From Annotation to Application

Implementing a structured metaphor modelling pipeline from schema design to user-facing system.

This case study demonstrates how a linguistically grounded semantic modelling framework can be implemented as a layered system. It illustrates the progression from schema-driven annotation to rule-based tagging and, finally, to human-facing application.

Key idea

Linguistic categories can be formalised as structured infrastructure: schema → taxonomy → executable rules → interpretable output.

Step 1 · Schema-Driven Annotation

Systematic modelling begins with explicit category design. Rather than applying generic labels, annotation schemas encode theoretical distinctions drawn from metaphor research, stance analysis, and intercultural communication.

Clear separation between metaphorical and literal usage
Encoding of stance and evaluative positioning
Documentation of translation shifts
Versioned schemas preserving modelling assumptions

Step 2 · Rule-Based Metaphor Tagging

Annotated datasets inform the development of structured taxonomies. Conceptual domains (e.g. FIRE → intensity, WEIGHT → burden, SHARP OBJECTS → intrusion, CONFINEMENT → loss of control) are formalised into hierarchical systems.

Detection rules distinguish figurative from literal usage in context. Outputs are evaluated against human annotation using precision, recall, F1, and agreement metrics.

Gold-standard annotations
        ↓
Rule-based taxonomy
        ↓
Tagger implementation
        ↓
Evaluation (Precision / Recall / F1 / κ)

Step 3 · Human-Facing Application

The final stage demonstrates how modelling infrastructure can produce interpretable outputs. Pain Tagger and Explain My Pain serve as domain-specific implementations of this architecture.

Descriptors mapped to structured conceptual domains
Sensory and emotional entailments formalised
Readable summaries generated from rule-based logic

This stage illustrates how research-grade semantic modelling can be translated into accessible systems without relying on opaque machine learning.

Architectural Principle

Human interpretation
        ↓
Formal schema
        ↓
Executable rules
        ↓
Structured output

The value of this pipeline lies not in a single tool, but in the ability to design and implement structured semantic systems across domains — from health communication to political discourse or intercultural analysis.