Taxonomy (information architecture)

Definition

A taxonomy in information architecture is a hierarchical classification scheme that organises content into categories and subcategories, providing a structured framework for navigation, discovery, and filtering. Unlike flat tags or free-text labels, a taxonomy imposes a controlled, consistent vocabulary with explicit parent-child relationships. In a legal AI system, the taxonomy defines how legal content is categorised — by tax type (income tax, VAT, registration duties), by jurisdiction (federal, Flemish, Walloon), by document type (legislation, case law, administrative guidance), and by topic area (deductible expenses, international taxation, procedural obligations).

Why it matters

• Consistent categorisation — a taxonomy ensures that the same topic is always labelled the same way, preventing inconsistencies like “corporate tax”, “vennootschapsbelasting”, and “impôt des sociétés” being treated as separate categories
• Navigational structure — taxonomies provide the backbone for browse-based navigation: users can drill down from “Tax types” → “Income tax” → “Corporate income tax” → “Deductible expenses” to find relevant content
• Faceted filtering — taxonomies enable structured filters in search results: filter by tax type, jurisdiction, document type, or time period to narrow results without changing the search query
• Knowledge gap identification — a well-maintained taxonomy reveals where content is sparse (categories with few documents) or where new categories are needed (emerging topics not yet classified)

How it works

A taxonomy consists of three elements:

Terms — the controlled vocabulary of category names. Each term has a preferred label (the canonical name), alternative labels (synonyms in different languages or common variations), and a definition. Terms are language-aware: the same concept has Dutch, French, and German labels.

Hierarchy — the parent-child relationships between terms. “Income tax” is a child of “Tax types”. “Corporate income tax” is a child of “Income tax”. The hierarchy can have multiple levels of depth, though practical taxonomies rarely exceed 4-5 levels.

Relationships — beyond hierarchy, taxonomies may include associative relationships (“related to”), equivalence relationships (“same as”), and scope notes (explaining what a term includes and excludes). These help users navigate between related but non-hierarchical topics.

Taxonomies are maintained by domain experts who add new terms as the legal landscape evolves (e.g., adding a category for new tax types introduced by legislation), merge or split categories as needed, and ensure cross-language consistency.

In a legal AI system, the taxonomy serves dual purposes: it organises the user-facing navigation and it provides metadata categories for document tagging during ingestion. Automated classification assigns taxonomy categories to new documents using text classifiers trained on the taxonomy’s structure.

Common questions

Q: How is a taxonomy different from an ontology?

A: A taxonomy is a hierarchical classification — it organises concepts into parent-child relationships. An ontology is richer — it defines types of entities, their properties, and the relationships between them, enabling logical reasoning. A taxonomy says “corporate income tax is a type of income tax”. An ontology additionally defines that corporate income tax has a rate, applies to specific entities, and interacts with specific deductions.

Q: How many categories should a taxonomy have?

A: Enough to be useful for navigation and filtering, but not so many that the categories become fragmented or overlapping. For a Belgian tax AI system, 50-200 leaf categories (grouped under 10-20 top-level categories) typically provides sufficient granularity without overwhelming users.