Is a crosswalk always one field to one field?

No. It can include one-to-many, many-to-one, conditional, or unsupported correspondences.

A converter can implement sufficiently precise rules, but domain review remains important when meanings differ.

Does a published crosswalk solve every local requirement?

No. Local fields, profiles, and application expectations may require documented adaptations and additional tests.

AI Glossary / Data engineering and architecture / Intermediate

Schema crosswalk

Definition

A schema crosswalk documents how elements in one metadata or data schema correspond to elements in another. It may identify equivalent fields, conditional mappings, transformations, and information that has no adequate destination. Crosswalks are especially common in libraries, archives, and data exchange. A crosswalk can guide implementation, but a list of field names alone is not necessarily a complete, executable conversion specification.

What it means

Two organizations may describe similar things using different structures and vocabulary. One separates a person's family name and given name; another has a single creator field. A crosswalk makes the proposed relationship explicit. It also reveals limitations. Combining two fields can be straightforward, while reconstructing the original two fields from a single string may be ambiguous. Good crosswalks document these asymmetries instead of implying that all mappings work equally well in both directions.

How it works

Start with the source and target schema versions and the purpose of exchange. Compare definitions, cardinalities, permitted values, data types, and context. For each mapping, state any selection condition, transformation, vocabulary conversion, and handling of missing or repeated values. Mark unmapped information and intentional losses. Implement the agreed rules in a converter, then validate representative records. Treat updates to either schema as a reason to review the crosswalk and its tests rather than assuming old mappings remain correct.

Three practical examples

Library exchange: a catalog team uses a documented crosswalk to export richer bibliographic descriptions into a simpler discovery format, acknowledging details that the simpler format cannot retain.
Research repositories: two institutions map their project, contributor, and identifier fields to a shared deposit format, specifying which identifier scheme each value uses.
Business integration: an analyst documents how a supplier's product measurements and units map to a purchasing system, including conversions and rules for unavailable measurements.

Think of it this way

Think of a bilingual form guide that explains which box on one form corresponds to which box on another. A box labeled “date” is not useful until the guide says whether it means creation, publication, or approval. The analogy is limited because a technical crosswalk can include conditional transformations and repeated structures, not just labels connected by arrows. It may also explicitly say that no suitable destination exists.

Worked example

Suppose the source has title, creator_given, creator_family, and issued_year. The target has title, creator, and date. For a toy record, title is “River Survey,” creator_given is “Maya,” creator_family is “Stone,” and issued_year is 2024. The crosswalk copies title, creates creator as “Stone, Maya,” and maps issued_year to date only when date is defined as publication date. The target record is valid under those assumptions. A reverse converter cannot safely split every creator value on a comma: an organization may contain punctuation, and some source names may already be formatted differently. The crosswalk should identify this limitation and avoid promising lossless round-tripping.

Illustrative case study

Fictional example: Pine Archive shares collection descriptions with a regional portal. Its initial mapping sends both digitization dates and original publication dates into one destination field. Reviewers notice that portal users could misinterpret the result. The teams agree on a date-type rule and retain additional dates in a supported qualified field where available. They test old, missing, and approximate dates before exchange. This invented scenario illustrates semantic review rather than a documented outcome for a real archive or metadata standard.

Common misconceptions and mistakes

Do not map fields solely because their names match. Do not assume “optional” in one schema means “unknown” in the other. Repeated source values may be lost when the target allows only one. A target-required field may need an explicit missing-data policy rather than an invented value. Also distinguish a crosswalk from schema matching: matching helps discover correspondences, while the crosswalk records chosen rules. A correct forward conversion does not establish that the reverse direction is lossless.

Business use

A useful crosswalk serves as a reviewable contract between data owners and implementers. Include examples beside difficult rules and assign ownership for changes. Measure rejected records and unmapped fields during trial conversions. Preserve source identifiers and provenance where the exchange permits it, so problems can be traced back. For AI datasets, document how label definitions and annotation categories map; changing a label name without checking its meaning can silently change the task the model is trained to perform.

Check your understanding

A source records “publication year,” while a target field named “date” means the last modification date. Should the values be copied directly because both are dates?

Answer: No. Their meanings differ. The crosswalk must find an appropriate destination or record that the information cannot be represented under the proposed mapping.

Related terms

Schema matching
Controlled vocabulary
Round-tripping (computer science)
Semantic interoperability
Metadata mapping
schema transformation
application profile
round-tripping

Learn next

map a five-field sample record to a simpler three-field format. List what is copied, transformed, merged, and lost. Ask someone who knows the source domain to explain the returned record without seeing the original, then investigate any changed interpretation.

Sources

Primary sources consulted: Library of Congress, MODS Conversions, https://www.loc.gov/standards/mods/mods-conversions.html — excerpt on maintained conversion stylesheets and local adaptation consulted. Library of Congress, MARC to Dublin Core Crosswalk, https://wwws.loc.gov/marc/marc2dc.html — mapping and round-trip limitation excerpt consulted. Examples and review workflow are original; full editorial review remains pending.

Previous: Schema-agnostic databases Next: Schema heterogeneity