reasoning, so truth and structural coherence remain logically independent, reflecting the separation introduced in the Disentanglement Operator (Definition 1.6). Nor does the framework provide a theory of agency: it neither attributes nor denies goals, inten- tions, preferences, values, autonomy, or moral status, and any computational system ca- pable of executing admissible transformations may be analyzed within the Scope Calcu- lus regardless of whether it is autonomous, externally controlled, deterministic, stochas- tic, biological, or mechanical. Nor does anything within the Scope Calculus determine whether a computational system ought to perform a particular action: mathematical ad- missibility is distinct from ethical acceptability, and an admissible transformation may be morally undesirable just as an ethically desirable action may require violating pre- viously established contextual assumptions, so questions of ethics require additional normative frameworks not developed in this monograph.
17.3 General Intelligence and Biological Realism The framework should not be interpreted as a definition of general intelligence: high structural semantic capacity does not imply competence across every domain, and sys- tems may achieve impressive task performance while remaining structurally fragile, since general intelligence depends on many additional considerations including learn- ing, abstraction, planning, memory, transfer, and adaptation, of which Structural Seman- tics addresses only one component. Nor is Structural Semantics intended as a biological theory: although examples throughout the text frequently reference neural networks and human cognition, no claim is made that the brain computes using the primitive alphabet {๐ซ, โ, โฌ๐ , ๐๐ }, or that cortical dynamics implement the Scope Calculus liter- ally; the framework should instead be interpreted as an abstract mathematical language capable of describing admissible organization across multiple possible substrates.
17.4 Completeness and Boundary Conditions No mathematical theory describing complex adaptive systems should be expected to explain every aspect of semantic behavior, and the present framework is intentionally incomplete: future extensions may incorporate probabilistic admissibility, temporal log- ics, multimodal reasoning, stochastic scope evolution, higher categorical structure, and continuous boundary fields, none of which are developed in the present monograph. The formal results proved throughout this work depend on explicit assumptions, among the most important being the existence of bounded scope geometries, admissibility- preserving transformations, well-defined primitive operations, canonical normal forms,
admissibility-preserving projection maps where required, and finite computational his- tories for reconstruction results. Violation of these assumptions invalidates the corre- sponding theorems, so every application of Structural Semantics should begin by veri- fying that these hypotheses are satisfied.
17.5 Research Programme Versus Finished Theory It is important to distinguish between those components already established mathemat- ically and those proposed for future empirical investigation: Volumes I and II primarily developed formal mathematics, while Volume III introduced computational diagnostics together with explicit experimental hypotheses whose value only empirical investiga- tion can determine. Accordingly, this monograph should be understood simultaneously as a completed mathematical framework and an incomplete empirical research program, and these two aspects should not be conflated.
17.6 Discussion Scientific theories often become controversial not because of their formal content but because claims exceeding that content are attributed to them. Considerable effort has therefore been devoted throughout this monograph to separating mathematical the- orem from philosophical interpretation and empirical hypothesis. This distinction is methodological rather than merely stylistic: readers may accept the formal mathemat- ics while remaining skeptical of the empirical proposals, and future experiments may invalidate portions of Volume III without affecting the mathematical consistency of the Scope Calculus developed in Volumes I and II. Such independence is a strength rather than a weakness.
17.7 Chapter Summary This chapter explicitly identified the limits of Structural Semantics. The framework does not claim to explain consciousness, truth, agency, ethics, or general intelligence, and pro- vides neither a comprehensive theory of cognition nor a philosophy of mind; instead it contributes a formal mathematical language for describing admissible structural trans- formations together with an empirical program for investigating their computational re- alization. Having established these boundaries, the monograph concludes with a final synthesis of its principal mathematical, computational, and philosophical contributions.
Status of Results: Chapter 17 Established background. The importance of explicit scope conditions, methodologi- cal restraint, and boundary assumptions in mathematical and scientific theories.
Formal developments. Formal clarification of the domain of applicability of Structural Semantics together with explicit non-claims concerning consciousness, truth, agency, ethics, biological realism, and general intelligence.
Empirical hypotheses. None. This chapter establishes interpretive limits rather than new scientific predictions.
Chapter 18
Conclusion: The Architecture of Structural Semantics
Mathematics does not close inquiry. It gives inquiry a structure within which to continue. โ authorโs formulation
The central objective of this monograph has been to investigate whether semantic organization can be described independently of particular computational substrates, bi- ological mechanisms, or philosophical commitments concerning consciousness. Begin- ning with this question, we developed a mathematical framework grounded in admis- sibility, structural preservation, and bounded scope geometries. The resulting theory, Structural Semantics, proposes that semantic organization is fundamentally a property of admissible transformations rather than a consequence of any particular physical im- plementation.
18.1 From Philosophy to Mathematics The opening chapters began with an epistemic clarification. Discussions of artificial intelligence frequently employ the term โunderstandingโ while referring to multiple distinct concepts including phenomenal consciousness, factual correctness, intentional agency, linguistic fluency, and contextual reasoning. Chapter 1 introduced the Disentan- glement Operator to separate these concepts into independent domains, permitting the construction of a mathematical theory without requiring prior resolution of longstand- ing philosophical debates concerning the nature of consciousness or intentionality. The objective was therefore modest but precise: rather than explaining mind itself, the the- ory sought to characterize one particular aspect of intelligent systems, the preservation of structural context under transformation.
18.2 The Scope Calculus The central mathematical contribution of the monograph is the Scope Calculus. Begin- ning with bounded scope geometries, primitive scope operations, boundary tensors, and admissibility relations, we developed a minimal algebra capable of describing context- preserving computation. Subsequent chapters demonstrated that these primitive oper- ations admit categorical organization, canonical normal forms, and rigorous rewriting
109
CHAPTER 18. CONCLUSION: THE ARCHITECTURE OF STRUCTURAL 110 SEMANTICS
properties, and Representation Invariance established that semantic organization de- pends on admissible structural transformations rather than particular representational substrates. Collectively these results provide the formal mathematical foundation of Structural Semantics.
18.3 Geometry of Semantic Organization Traditional semantic theories frequently emphasize symbols, logical propositions, lexi- cal relations, or statistical associations. Structural Semantics instead approaches seman- tic organization through geometry: context is represented by bounded scope geome- tries, semantic evolution becomes a sequence of admissible transformations, reasoning corresponds to trajectories constrained by boundary relations, and correctness becomes preservation of admissibility rather than merely production of plausible outputs. This geometric perspective permits methods from topology, category theory, algebra, and dynamical systems to contribute naturally to the study of semantic organization.
18.4 Computational Realization Volume III demonstrated one possible computational realization of the abstract mathe- matical framework: transformer residual streams were interpreted as activation trajecto- ries evolving through high-dimensional spaces, analytic signal methods extended these trajectories into complex-valued representations possessing amplitude, phase, synchro- nization, and frequency, and diagnostic algorithms were proposed for estimating ad- missibility from observable activation dynamics. Importantly, these computational con- structions should not be confused with the mathematical foundations developed earlier: the Scope Calculus remains substrate independent, and the transformer framework rep- resents one application rather than the definition of Structural Semantics itself.
18.5 Scientific Status Throughout the monograph, considerable effort has been devoted to maintaining clear distinctions among formal mathematical theorems, proved deductively within the Scope Calculus; computational algorithms derived from those mathematical structures; and empirical hypotheses requiring experimental evaluation. The mathematical validity of the Scope Calculus does not depend on the success of the experimental program pro- posed in Volume III, while successful empirical results would support the usefulness of the computational framework without altering the formal proofs presented earlier. This separation between mathematics and empirical science is one of the guiding principles of the entire work.
18.6 Relationship to Existing Research Structural Semantics occupies an intermediate position between several established dis- ciplines, incorporating ideas from topology, category theory, formal language theory, information geometry, signal processing, and mechanistic interpretability while remain- ing reducible to none of them. The resulting framework should be understood as a syn- thesis rather than a replacement, whose objective is to provide a common mathematical
18.7. LIMITATIONS AND FUTURE DIRECTIONS 111
language capable of relating diverse approaches to semantic organization through the shared concept of admissible transformation. The preceding chapters developed the conceptual architecture of Structural Seman- tics: the epistemic framing that separates structural capacity from phenomenology, the Scope Calculus that formalizes admissible transformation, and the diagnostic apparatus that applies it to transformer representations. The appendices supplied the supporting mathematical machinery this architecture was built on rather than material set aside from it โ lambda calculus and abstract rewriting theory in Appendices A and B, cate- gorical semantics in Appendices C and D, topology and sheaf theory in Appendices E and F, and measure-theoretic and probabilistic foundations in Appendix G, together with the collected proofs of Appendix I. Each was invoked at the point in the main text where its results became relevant rather than gathered as an afterthought, and together they are intended to show that the framework rests on a unified mathematical founda- tion rather than a collection of loosely related techniques assembled for the occasion.
18.7 Limitations and Future Directions Structural Semantics does not provide a theory of consciousness, ethics, intentionality, biological cognition, or general intelligence, nor does it claim that transformer activation trajectories literally constitute physical waves or that analytic signal methods uniquely capture semantic organization; these questions remain outside the formal scope of the present work, and recognition of these limits strengthens rather than weakens the math- ematical precision of the framework itself. The present work opens numerous avenues for future research. From the mathe- matical perspective, extensions to stochastic scope geometries, continuous admissibility fields, enriched categories, and higher-dimensional rewriting systems appear particu- larly promising. From the computational perspective, large-scale empirical evaluation of activation trajectory diagnostics remains an immediate priority, with further oppor- tunities in multimodal foundation models, autonomous agents, program synthesis, sci- entific reasoning, formal verification, and hybrid symbolicโneural architectures. More broadly, the framework suggests that admissibility may provide a common mathemati- cal language for analysing structured computation across a wide variety of domains.
18.8 What Structural Semantics Does Not Claim Chapter 17 stated the non-claims of this monograph at length and gave the reasoning behind each. They are worth restating briefly here, at the point a reader is most likely to be forming an overall impression of what the book has argued, since a summary chapter is exactly where an unstated ambition is most easily read into a text that never asserted it. Structural Semantics does not claim to have shown that any computational system is conscious, that consciousness is unnecessary for understanding, that its formal results settle any question in the philosophy of mind, or that admissibility is a complete or even approximately complete theory of meaning. It does not claim that the resonant reading of activation trajectories in Volume III describes a physical process rather than an ana- lytic one, and it does not claim that the empirical hypotheses proposed there are estab- lished rather than proposed. What has been shown, within the scope this monograph set for itself, is narrower and, on its own terms, complete: that admissible structural transformation admits a rigorous mathematical treatment independent of these further
CHAPTER 18. CONCLUSION: THE ARCHITECTURE OF STRUCTURAL 112 SEMANTICS
questions, and that this treatment generates concrete, falsifiable predictions when ap- plied to contemporary neural architectures. Everything beyond that remains exactly what Chapter 17 already said it was โ open.
18.9 Final Remarks The principal contribution of this monograph is not the assertion that semantic orga- nization has been completely explained; no single mathematical framework should be expected to resolve every question concerning language, cognition, or intelligence. In- stead, the contribution lies in demonstrating that one important aspect of semantic orga- nization can be studied independently, formalized precisely, and connected to existing areas of mathematics through a common structural language. Whether future empiri- cal investigation ultimately confirms or rejects the diagnostic program proposed in Vol- ume III, the mathematical framework developed throughout this work remains available as a language for reasoning about admissible transformations and context-preserving organization. In this sense, Structural Semantics should be viewed less as the end of a research program than as its beginning.
Principal Contributions of This Monograph For clarity, the principal original contributions of this work are summarized as follows: the formal definition of Structural Semantics as a substrate-independent theory of ad- missible structural transformations; the Scope Calculus together with its primitive op- erational alphabet; Bounded Scope Geometry and the Boundary Constraint Tensor; the categorical formulation of admissible transformations through the category Scope; the Representation Invariance Theorem establishing semantic equivalence under admissibility- preserving projections; the interpretation of transformer residual streams as activation trajectories within a dynamical systems framework; the application of analytic signal methods to activation trajectory analysis; a computational diagnostic architecture based on admissibility estimation; and an explicit program of empirical hypotheses, statistical controls, and falsification criteria for evaluating these diagnostic operators.
Closing Statement Mathematics has long provided languages for describing quantity, symmetry, computa- tion, probability, and geometry. The present work proposes that admissibility deserves recognition as an additional organizing principle through which semantic organization may be investigated. Whether this proposal ultimately becomes a foundational compo- nent of future semantic theory or remains a specialized mathematical framework will de- pend not on philosophical preference, but on the continued interaction of formal proof, computational implementation, and empirical investigation. That interaction has only begun.
Status of Results: Chapter 18 Established background. Summary and synthesis of the mathematical, computational, and empirical developments presented throughout the monograph.
Formal developments. None. This chapter consolidates previously established results rather than introducing new mathematics.
Empirical hypotheses. No new hypotheses are introduced. The chapter summarizes the existing research program and identifies directions for future investigation.
CHAPTER 18. CONCLUSION: THE ARCHITECTURE OF STRUCTURAL 114 SEMANTICS
Appendix A
Lambda Calculus and Scope Calculus
The purpose of this appendix is to establish a precise correspondence between the sim- ply typed ๐-calculus and the Scope Calculus. Rather than identifying the two systems directly, we construct a translation preserving operational behavior; the principal result is that ๐ฝ-reduction corresponds to a canonical scope transition generated by primitive scope operations. Throughout this appendix we assume familiarity with the simply typed ๐-calculus as presented in Barendregt [6] and Pierce [47].
A.1 Translation of Lambda Terms Definition A.1 (Scope translation). There exists a translation ๐ฏ โถ ฮ โ Scope defined recursively on the structure of a lambda term by ๐ฏ(๐ฅ) = S๐ฅ for a variable ๐ฅ, ๐ฏ(๐๐ฅ.๐ ) = โฌ๐ (๐ฅ, ๐ฏ(๐ )) for an abstraction, and ๐ฏ(๐ ๐ ) = ๐ซ(๐ฏ(๐ ), ๐ฏ(๐ )) for an application.
Definition A.2 (Beta reduction). For lambda terms (๐๐ฅ.๐ )๐, the ordinary reduction rule is (๐๐ฅ.๐ )๐ โ๐ฝ ๐ [๐ฅ โถ= ๐ ], substituting ๐ for free occurrences of ๐ฅ in ๐.
A.2 The Correspondence Theorem Theorem A.3 (Beta reduction correspondence). Let ๐ฏ โถ ฮ โ Scope be the translation of Definition A.1. Every beta reduction (๐๐ฅ.๐ )๐ โ๐ฝ ๐ [๐ฅ โถ= ๐ ] corresponds to an admissible primitive scope transition ๐ซ โ โฌ๐ โ ๐๐ ; equivalently, ๐ฏ((๐๐ฅ.๐ )๐) reduces, through canonical Scope Calculus reduction, to ๐ฏ(๐ [๐ฅ โถ= ๐ ]).
Proof. Application constructs a temporary evaluation scope containing the function ob- ject and the argument object, into which the Bind operator introduces a child scope where the formal parameter is locally defined. The Pop operator then transfers control into that child scope, so that Collapse can remove the temporary evaluation frame once substitution is complete. Because canonical reduction preserves admissibility (Theo- rem 9.3), the resulting scope geometry is isomorphic to the translated substitution term, so beta reduction is represented by the admissible scope transition ๐ซ โ โฌ๐ โ ๐๐ .
This correspondence should be read as a mathematical equivalence between two formal systems established through an explicit translation, not as an identification of
115
syntax: nothing here claims that ๐ซ, โฌ๐ , and ๐๐ are beta reduction, only that beta reduc- tion is faithfully represented by a specific composition of them under ๐ฏ. The remaining appendices build on this translation, progressively situating the Scope Calculus within rewriting theory, category theory, topology, sheaf theory, and measure theory.
Appendix B
The Scope Calculus as an Abstract Reduction System
The purpose of this appendix is to place the Scope Calculus within the well-developed mathematical theory of abstract reduction systems. Throughout the main body of the monograph, reduction has been treated operationally; here we establish the correspond- ing rewriting theory, proving termination, confluence, and the existence of canonical normal forms under explicit hypotheses, following the general framework of abstract rewriting systems developed by Newman and by Church and Rosser, adapted to the primitive scope algebra.
B.1 Reduction Systems Definition B.1 (Scope terms). Let ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } = {๐ซ, โ, โฌ๐ , ๐๐ } be the primitive scope alphabet. The set of finite primitive expressions is defined inductively by ๐ก โถโถ= ๐ โฃ ๐ซ(๐ก) โฃ โ(๐ก) โฃ โฌ๐ (๐ก, ๐ก) โฃ ๐๐ (๐ก).
Definition B.2 (Reduction relation). A reduction relation โโ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ ร ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ is generated by the primitive rewrite rules below. If ๐ก โ ๐ข, then ๐ถ[๐ก] โ ๐ถ[๐ข] for every admissible context ๐ถ, extending the relation to arbitrary contexts in the standard manner.
B.2 Primitive Rewrite Rules The Scope Calculus employs four fundamental rewrite rules, restated here in their fully general term-rewriting form (the elementary word-level rules of Chapter 7 are the spe- cial case where each subterm is a variable or ๐): identity elimination, โฌ๐ (๐ฅ, ๐ฅ) โ ๐ฅ; empty collapse, ๐๐ (๐) โ ๐; Pop elimination, ๐ซ(โฌ๐ (๐ฅ, ๐ก)) โ ๐ก; and refusal stability, โ(โ(๐ก)) โ โ(๐ก). These four reductions generate every higher reduction used through- out the main text.
B.3 Termination Definition B.3 (Reduction measure). Define ๐ โถ ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }โ โ โ by ๐(๐) = 0 and, recursively, ๐(๐ซ(๐ก)) = 1 + ๐(๐ก), ๐(๐๐ (๐ก)) = 1 + ๐(๐ก), ๐(โ(๐ก)) = 1 + ๐(๐ก), and ๐(โฌ๐ (๐ , ๐ก)) = 1 + ๐(๐ ) + ๐(๐ก).
117
Lemma B.4. Every primitive rewrite strictly decreases ๐.
Proof. Immediate verification for each rewrite rule: each removes at least one construc- tor while introducing no additional constructors, so ๐(๐กโฒ ) < ๐(๐ก) whenever ๐ก โ ๐กโฒ .
Theorem B.5 (Strong normalization). Every reduction sequence terminates.
Proof. Since ๐ takes values in โ, no infinite strictly decreasing chain can exist, so every reduction sequence is finite.
B.4 Confluence and Canonical Normal Forms Strong normalization alone does not guarantee uniqueness of normal forms; we there- fore investigate confluence.
Definition B.6. A reduction system is locally confluent if ๐ฅ โ ๐ฆ and ๐ฅ โ ๐ง imply the existence of ๐ค such that ๐ฆ โโ ๐ค and ๐ง โโ ๐ค.
Lemma B.7. The primitive Scope Calculus is locally confluent.
Proof. The proof proceeds by examining every critical overlap between primitive rewrite rules. Since only four primitive rules exist, the overlap set is finite, and each overlap reduces to a common descendant after at most two additional rewrite steps, so every critical pair joins.
Theorem B.8 (Confluence). The Scope Calculus is confluent.
Proof. By the preceding theorem the system is terminating, and by the preceding lemma it is locally confluent; Newmanโs Lemma therefore implies global confluence.
Corollary B.9. Every primitive scope expression possesses a unique normal form.
Proof. Confluence implies uniqueness of normal forms, and termination guarantees ex- istence.
B.5 Categorical Interpretation The reduction system constructed above induces a category whose objects are normal- form scope geometries and whose morphisms are equivalence classes of primitive reduc- tion sequences, with composition induced by concatenation followed by normalization, identity morphisms corresponding to empty reduction sequences, and associativity fol- lowing from associativity of concatenation. The rewriting system therefore supplies an alternative construction of the category Scope introduced in Chapter 8. Appendix A established a translation ๐ฏ โถ ฮ โ Scope; the present appendix has shown that ๐ฏ maps beta reduction into canonical primitive reduction, so the simply typed lambda calculus embeds into the Scope Calculus as a terminating, confluent rewrit- ing subsystem.
B.6 Discussion The importance of this appendix extends beyond technical completeness. Termination guarantees that scope evaluation always completes; confluence guarantees that evalu- ation order does not affect the resulting semantic object; and canonical normal forms justify the use of unique semantic representatives throughout Chapters 8โ10. Without these results, the Representation Invariance Theorem would depend on evaluation strat- egy; with them, semantic equivalence becomes an intrinsic property of the Scope Calcu- lus itself. In summary, this appendix has established a formal abstract reduction system for the primitive alphabet, proved strong normalization and local and global confluence, established the existence and uniqueness of canonical normal forms, and reconstructed Scope directly from rewriting theory.
Appendix C
Cartesian Closed Categories and the Scope Calculus
The previous two appendices established complementary descriptions of the Scope Cal- culus: Appendix A constructed a translation from the simply typed ๐-calculus into Scope Geometry, and Appendix B showed that the primitive algebra forms a terminat- ing, confluent abstract reduction system. The objective of the present appendix is to place these constructions within categorical semantics, following Lambek and Scott, by asking under what additional hypotheses Scope can interpret the simply typed lambda calculus as a Cartesian closed category (CCC). Remark C.1. Every result in this appendix beyond ยงC.1 is conditional on Scope possess- ing finite products and exponential objects. Nothing in Chapters 6โ9 establishes that these exist for arbitrary bounded scope geometries; the appendix states the hypothesis explicitly wherever it is used, rather than presupposing it.
C.1 Cartesian Closed Structure Definition C.2 (Terminal object, product, exponential). A category ๐ has a terminal ob- ject 1 if for every object ๐ด there is a unique morphism ๐ด โ 1. Objects ๐ด, ๐ต have a binary product ๐ด ร ๐ต, equipped with projections ๐1 , ๐2 , if it satisfies the usual universal property; they have an exponential ๐ต๐ด if Hom(๐ ร ๐ด, ๐ต) โ Hom(๐, ๐ต๐ด ) naturally in ๐. A category possessing a terminal object, finite products, and exponentials is Cartesian closed.
Theorem C.3 (Conditional Cartesian closure). If Scope admits finite admissible products and exponential scope geometries in the sense above, then Scope is Cartesian closed.
Proof. Immediate from the preceding definition, given the stated hypothesis.
Within Scope Geometry, an exponential ๐ต๐ด (where it exists) is interpreted as the scope of admissible transformations from ๐ด to ๐ต, so that functions become first-class scope objects.
C.2 Interpretation of Types and Terms Under the hypothesis that Scope is Cartesian closed, the simply typed lambda calculus receives a standard categorical interpretation: [ ๐ด] = ๐ด โ Scope for a base type, [ ๐ด โ
121
๐ต] = ๐ต๐ด for function types, [ ๐ด ร ๐ต] = ๐ด ร ๐ต for product types, and [ 1] = 1 for the unit type, so that every typing judgement becomes a categorical statement. Lambda abstraction ๐๐ฅ.๐ corresponds to currying: under the translation of Ap- pendix A, ๐ฏ(๐๐ฅ.๐ ) = โฌ๐ (๐ฅ, ๐ฏ(๐ )), and categorically this is represented by the trans- pose ฮ(๐) โถ ๐ โ ๐ต๐ด , so binding becomes the categorical construction producing an exponential object. Application ๐ ๐ is interpreted through the evaluation morphism eval โถ ๐ต๐ด ร ๐ด โ ๐ต; within the Scope Calculus, evaluation corresponds to the admissi- ble sequence ๐ซโ๐๐ , with the evaluation scope entered, the bound environment activated, and the temporary computation collapsed after completion.
Theorem C.4 (Categorical beta correspondence). Under the Cartesian closure hypothesis, beta reduction (๐๐ฅ.๐ )๐ โ๐ฝ ๐ [๐ฅ โถ= ๐ ] commutes with the translation ๐ฏ: applying eval- uation to the curried translation of ๐๐ฅ.๐ and ๐ฏ(๐ ) and reducing to canonical form yields ๐ฏ(๐ [๐ฅ โถ= ๐ ]).
Proof. Application corresponds to evaluation and lambda abstraction to exponential transpose; the universal property of exponentials guarantees that evaluation following currying commutes. Since primitive reduction is confluent (Appendix B), normaliza- tion is unique, so the translated computation reduces uniquely to the translated substi- tuted term.
C.3 CurryโHowardโLambek Interpretation The categorical semantics, when it applies, induces a logical interpretation in which types correspond to propositions, morphisms to proofs, scope transformations to proof normalization, and primitive reductions to proof simplification: canonical scope reduc- tion is simultaneously program evaluation, proof normalization, categorical composi- tion, and scope simplification, placing the Scope Calculus within the broader Curryโ HowardโLambek correspondence. The Representation Invariance Theorem (Theorem 9.10) acquires a categorical read- ing under this interpretation: different computational substrates correspond to different concrete categories, and projection into Scope forgets implementation-specific structure while preserving admissible composition, so representation invariance becomes, in this reading, a statement about commuting categorical diagrams rather than about coordi- nate representations.
Conjecture C.5 (Universal property of Scope Geometry). Among categories generated by admissible computational transitions, Scope is initial with respect to primitive scope- preserving functors.
This is stated as a conjecture, not a theorem: nothing established in this monograph proves it, and it is included to indicate a direction for future work rather than a result relied upon elsewhere in the text. If established, every admissibility-preserving compu- tational system would factor uniquely through Scope.
C.4 Discussion The value of the Cartesian closed reading is not mathematical elegance for its own sake, but compatibility: under the stated hypothesis, the Scope Calculus connects to the stan- dard semantic foundations of functional programming, typed lambda calculus, proof
theory, and categorical logic, rather than replacing them. Scope Geometry extends these theories by making admissibility an explicit primitive mathematical object, uni- fying functional evaluation, scope management, proof normalization, and admissible computation within a common categorical language. This appendix has established the conditional Cartesian closed interpretation of Scope, the interpretation of lambda abstraction as binding and application as admissible evaluation, a strengthened cate- gorical proof of the beta correspondence, the CurryโHowardโLambek reading, and an explicitly labeled open conjecture concerning a universal property of Scope Geometry.
Appendix D
Higher Categories, Double Categories, and Scope Geometry
Appendix C showed that Scope admits a Cartesian closed interpretation under appro- priate hypotheses, with objects representing bounded scope geometries and morphisms representing admissible transformations. Many computational systems, however, ex- hibit two independent notions of composition simultaneously: a compiler performs both control-flow and data-flow composition, a proof assistant distinguishes proof re- finement from proof composition, and Scope Geometry itself naturally possesses both admissible computational transitions and refinement of scope description. This obser- vation motivates the introduction of higher categorical structure, developed here as a conditional extension rather than as an established part of the core framework.
D.1 Double Categories Definition D.1 (Double category). A double category ๐ป consists of objects, horizontal morphisms, vertical morphisms, and squares (2-cells), together with compatible hori- zontal and vertical composition satisfying the interchange law. Unlike an ordinary cate- gory, a double category allows two independent notions of composition to coexist. Definition D.2 (Scope double category, conditional). Suppose Scope admits a well- defined refinement relation on scope geometries, compatible with admissible transitions in the sense made precise below. Under this hypothesis, define ๐๐๐ ๐ก๐ by taking objects to be scope geometries, horizontal morphisms to be admissible computational transi- tions, vertical morphisms to be scope refinement operations, and squares to be admissi- ble commuting refinements. In this construction, the primitive scope operators become generators for the higher- dimensional cells: ๐ซ corresponds to entering a neighboring scope, โฌ๐ to constructing a new refinement layer, ๐๐ to evaluation of a completed square, and โ to a degenerate square whose refinement is forbidden. Scope computation, read this way, is no longer merely sequential but possesses internal geometric structure. Horizontal composition, for ๐ โถ S1 โ S2 and ๐ โถ S2 โ S3 , coincides with the ordinary composition of Chapter 8; vertical composition, for refinements ๐1 , ๐2 , corresponds to successive refinement of ad- missibility constraints and changes descriptive resolution rather than execution order. Theorem D.3 (Interchange law, conditional). For every compatible square in a well-defined ๐๐๐ ๐ก๐, (๐ผ โ๐ฃ ๐ฝ) โโ (๐พ โ๐ฃ ๐ฟ) = (๐ผ โโ ๐พ) โ๐ฃ (๐ฝ โโ ๐ฟ).
125
APPENDIX D. HIGHER CATEGORIES, DOUBLE CATEGORIES, AND SCOPE 126 GEOMETRY
Proof. Both compositions produce identical refinement histories, since horizontal com- putation preserves admissibility while vertical refinement preserves scope inclusion, so the resulting square commutes.
The interchange law guarantees that, under the stated hypotheses, refinement order and execution order remain compatible. A functor โ โถ Scope โ Scope that refines every scope by introducing additional internal structure while preserving identities, compo- sition, and admissibility models progressively more detailed semantic analysis without altering external computational behavior; repeated refinement then produces a hierar- chical tower S(0) โ S(1) โ S(2) โ โฏ of increasingly refined semantic descriptions, each preserving admissibility while increasing structural detail.
D.2 Bicategorical Relaxation Many computational transformations preserve admissibility only up to canonical equiv- alence, so strict equality may be unnecessarily restrictive.
Definition D.4 (Bicategorical Scope Geometry, conditional). A bicategorical Scope Ge- ometry replaces equalities ๐ โ ๐ = โ by coherent natural isomorphisms ๐ โ ๐ โ โ.
This relaxation accommodates optimization, parallel evaluation, lazy computation, and compiler transformations, all of which preserve semantics without preserving iden- tical execution histories.
Theorem D.5 (Coherence, conditional). If primitive reductions are confluent, every admissi- ble bicategorical diagram generated by primitive scope operations commutes after normalization.
Proof. Appendix B established unique normal forms, so every composite reduces uniquely, and hence every coherence diagram commutes after normalization.
D.3 Computational Interpretation Within practical programming systems, horizontal morphisms correspond to execution, vertical morphisms to abstraction refinement, squares to verified implementation, and higher cells to proofs that independently constructed implementations preserve admis- sibility. Higher Scope Geometry, read this way, provides a candidate mathematical lan- guage for verified compilers, proof assistants, interactive theorem provers, incremental program synthesis, and hierarchical autonomous agents โ though, as throughout this appendix, this reading is offered as a direction rather than as an established result.
Conjecture D.6 (Toward higher scope categories). The constructions above generalize to (โ, ๐) scope categories, whose higher morphisms would encode increasingly sophis- ticated notions of semantic equivalence, proof refinement, program transformation, and admissibility preservation, potentially connecting to homotopy type theory.
D.4 Discussion The progression from ordinary categories through Cartesian closed categories to double categories, bicategories, and conjecturally to higher categories illustrates a theme of this
monograph: Scope Geometry is not merely an operational formalism, but appears com- patible with progressively richer categorical structures while preserving its primitive operational interpretation, provided the relevant hypotheses (refinement relations, ad- missible products and exponentials, confluent bicategorical diagrams) are made explicit rather than assumed. This appendix has developed the double-categorical formulation of Scope Geometry conditionally on the existence of a compatible refinement relation, identified the primitive operators as generators of higher-dimensional cells, proved the interchange law and a coherence theorem under their respective stated hypotheses, and outlined hierarchical refinement towers and a conjectural extension toward higher cate- gory theory.
APPENDIX D. HIGHER CATEGORIES, DOUBLE CATEGORIES, AND SCOPE 128 GEOMETRY
Appendix E
Topological Semantics of Scope Geometry
The preceding appendices developed the Scope Calculus from the perspectives of rewrit- ing systems and category theory. We now introduce a topological formulation, showing that admissibility admits a natural interpretation in terms of continuity, closure, interi- ors, boundaries, and connected components. Where the earlier appendices emphasized algebraic computation, the present one studies the geometric organization of semantic spaces, assuming ordinary point-set topology as developed by Munkres [40].
E.1 Scope Spaces Definition E.1 (Scope topological space). A scope space is a pair (S, ๐ ) where S is the underlying set of semantic configurations and ๐ is a topology on it satisfying the usual axioms. The topology describes semantic neighborhoods rather than geometric distance: nearby points correspond to semantically similar admissible configurations. Definition E.2 (Interior, closure, boundary). The interior operator is Int(๐ด) = โ{๐ โ ๐ โถ ๐ โ ๐ด}; the closure operator is Cl(๐ด) = โ{๐น โถ ๐น closed, ๐ด โ ๐น }; and the boundary of ๐ด is ๐๐ด = Cl(๐ด) โ Int(๐ด). Interior points correspond to configurations whose admissibility remains stable un- der sufficiently small perturbations โ configurations lying safely within a semantic re- gion. Boundary points separate admissible from inadmissible semantic regions, so the boundary constraint of Chapter 6 admits a topological reading as an operator govern- ing movement across ๐๐ด. Closure represents completion of semantic neighborhoods: configurations lying in the closure may be approached arbitrarily closely through ad- missible computation.
E.2 Continuity and Admissible Maps Definition E.3 (Continuity). A scope transformation ๐ โถ (S, ๐1 ) โ (S, ๐2 ) is continuous if ๐ โ1 (๐ ) is open whenever ๐ is open. Continuity expresses preservation of semantic neighborhoods: small admissible per- turbations remain small after transformation. The principal topological object of Struc- tural Semantics is stronger than mere continuity.
129
Definition E.4 (Admissible map). A continuous map ๐ โถ S โ S is admissible if ๐(Int(๐ด)) โ Int(๐(๐ด)) and ๐(๐๐ด) โ ๐(๐(๐ด)).
Admissible maps therefore preserve both semantic interiors and semantic bound- aries. Two configurations ๐ฅ, ๐ฆ lie in the same admissible component if they belong to a common connected subset of S; connected components represent maximal regions through which continuous admissible computation may proceed, and moving between disconnected components necessarily requires violating some admissibility constraint. A semantic region ๐พ โ S is compact if every open cover possesses a finite subcover; compact regions admit useful properties for this theory, including that continuous ad- missible functions attain extrema on them and that optimization over admissible states becomes well behaved.
E.3 Fixed Points and Homotopy Topological fixed-point theory connects naturally with the repair operators introduced in the reconstruction chapters.
Theorem E.5 (Topological repair principle). Let ๐พ be a compact convex admissible region and ๐ โถ ๐พ โ ๐พ a continuous repair operator. Then ๐ possesses at least one fixed point.
Proof. Immediate from Brouwerโs Fixed Point Theorem.
This theorem provides a classical topological justification for the existence of re- paired admissible configurations. Two admissible maps ๐, ๐ โถ S โ S are homotopic if there is a continuous ๐ป โถ S ร [0, 1] โ S with ๐ป(๐ฅ, 0) = ๐(๐ฅ) and ๐ป(๐ฅ, 1) = ๐(๐ฅ), so that homotopy represents continuous deformation of admissible computations. Loops in semantic space measure global computational structure through the fundamental group ๐1 (S), the homotopy classes of admissible loops based at a chosen configuration; a non-trivial fundamental group indicates a semantic obstacle that cannot be removed through continuous admissible deformation. The algebraic-topological invariants ๐๐ , homology, cohomology, and Euler characteristic remain unchanged under admissible homotopy equivalence, and so provide candidate semantic invariants independent of any particular representation, suggesting a future connection with persistent homology and topological data analysis.
E.4 Discussion Topology supplies an additional geometric interpretation of the Scope Calculus: primi- tive operations become continuous movements between semantic regions, admissibility becomes boundary preservation, repair corresponds to continuous deformation toward fixed points, and representation invariance becomes topological equivalence rather than coordinate equality. This perspective complements, rather than replaces, the categorical interpretation of the preceding appendices. In summary, this appendix has introduced scope topological spaces, given interior, closure, and boundary readings of admissibil- ity, defined continuous and admissible maps, established connected components and compactness for admissible regions, proved a topological repair theorem via Brouwerโs theorem, and introduced the homotopy interpretation of semantic evolution together with the fundamental group as a candidate semantic invariant.
Appendix F
Sheaf Theory and Local-to-Global Reconstruction
The preceding appendix gave a topological interpretation of Scope Geometry, in which semantic configurations were points of a topological space and admissibility was bound- ary preservation. Topology alone, however, cannot fully describe semantic organiza- tion: many computational systems possess information that is locally consistent while globally inconsistent, and conversely global coherence frequently emerges only through reconciling multiple local observations. The mathematical language designed for this problem is sheaf theory, originally developed in algebraic topology and algebraic geom- etry and now applied throughout differential geometry, distributed systems, and logic. Within Structural Semantics, sheaf theory provides the mathematical language of recon- struction.
F.1 Presheaves and Sheaves
Let (S, ๐ ) be a scope topological space and let ๐ฐ = {๐๐ }๐โ๐ผ be an open cover, S = โ๐ ๐๐ , each ๐๐ representing a locally observable semantic region; no single region need contain sufficient information to reconstruct the entire computational state.
Definition F.1 (Scope presheaf). A scope presheaf โฑ assigns a set โฑ(๐ ) to every open ๐ โ S, together with restriction maps ๐๐๐ โถ โฑ(๐ ) โ โฑ(๐ ) for ๐ โ ๐, satisfying ๐๐๐ = id and ๐๐ ๐ โ ๐๐๐ = ๐๐๐ .
Intuitively โฑ(๐ ) contains every semantic description observable within ๐. An ele- ment ๐ โ โฑ(๐ ) is a local section, representing a partial semantic description extracted from a restricted computational context โ a local variable environment, an attention neighborhood, a proof fragment, or a segment of an activation trajectory. Restriction ๐๐๐ (๐ ) for ๐ โ ๐ models information loss under localization.
Definition F.2 (Scope sheaf). A presheaf โฑ is a sheaf if it satisfies local identity (if ๐ , ๐ก โ โฑ(๐ ) agree on every member of an open cover of ๐, then ๐ = ๐ก) and gluing (if sections ๐ ๐ โ โฑ(๐๐ ) agree on overlaps, ๐๐๐ ,๐๐ โฉ๐๐ (๐ ๐ ) = ๐๐๐ ,๐๐ โฉ๐๐ (๐ ๐ ), then there is a unique ๐ โ โฑ(๐ ) restricting to every ๐ ๐ ).
131
F.2 Reconstruction as Gluing The gluing axiom has a direct computational reading: compatible local semantic de- scriptions reconstruct a unique global semantic configuration, a principle underlying distributed reasoning, proof composition, compiler passes, modular verification, and hierarchical language understanding.
Definition F.3 (Reconstruction operator). Let ๐ฐ = {๐๐ } be an admissible cover. Define โ โถ โ๐ โฑ(๐๐ ) โ โฑ(S) by โ(๐ 1 , โฆ , ๐ ๐ ) = Glue(๐ 1 , โฆ , ๐ ๐ ) whenever the sections agree on overlaps.
Reconstruction, in this reading, is precisely the sheaf gluing operator. Not every family of local observations admits gluing: the reconstruction defect is 0 whenever gluing succeeds uniquely and positive otherwise, giving a sheaf-theoretic reading of the recon- struction defect discussed in Chapter 9 as a measure of failure of global compatibility rather than of numerical error alone. The stalk at a configuration ๐ฅ, โฑ๐ฅ = lim โฑ(๐ ), โโ๐ฅโ๐ collects every local semantic observation arbitrarily close to ๐ฅ; a scope-preserving trans- formation ๐ โถ S โ S induces a sheaf morphism that preserves restriction maps, com- mutes with gluing, and therefore preserves admissible reconstruction.
F.3 Cohomological Obstructions When reconstruction fails, the obstruction can itself be measured. Given an admissible open cover, the associated ฤech complex computes compatibility between overlapping local descriptions, and the resulting cohomology groups ๐ปฬ ๐ (S, โฑ) measure increasingly global reconstruction obstructions; in particular a nontrivial ๐ปฬ 1 indicates that locally compatible semantic fragments cannot be assembled into a globally admissible config- uration.
Conjecture F.4 (Cohomological admissibility). A computational system is globally ad- missible whenever the associated reconstruction cohomology vanishes: ๐ปฬ 1 (S, โฑ) = 0.
This is stated as a conjecture rather than a theorem; it generalizes the reconstruc- tion results of Chapter 9 and suggests a connection between admissibility and algebraic topology that is not established here.
F.4 Discussion Sheaf theory provides one of the strongest mathematical interpretations of Structural Se- mantics developed in this monograph: Scope Geometry describes admissible regions, topology describes continuity, category theory describes composition, and sheaf the- ory explains how local admissibility reconstructs global semantic organization. Read this way, reconstruction is not a heuristic procedure but an instance of one of the cen- tral local-to-global principles of modern mathematics. This appendix has introduced scope presheaves and sheaves, local sections and restriction morphisms, sheaf gluing as a reading of semantic reconstruction, the reconstruction defect as gluing failure, stalk se- mantics, sheaf morphisms, and a ฤech-cohomological reading of reconstruction failure together with the associated open conjecture.
Appendix G
Measure Theory, Entropy, and Admissibility
The preceding appendices developed the Scope Calculus through rewriting theory, cat- egory theory, topology, and sheaf theory, each providing an increasingly abstract struc- tural language for admissible computation. The present appendix instead introduces quantitative measures on semantic spaces, drawing on measure theory, probability, and information theory, so that admissibility becomes not merely a binary property but a measurable quantity. Throughout, (S, ฮฃ, ๐) denotes a measurable scope space.
G.1 Measures and Admissibility Definition G.1 (Scope ๐-algebra and semantic measure). A scope ๐-algebra ฮฃ is a collec- tion of subsets of S containing โ and closed under complements and countable unions; its elements are measurable semantic regions. A semantic measure ๐ โถ ฮฃ โ [0, โ] satisfies ๐(โ ) = 0 and countable additivity over disjoint measurable sets. Unlike a probability measure, a semantic measure need not satisfy ๐(S) = 1; normalizing ๐ by ๐(S) recovers a probability measure as one special instance of semantic measurement.
Definition G.2 (Admissibility measure). Let ๐ด โ S with ๐(๐ด) > 0. Its admissibility is ๐ผ(๐ด) = ๐(๐ด โฉ Adm)/๐(๐ด), so that ๐ผ = 1 indicates complete admissibility and ๐ผ = 0 indicates complete inadmissibility.
For a stochastic process ๐ โถ ฮฉ โ S, the expected admissibility ๐ธ[๐ผ] = โซ ๐ผ(๐) ๐๐ measures the average structural integrity of the process.
G.2 Entropy and Mutual Information For a probability distribution ๐ = (๐๐ ), the Shannon entropy is ๐ป(๐ ) = โ โ๐ ๐๐ log ๐๐ ; within Structural Semantics, entropy measures uncertainty over admissible semantic configurations rather than merely over tokens. For a semantic state ๐ and an observa- tion ๐, the conditional entropy ๐ป(๐ โฃ ๐ ) measures remaining uncertainty after obser- vation, and the mutual information ๐ผ(๐; ๐ ) = ๐ป(๐) โ ๐ป(๐ โฃ ๐ ) measures preserved structure: large mutual information indicates that observation preserves semantic or- ganization, small mutual information suggests context loss. For an admissible distribu- tion ๐ and an observed distribution ๐, the relative entropy ๐ทKL (๐ โ๐) = โ๐ ๐๐ log(๐๐ /๐๐ ) measures deviation from admissible behavior.
133
Definition G.3 (Reconstruction entropy). Let ๐ be the reconstruction operator of Ap- pendix F. Define the reconstruction entropy ๐ป๐ = ๐ป(๐ (๐)).
Theorem G.4 (Entropy monotonicity). If ๐ is an admissibility-preserving reconstruction operator, then ๐ป(๐ (๐)) โค ๐ป(๐).
Proof. Under admissibility, reconstruction identifies compatible semantic regions with- out creating new ambiguity, so the image partition refines rather than enlarges the un- derlying uncertainty, and entropy cannot increase.
Observation updates admissibility in the usual Bayesian manner: for new evidence ๐ธ, ๐ (๐ด โฃ ๐ธ) = ๐ (๐ธ โฃ ๐ด)๐ (๐ด)/๐ (๐ธ), so the posterior updates admissibility estimates while preserving measure-theoretic consistency.
G.3 Measure-Preserving Dynamics A scope transformation ๐ โถ S โ S is measure preserving if ๐(๐ โ1 (๐ด)) = ๐(๐ด) for every measurable ๐ด, so that measure-preserving admissible transformations conserve seman- tic volume. Such a transformation is ergodic if every invariant measurable subset has measure 0 or 1, connecting long-running computation with statistical semantic equilib- rium.
G.4 Discussion This appendix extends Structural Semantics from qualitative to quantitative analysis: topology supplied continuity, category theory supplied composition, sheaf theory sup- plied reconstruction, and measure theory now supplies uncertainty, probability, entropy, and statistical inference, so that admissibility can be studied simultaneously from geo- metric, algebraic, topological, and statistical perspectives. This appendix has introduced measurable scope spaces and semantic measures, defined the admissibility measure and expected admissibility, related entropy, conditional entropy, and mutual information to semantic uncertainty and its preservation, proved an entropy-monotonicity theorem for admissibility-preserving reconstruction, and introduced measure-preserving and er- godic scope dynamics.
Appendix H
Notation and Mathematical Conventions
This appendix collects, in one place, every symbol, operator, and convention used through- out the monograph, as promised in the front-matter Notation and Conventions section. Symbols are grouped by the cluster of chapters in which they are introduced; within each group they appear in the order first used.
H.1 Scope Calculus (Chapters 1, 6โ10) The category of scope geometries is Scope; a scope object is written S = โจโฌ, โฐ, ๐ฎchild โฉ, with โฌ the boundary (a collection of admissibility predicates), โฐ the admissible interior, and ๐ฎchild the collection of child scopes. A scope morphism is ๐ โถ S๐ โ S๐ . The primitive alphabet is ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } = {๐ซ, โ, โฌ๐ , ๐๐ }, where ๐ซ promotes an entity from a child scope to its parent, โ rejects an inadmissible transition, โฌ๐ creates an admissible dependency within a scope, and ๐๐ resolves a scope into a canonical result. A computa- tional system is ๐ = (๐, ๐ฏ), with ๐ a state domain and ๐ฏ the set of admissible transition operators ๐ โถ ๐ โ ๐. The category of Whiteheadian processes, used only in Chapters 4, 5, and 10, is Process; the translation functor between the two is ๐น โถ Process โ Scope.
H.2 Whitehead Reconstruction (Chapters 4โ5, 10) An actual occasion is ๐ธ; antecedent data is ๐ท; a prehension is ๐ = โจ๐ท, ๐ฃ, ๐ดโฉ, with ๐ฃ โ {+, โ} the valuation polarity and ๐ด the Subjective Form; the satisfaction state is ๐. These symbols are kept visually distinct from the Scope Calculus notation throughout, since no identification between the two systems is assumed before the translation functor of Chapter 10 makes any correspondence explicit.
H.3 Transformer Analysis (Chapters 11โ15) The residual stream tensor is X โ โ(๐ฟ+1)ร๐ ร๐model , with ๐ฟ the number of layers, ๐ the sequence length, and ๐model the embedding dimension; x๐,๐ก is the residual vector for to- ken ๐ก after layer ๐. An activation trajectory is ๐พ๐ก = (x0,๐ก , โฆ , x๐ฟ,๐ก ); the layerwise Jacobian is ๐ฝ๐ = ๐x๐+1 /๐x๐ ; discrete curvature is ๐ ๐ ; the local layerwise Lyapunov exponent is ๐๐ = log ๐max (๐ฝ๐ ). The analytic activation signal is z๐ = x๐ + ๐ โ(x๐ ), with โ the discrete
135
Hilbert transform; instantaneous amplitude and phase are ๐ด๐ and ๐๐ ; instantaneous fre- quency is ๐๐ = ๐๐+1 โ ๐๐ ; the Phase Locking Value is PLV โ [0, 1]; the Marine Operator is โณ โถ ฮ โ โ๐ on the space ฮ of activation trajectories. The Doppler velocity is ๐ฃ๐ and the Marine salience gate is ๐บ๐ . The diagnostic state vector is d๐ก = (๐ด, ๐, ๐, PLV, ๐ , ๐); the admissibility estimator is ๐ โถ d โ [0, 1]; the context collapse threshold is ๐ โ (0, 1).
H.4 General Conventions Bold uppercase symbols denote categories or structured spaces; calligraphic symbols denote operators, constraint systems, or families of sets; lowercase Greek letters denote morphisms, transformations, or parameters. Once a symbol is assigned a formal mean- ing in a given cluster of chapters, it is preserved throughout the monograph and is not reassigned. Definitions, propositions, lemmas, theorems, and corollaries share a single numbering sequence within each chapter, restarting at each chapter; equations are num- bered within chapters. Every theoremโs hypotheses are stated explicitly in its statement rather than left to surrounding prose.
Appendix I
Collected Proofs
This appendix collects the proofs deferred from the main text for length, principally the structural-induction argument underlying Operational Completeness (Chapter 7) and the closure and decomposition arguments underlying the foundational theorems of Chapter 9. Each proof below restates its theorem for reference.
I.1 Operational Completeness (Theorem 7.6) Theorem. Every admissible transformation of bounded scope geometries admits a finite presentation as a composition of primitive operations from ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ }.
Proof. We argue by induction on the number of atomic modifications comprising an admissible transformation, where an atomic modification changes exactly one of โฌ, โฐ, or ๐ฎchild at a single scope. Base case. A transformation comprising zero atomic modifications is the identity, presented by the empty primitive sequence. Inductive step. Suppose every admissible transformation comprising ๐ atomic modifi- cations admits a finite primitive presentation. Let ๐ comprise ๐+1 atomic modifications, and let ๐ โฒ be ๐ with its final atomic modification removed, so ๐ โฒ comprises ๐ modifica- tions and, by the inductive hypothesis, has a finite primitive presentation ๐คโฒ . The final modification either introduces new admissible information into โฐ (representable by โฌ๐ , if it creates a dependency, or by an update already captured within a preceding โฌ๐ or ๐ซ otherwise), rejects an inadmissible candidate transition (representable by โ), trans- fers information from a child scope into its parent (representable by ๐ซ), or resolves the scope into a canonical result (representable by ๐๐ ); one of these four cases must apply because Chapter 6 characterized every admissibility-relevant change to a scope geom- etry as falling into exactly one of them. Appending the corresponding primitive to ๐คโฒ yields a finite primitive presentation of ๐. By induction, every admissible transforma- tion with finitely many atomic modifications admits a finite primitive presentation, and every admissible transformation has finitely many atomic modifications by the finite- ness assumption of Chapter 9.
I.2 Local Confluence Verification (Chapter 7, ยง7 and Appendix B) The elementary rewrite rules (R1)โ(R4) of Chapter 7 give rise to a finite set of critical pairs, since each ruleโs left-hand side has bounded depth and there are only four rules. Enumerating the overlaps: (R1) (identity elimination) can overlap with any other rule
137
only at the empty word, where all rules agree trivially since ๐ has no further structure to rewrite. (R2) and (R3) (consecutive refusal and consecutive collapse) overlap only with themselves, and each such overlap (โโโ, ๐๐ ๐๐ ๐๐ ) reduces to the single-operator form by two applications of the same rule regardless of order, so both branches join immediately. (R4) (empty promotion) overlaps with (R1) only at ๐ซ๐ itself, where both rules yield ๐ directly. No other overlaps exist among (R1)โ(R4), since each ruleโs left- hand side pattern is disjoint from the others outside the cases just enumerated. Hence every critical pair joins within at most two steps, establishing local confluence as asserted in Chapter 7 and used in Appendix B.
I.3 Closure, Decomposition, and Reconstruction (Chapter 9) The Closure Theorem (Theorem 9.1) follows by induction on the primitive presentation of ๐ furnished by Proposition 8.4: each of ๐ซ, โ, โฌ๐ , and ๐๐ was defined in Chapter 7 to act only on admissible states and to produce a well-formed bounded scope geometry satisfying Definition 6.5 as output (boundary inheritance is preserved by construction, since none of the four primitives removes an inherited constraint), so a finite composi- tion of them, applied to a well-formed input, returns a well-formed output. The Scope Decomposition Theorem (Theorem 9.2) is Proposition 8.4 restated at the level of Scope; the induction is the one given in ยงI7.6 above, transported along the iden- tification of morphisms with admissible transformations. The State Reconstruction Theorem depends on collapse operations recording their boundary history; under that recording assumption, the parent scope S๐ at any step is determined as the unique bounded scope geometry whose interior and boundary agree, on every child region present at that step, with the recorded pre-collapse data โ uniqueness follows because two candidate parents agreeing with all recorded child data on their boundaries would, by the sheaf-theoretic local-identity property of Appendix F applied to the cover by child scopes, coincide.
Bibliography
[1] Shun-ichi Amari and Hiroshi Nagaoka. Methods of Information Geometry. American Mathematical Society, 2016.
[2] Anthropic. Towards monosemanticity: Decomposing language models with dic- tionary learning. Transformer Circuits Thread, 2023.
[3] W. Ross Ashby. An Introduction to Cybernetics. Chapman & Hall, 1956.
[4] Steve Awodey. Category Theory. Oxford University Press, 2010.
[5] Franz Baader and Tobias Nipkow. Term Rewriting and All That. Cambridge Univer- sity Press, 1998.
[6] Henk P. Barendregt. The Lambda Calculus: Its Syntax and Semantics. North-Holland, 1984.
[7] Ned Block. On a confusion about a function of consciousness. Behavioral and Brain Sciences, 18(2):227โ247, 1995.
[8] Tom B. Brown et al. Language models are few-shot learners. arXiv:2005.14165, 2020.
[9] Stanley Burris and H. P. Sankappanavar. A Course in Universal Algebra. Springer, 1981.
[10] David J. Chalmers. The Conscious Mind: In Search of a Fundamental Theory. Oxford University Press, 1996.
[11] Andy Clark. Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science, volume 36. 2013.
[12] Eugenio Coseriu. Principios de Semรกntica Estructural. Gredos, 1977.
[13] Ferdinand de Saussure. Cours de Linguistique Gรฉnรฉrale. Payot, 1916.
[14] Samuel Eilenberg and Saunders Mac Lane. General theory of natural equivalences. Transactions of the American Mathematical Society, 58:231โ294, 1945.
[15] Nelson Elhage et al. A mathematical framework for transformer circuits. An- thropic, 2021.
[16] Jerry A. Fodor. The language of thought. 1975.
[17] Gottlob Frege. รber sinn und bedeutung. Zeitschrift fรผr Philosophie und philosophis- che Kritik, 100:25โ50, 1892.
139
[18] Karl Friston. The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11:127โ138, 2010.
[19] Karl Friston. A free energy principle for a particular physics. arXiv:1906.10184, 2019.
[20] Gerhard Gentzen. Untersuchungen รผber das logische schlieรen. Mathematische Zeitschrift, 1935.
[21] Mor Geva et al. Transformer feed-forward layers build predictions by promoting concepts in the vocabulary space. arXiv:2203.14680, 2022.
[22] Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016.
[23] Algirdas Julien Greimas. Sรฉmantique Structurale. Larousse, 1966.
[24] Martin Heidegger. Being and Time. Niemeyer, 1927.
[25] Louis Hjelmslev. Omkring Sprogteoriens Grundlรฆggelse. Munksgaard, 1943.
[26] Louis Hjelmslev. Prolegomena to a Theory of Language. University of Wisconsin Press, 1961.
[27] Jordan Hoffmann et al. Training compute-optimal large language models. arXiv:2203.15556, 2022.
[28] Jakob Hohwy. The Predictive Mind. Oxford University Press, 2013.
[29] John E. Hopcroft, Rajeev Motwani, and Jeffrey D. Ullman. Introduction to Automata Theory, Languages, and Computation. Addison-Wesley, 2006.
[30] Edmund Husserl. Ideas: General Introduction to Pure Phenomenology. Niemeyer, 1913.
[31] Jenann Ismael. Curieโs principle. Synthese, 110(2):167โ190, 1997.
[32] Jenann Ismael and Bas C. van Fraassen. Symmetry as a guide to superfluous theo- retical structure. In Katherine Brading and Elena Castellani, editors, Symmetries in Physics: Philosophical Reflections, pages 371โ392. Cambridge University Press, 2003.
[33] Jared Kaplan et al. Scaling laws for neural language models. arXiv:2001.08361, 2020.
[34] Cheng Luo, Zefan Cai, and Junjie Hu. Multi-head attention residuals. arXiv:2607.27230, 2026.
[35] John Lyons. Structural Semantics. Blackwell, 1963.
[36] John Lyons. Semantics. Cambridge University Press, 1977.
[37] Saunders Mac Lane. Categories for the Working Mathematician. Springer, 1998.
[38] Samuel Marks et al. Sparse feature circuits. arXiv, 2024.
[39] Maurice Merleau-Ponty. Phenomenology of Perception. Gallimard, 1945.
[40] James R. Munkres. Topology. Prentice Hall, 2000.
[41] Thomas Nagel. What is it like to be a bat? The Philosophical Review, 83(4):435โ450, 1974.
[42] Neel Nanda et al. Progress measures for grokking via mechanistic interpretability. arXiv:2301.05217, 2023.
[43] Allen Newell and Herbert A. Simon. Computer Science as Empirical Inquiry: Symbols and Search, volume 19. 1976.
[44] Chris Olah, Nick Cammarata, Ludwig Schubert, Gabriel Goh, Michael Petrov, and Shan Carter. Zoom in: An introduction to circuits. Distill, 2020.
[45] OpenAI. Gpt-4 technical report. arXiv:2303.08774, 2023.
[46] Charles Sanders Peirce. Collected Papers of Charles Sanders Peirce. Harvard University Press, 1931.
[47] Benjamin C. Pierce. Types and Programming Languages. MIT Press, 2002.
[48] Hilary Putnam. Psychological predicates. Art, Mind, and Religion, pages 37โ48, 1967.
[49] Dana Scott. Outline of a mathematical theory of computation. Technical Monograph PRG-2, Oxford University Computing Laboratory, 1970.
[50] John R. Searle. Minds, brains, and programs. Behavioral and Brain Sciences, 3(3): 417โ457, 1980.
[51] Thomas L. Short. Peirceโs Theory of Signs. Cambridge University Press, 2007.
[52] Adly Templeton et al. Scaling monosemanticity: Extracting interpretable features from claude 3 sonnet. Transformer Circuits Thread, 2024.
[53] Jost Trier. Der deutsche wortschatz im sinnbezirk des verstandes. Universitรคtsverlag Winter, 1931.
[54] Alan M. Turing. Computing machinery and intelligence. Mind, 59(236):433โ460, 1950.
[55] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in Neural Information Processing Systems, 2017.
[56] Alfred North Whitehead. Process and Reality. Macmillan, 1929.
[57] Norbert Wiener. Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press, 1948.
[58] Glynn Winskel. The Formal Semantics of Programming Languages. MIT Press, 1993.
Index of Symbols
๐ด (Subjective Form), 135 ๐๐ (instantaneous frequency), 136 ๐ (admissibility estimator), 136 ๐ด๐ (instantaneous amplitude), 136 ๐ (prehension), 135 ๐๐ (instantaneous phase), 136 โฌ (boundary), 135 PLV (Phase Locking Value), 136 โฌ๐ (Bind operator), 135 ๐ซ (Pop operator), 135 Process (category of processes), 135 ๐๐ (Collapse operator), 135
๐ (state domain), 135 โ (Refuse operator), 135 d๐ก (diagnostic state vector), 136 ๐ท (antecedent data), 135 S (scope object), 135 ๐model (embedding dimension), 135 ๐ (satisfaction state), 135 ๐ฎchild (child scopes), 135 โฐ (admissible interior), 135 Scope (category of scope geometries), ๐ธ (actual occasion), 135 135 ฮฃ = {๐ซ, โ, โฌ๐ , ๐๐ } (primitive ๐น (translation functor), 135 alphabet), 135 ฮ (space of activation trajectories), 136 ๐ (scope morphism), 135 ๐พ๐ก (activation trajectory), 135 ๐บ๐ (Marine salience gate), 136 ๐ฏ (transition set), 135 ๐ (sequence length), 135 โ (discrete Hilbert transform), 135 ๐ (context collapse threshold), 136
๐ฝ๐ (layerwise Jacobian), 135 ๐ฃ (valuation polarity), 135 ๐ฃ๐ (Doppler velocity), 136 ๐ ๐ (discrete curvature), 135
๐ฟ (number of layers), 135 X (residual stream tensor), 135 ๐๐ (local Lyapunov exponent), 135 x๐,๐ก (residual vector), 135
โณ (Marine Operator), 136 z๐ (analytic activation signal), 135
143