My work began in linguistics, in disciplines concerned with how meaning is constructed, negotiated, and patterned across real-life language use. Over years of analysing conversations, metaphors, and discourse, I developed structured interpretive frameworks to make complex language visible and systematically describable.
What I did not fully anticipate at the time was how directly that training would translate into structured data work, annotation frameworks, and the design of interpretable language systems.
Core principle
Moving from language to systems is not a career shift, it is a shift in operational context. The method remains disciplined modelling of meaning.
Research as structured modelling
Academic linguistics is already a form of modelling. Conversation analysis requires transcription conventions and sequential coding. Corpus linguistics depends on controlled tag sets and reproducible search strategies. Metaphor research involves clustering expressions and defining conceptual mappings that hold across datasets.
In each case, the task is structural: define categories, test boundaries, refine edge cases, and document decisions for replication.
From interpretation to annotation
Professional annotation introduces scale, speed, and strict guideline adherence. Unlike academic research, schemas are predefined and must be applied consistently across large datasets.
High-quality annotation requires:
- Mutually exclusive, clearly defined categories
- Explicit handling of ambiguity
- Awareness of downstream propagation effects
- Transparent decision logic supporting agreement
Annotation is structured linguistic judgement under constraint. The interpretation remains, but it is operationalised.
Language and code
Programming is another formal language environment. It depends on syntax, compositional structure, and explicit rule definition. Ambiguity, if left unresolved, produces systemic error.
- Sentences and functions rely on compositional logic
- Discourse coherence mirrors architectural coherence
- Context determines interpretation in both natural and formal systems
- Structural decisions shape downstream behaviour
When building rule-based pipelines or structured JSON outputs, the objective is not only functional output but traceable internal logic.
From data to tool
A concrete example of this trajectory is Explain My Pain, a deterministic prototype derived from corpus-based metaphor research.
The system translates structured taxonomies into rule-based matching logic, generating bilingual outputs for patients and clinicians.
input → taxonomy match → validated schema → report renderingThe architecture preserves separation between input, processing, and output. The logic remains inspectable and auditable.
Convergence
Research trained pattern recognition. Annotation trained operational discipline. Programming enables formal encoding and deployment.
The through-line is continuity of method: modelling meaning carefully, documenting decisions rigorously, and building interpretable systems.
In an era dominated by opaque statistical systems, there remains a place for deterministic language architectures that make reasoning visible.