A crosswalk exists because two systems can describe the same world without dividing it in the same way.
One library uses one subject vocabulary. Another uses a different scheme. A government changes industry codes. A school syllabus changes topic structure. A medical terminology is revised. A company replaces one product taxonomy with another. An archive inherits records created under several cataloguing standards.
The labels may look similar while the meanings differ. The labels may look different while the meanings overlap. One category may split into three. Three may merge into one. Some concepts may have no useful equivalent at all.
A classification crosswalk records those relationships explicitly.
Quick answer: what is a classification crosswalk?
A classification crosswalk is a structured mapping between concepts or categories in two classification systems, vocabularies, taxonomies or ontology versions.
- Exact: concepts are equivalent for the intended use.
- Broader: source concept is narrower than the target.
- Narrower: source concept is broader than the target.
- Partial: meanings overlap but are not equivalent.
- Split: one old category maps to several new categories.
- Merge: several old categories map to one new category.
- Related: concepts are connected but should not be substituted.
- Unmappable: no defensible target exists.
Crosswalking is the continuity layer between the classification methods described in How to Categorise Anything, How Controlled Vocabularies Work and How Ontologies Work.
1. Why crosswalks are necessary
Knowledge systems evolve independently. They are built by different institutions, for different users, in different periods and for different operational jobs.
Expecting them to use identical categories is unrealistic.
Crosswalks allow them to communicate without pretending they were designed from one master schema.
2. Do not map labels by spelling alone
Two systems may use the same word differently.
“Urban”, “primary”, “advanced”, “archive”, “company”, “student” and “emergency” can all have system-specific definitions.
A crosswalk should compare concepts, not strings.
3. Different labels can represent the same concept
One system may say “Additional Mathematics”. Another says “A-Math”. One says “myocardial infarction”. Another stores “heart attack”. One uses a current place name while an older archive uses a historical name.
Where the underlying concept is genuinely equivalent, a crosswalk can record an exact or close equivalence.
4. Exact mappings are rarer than they look
An exact mapping means substitution is safe for the intended task.
That requires more than similar names. Definitions, scope, granularity, time and authority should align closely enough that records can move without changing meaning.
Use exact mappings conservatively.
5. Broader mappings preserve direction
Suppose System A has “electric passenger car” and System B only has “car”.
The A concept maps to a broader B concept.
Mapping upward preserves that every electric passenger car is a car, but the reverse is not true.
6. Narrower mappings require caution
If the source concept is “car” and the target concept is “electric passenger car”, assigning every source record to the narrower target would invent information.
A narrow target requires evidence the source data may not contain.
7. Partial mappings admit overlap
Two categories may overlap substantially without one containing the other.
For example, one curriculum may organise a topic around “forces” while another organises related material under “motion and interactions”. The content overlaps, but the conceptual boundaries may not be identical.
Marking partial equivalence is more honest than forcing an exact mapping.
8. One-to-many mappings signal a split
An older category may later be divided into several more precise concepts.
Migration then depends on whether old records contain enough evidence to choose the correct new branch.
If not, preserve an ambiguous split rather than guessing.
9. Many-to-one mappings signal a merge
Several older categories may become one broader category in a new scheme.
New records can use the merged concept while historical records retain their original distinctions in provenance.
10. Some concepts should remain unmapped
Not every concept has an equivalent.
A local legal category, cultural concept, historical institution or specialist scientific term may simply have no defensible target in another system.
“Unmappable” is a valid result. False equivalence is worse than a gap.
11. Crosswalks need identifiers on both sides
Labels change. Mapping concept IDs rather than display strings makes the crosswalk more durable.
A robust mapping record should identify:
- source system;
- source concept ID;
- target system;
- target concept ID;
- mapping type;
- confidence;
- evidence;
- version;
- review status.
12. Crosswalks are directional
A mapping from A to B may not behave the same as B to A.
If A is narrower than B, translating A → B loses specificity. Translating B → A may be impossible without more evidence.
Direction must be explicit.
13. Crosswalks can be lossy
When a detailed taxonomy maps into a coarse one, information is lost.
The crosswalk should record that loss instead of hiding it.
This matters for analytics. A report produced after migration may look comparable while actually operating at lower resolution.
14. Crosswalks can also be underdetermined
If a broad old category maps to several narrow new categories, the target cannot be determined from the old label alone.
Additional metadata, text review or human judgement may be required.
15. Time changes mappings
Concepts and institutions evolve.
A mapping valid in 2020 may be wrong in 2026 because one system changed definitions or introduced new categories.
Crosswalk records need effective dates and version references.
16. Version-to-version crosswalks preserve history
A system should know how Version 1 maps to Version 2, and Version 2 to Version 3.
This allows old records to remain interpretable and supports migration without deleting historical meaning.
17. Do not overwrite old classifications
If a record was classified under an earlier scheme, preserve that original assignment.
A migrated target classification can be stored alongside it with a provenance trail.
History should be translated, not erased.
18. Crosswalking controlled vocabularies
Controlled vocabularies often map relatively well because both sides already distinguish concepts from labels.
Compare preferred terms, definitions, scope notes, broader terms and identifiers before assigning equivalence.
19. Crosswalking taxonomies
Taxonomy mapping must consider hierarchy.
If one target category is mapped as equivalent to a source category but their parent structures imply different scope, the equivalence may be misleading.
Inspect the neighbourhood around the node, not only the node label.
20. Crosswalking faceted systems
Faceted systems may differ not only in values but in dimensions.
One system may have one broad “level” facet while another separates curriculum stage, reading difficulty and audience.
Mapping values without mapping the facet meaning can produce category errors.
21. Crosswalking ontologies is deeper
Ontology alignment may involve classes, properties, relationships, domains, ranges and logical constraints.
Two classes can look similar while participating in different relationship models.
Ontology crosswalks therefore require semantic review beyond label matching.
22. Preserve mapping provenance
For each mapping, record:
- who proposed it;
- who approved it;
- which definitions were compared;
- which examples were tested;
- which version was used;
- which caveats apply.
A crosswalk is itself a knowledge claim and should be auditable.
23. Confidence belongs in the mapping
Some mappings are obvious. Others are expert judgements.
Store confidence or review status rather than presenting every mapping as equally certain.
24. Human review is essential for consequential mappings
Automatic matching can generate candidates from label similarity, definitions, examples and embeddings.
But when mappings affect legal status, medical classification, financial reporting or historical interpretation, high-impact changes should receive domain review.
25. AI can propose crosswalks
AI is useful for discovering candidate correspondences across large vocabularies.
It can compare labels, definitions, hierarchy and example records.
The output should still state mapping type, confidence and evidence rather than silently replacing one system with another.
26. Embeddings can reveal hidden overlap
Semantic embeddings may surface concepts whose labels differ greatly but whose descriptions are similar.
This improves recall in candidate generation, but semantic closeness does not prove equivalence.
27. Test mappings with real records
Do not validate a crosswalk only at the concept-definition level.
Take representative records classified under the source concept and see where they land in the target system.
If many records look wrong, the mapping is wrong or too coarse.
28. Test reverse mappings too
Map A → B, then examine whether B → A reconstructs the original meaning.
Lossy mappings will not round-trip perfectly. That is acceptable if the loss is known and documented.
29. Round-trip loss is a useful metric
If records translated A → B → A return to different categories, measure how often and why.
This reveals how much semantic detail the target system cannot preserve.
30. Crosswalks affect analytics
If old and new categories are aggregated without understanding splits and merges, trend lines can become misleading.
A rise in one category may reflect reclassification rather than a change in the world.
Analytical systems should know when a schema change affects comparability.
31. Crosswalks support federated search
A user can search one concept while the system expands the query into equivalent or related concepts in several external vocabularies.
This improves discovery across libraries and archives without forcing them to abandon their local classification systems.
32. Crosswalks support interoperability
Different systems can exchange data when they understand how their categories correspond.
Interoperability does not require identical internal models. It requires explicit translation contracts.
33. Crosswalks can expose irreducible differences
Sometimes the most important output is discovering that two systems do not mean the same thing.
That difference may reflect culture, jurisdiction, scientific theory, institutional purpose or historical period.
A good crosswalk preserves the distinction rather than flattening it.
34. Crosswalks need governance
Mappings change when either source or target changes.
Assign ownership for:
- mapping proposals;
- domain review;
- approval;
- version updates;
- migration testing;
- retirement of obsolete mappings.
Crosswalk maintenance is not a one-time conversion job.
35. Monitor unmapped concepts
An increasing number of unmapped source concepts may mean the target taxonomy is aging, the source is expanding or the mapping project has fallen behind.
Unmapped counts are a health signal.
36. Monitor ambiguous mappings
If one source concept repeatedly maps to several targets with low confidence, the difference may be structural rather than a documentation problem.
Do not force one target merely to make the table look complete.
37. A practical crosswalk record
- source scheme ID;
- source scheme version;
- source concept ID;
- source label;
- target scheme ID;
- target scheme version;
- target concept ID;
- target label;
- mapping type;
- direction;
- confidence;
- evidence note;
- reviewer;
- effective date;
- status.
This record can support migration, federated search and audit.
38. A complete crosswalk test
- Compare definitions.
- Compare scope.
- Compare hierarchy.
- Compare time validity.
- Compare authority.
- Assign mapping type.
- Test representative records.
- Test reverse direction.
- Measure information loss.
- Record uncertainty.
- Review high-impact mappings.
- Version the result.
39. The eduKateSingapore crosswalk approach
The Interlibrary Discovery & Crosswalk Wing sits inside a larger principle: knowledge systems should be able to meet without being forced into one vocabulary prematurely.
A crosswalk allows local ownership to remain intact while making traversal possible between libraries, disciplines and classification schemes.
This is especially important in a large knowledge estate where science, medicine, civilisation, education and archival systems may each require their own canonical structures.
40. The deeper idea
A crosswalk is not an attempt to prove that two maps are identical.
It is a disciplined account of how one map can be translated into another, where meaning is preserved, where it becomes broader or narrower, and where translation fails.
Interoperability begins when difference is described precisely enough to cross.
Final answer
Build classification crosswalks by mapping concepts, not labels. Record exact, broader, narrower, partial, split, merge and unmappable relationships explicitly. Preserve direction, provenance, confidence, versions and time. Test mappings against real records and measure information loss. Keep original classifications so migration does not erase history.
A good crosswalk does not make two knowledge systems the same. It lets them understand one another without losing the differences that made each system useful in the first place.