This schema is helpful for writing “movable” RDF schemas. (We use the term “schema” to include “ontology” and “vocabulary definition”, since the terms are equivalent in this context.) RDF schemas are generally useful for providing interoperability between loosely coupled systems, but they require URLs which remain stable over many years.

Movable schemas, in contrast, allow interoperability even when moved to different URLs. For example, while the original version of this document was (and perhaps still is) hosted at https://sandhawke.github.io/mov, it is possible (and recommended) to use this schema in ways which do not involve trusting either github or the author, such as by forking it or copying it to your own server.

In fact, due to limitions of github hosting, the namespace URL for this schema is different for different formats. This would normally break interoperability, but with movable schemas, it’s fine.

The provided formats and locations/namespaces are:

Format Media Type Namespace URL
Turtle text/turtle https://sandhawke.github.io/mov/schema.ttl#
JSON-LD application/ld+json https://sandhawke.github.io/mov/schema.jsonld#
JSON-LD embedded in HTML text/html https://sandhawke.github.io/mov#

For JSON-LD, the location/namespace is also the @context URL.

Concept

The basic concept is that we identify each RDF property (and class, etc) using some definition text, not its URI. It still has a URI, but the URI doesn’t have to match other references to the property, because the match will also be done using the definition text. What’s needed is that the same definition text be used by each data source being merged and each system consuming the data. In some ways this is more cumbersome than using the same URI, but in other ways it is simpler.

Practice

In practice, to create a movable schema, you need to provide suitable definition text for each resource in your schema. Those definition are provided using this mov schema. The definitions should each be a few lines of text, specifying the item well enough that a careful expert reader who is confident they understand the text is very likely to be right. (For machine purposes, the important thing is that the text has enough entropy that it is statistically unlikely to accidentally be the same as someone else’s definition of another term. By requiring the definitions to be meaningful and unambiguous to people, we can have relatively short strings which should only be coincidentally selected when the meaning is the same.)

These definitions can be embedded in your data or provided nearby, like by a schema file in the same directory. You can also point to the original provider, as you would be with a non-movable schema, if you happen to trust them sufficiently.

To consume movable schemas, with all the data integration benefits of non-movable schemas, an additional software layer is required to first do the merging based on definition matches. This layer can be built into tools or executed before the data is loaded into conventional tools. (It might be possible to build a translating proxy, but https would make this very difficult.)

Terms

mov:itemdef

This can be used for defining any resource. In practice, it should only be used for resources which are not an rdf:Property or an rdfs:Class, since they are more easily defined using mov:propdef and mov:classdef.

Example

PREFIX mov: <https://sandhawke.github.io/mov/schema.ttl#>
PREFIX : <.#>
:Tatooine mov:itemdef "A fictional location, a planet called 'Tatooine' in the Star Wars franchise. It was introduced in the 1977 film _Star Wars_ as the home of protagonist Luke Skywalker.".

mov:propdef

This is used for defining RDF properties, such that they can be textually matched even when using different URIs.

Example 1

This example shows definition and instance data in same file, and uses a quite verbose dictionary-style definition.

PREFIX mov: <https://sandhawke.github.io/mov/schema.ttl#>
PREFIX : <.#>
:familyName mov:propdef "The family name of some person. This name is usually assigned at birth, identical to or derived from the parents' family names. Siblings in the same family are usually given the same family name, and family names usually only change with adoption or, for some people in some cultures, marriage. Typically, a person's name is their family name and given name combined in an order that depends on context.".
:Luke :familyName "Skywalker".

Example 2

This example shows the definition and instance data in different files on the same site. It also uses template-style definition, instead of dictionary-style. This style makes use of the feature that matching ignores contents in square brackets.

This example also pushes the limit of short definitions. Given the wide consensus on what “family name” means, this is probably okay for most applications.

At https://example.org/schema:

PREFIX ex: <https://example.org/schema#>
ex:familyName mov:propdef "[subject ref] has the family name [value string].".

### definition from  <https://sandhawke.github.io/mov/schema.ttl#>

At https://example.org/data:

PREFIX mov: <https://sandhawke.github.io/mov/schema.ttl#>
PREFIX ex: <https://example.org/schema#>
:Luke ex:familyName "Skywalker".

In practice the mov: schema definitions might also be copied into https://example.org/schema, to avoid external dependencies.

mov:classdef

tbd

Modifying your definitions

Often, after some experience using term definitions you developed, you will think of better wording, or otherwise want to change the text. We suggest keeping the old wording if anyone might be using it or interested in it, and just adding the new wording, as a new definition.

There are two ways to add these new defintions:

  1. If the change is purely editorial and does not reflect a change in your intended meaning, then the change is nonbreaking. You can simply add the new definition to the same term. This asserts you consider the definitions to be synonymous.

  2. If the change does reflect a change in your intended meaning, then it is substantive and is a breaking change. If systems conflate the two meanings, they might get incorrect results. For a change like this, you need to make up a new term URI and put the new definition on that term. (You can rename terms if, and only if, all supported consumers are movable-schema-aware.)

Consumers do have the ability to override your decision, above, in that they can choose which definitions to use for each of the terms they consume. When you treat the change as editorial/nonbreaking, they might decided not to use one of the definitions you equate, because they do not consider them sufficiently synonymous. Alternatively, if you treat the change as substantive/breaking, they might decide for their application the definitions are close enough and equate them.

Implementations

schemove

Technical Details