This paper presents a new OWL RL ontology, the Reasoning Violations Ontology (RVO), which describes both ABox and TBox reasoning errors produced by DL reasoners. This is to facilitate the integration of reasoners into data engineering tool-chains. The ontology covers violations of OWL 2 direct semantics and syntax detected on both the schema and instance level over the full range of OWL 2 and RDFS language constructs. Thus it is useful for reporting results to other tools when a reasoner is applied to linked data, RDFS vocabularies or OWL ontologies, for example for quality evaluations such as consistency, completeness or integrity. RVO supports supervised or semi-supervised error localisation and repair by de ning properties that both identify the statement or triple where a violation is detected, and by providing context information on the violation which may help the repair process. In a case study we show how the ontology can be used by a reasoner and a supervised repair process to accelerate high quality ontology development and provide automated constraint checking feedback on instance data. RVO is also being used to enable integration of reasoning results into multi-vendor data quality tool chains within the ALIGNED H2020 project.
A common task performed with Semantic Web reasoners is the detection and
reporting of errors or inconsistencies found in an ontology. This task frequently
occurs within the ontology authoring, interlinking, classi cation, quality
analysis and evolution phases of the linked data lifecycle [
To the best of our knowledge, there are no ontologies that have been developed
for the speci c purpose of describing reasoning violations. However, similar
problems have been addressed from di ering perspectives. One of the most relevant
contributions is OOPS! [
The purpose of RVO is to enable a reasoner to describe reasoning errors
detected in an input ontology in order to facilitate the integration of reasoners
into semantic web toolchains. It is de ned as a simple OWL 2 ontology that
is amenable to RDFS-based interpretations or use as a linked data vocabulary
without any dependence on reasoning. In future, an RDFS version of the
ontology is planned, in order to support interpretation by RDFS reasoners. A
permanent identi er for the ontology has been registered with the W3C
permanent identi er community group. The full source of the ontology is published
online4 and meta-data have been added to facilitate LODE-based [
a limited number of problems due to incomplete knowledge. The closed world interpretation assumes that you have provided all relevant aspects of the model and is able to detect a much wider range of violations, e.g. missing or misspelled term de nitions. This is often useful during ontology development or in a system that interprets OWL as a constraint language. 3.1
The ontology can be divided into two layers. The top layer consists of the base classes and their properties and the bottom layer is a vocabulary which de nes the hierarchical structure of violations identi ed so far (see 3.2).
Figure 1 shows the top tier of the ontology which represents general metadata about a Violation as well as related properties, elements and classes. Class violations are used for reporting issues regarding the TBox and instance violations ABox in general. Therefore, class violations are reported when e.g. property domains are missing, subsumption errors are detected, or class and property cycles are found. Instance violations show instances which are not elements of valid classes, cardinalities which are incorrect, property constraints that are violated, literals and objects which are confused, etc. 3.2
We have organised reasoning errors in violation classes and put them in a hierarchical structure. The top level di erentiation of violations is instance or schema, depending on whether the violation occurred in the ABox (instances) or TBox (schema). Here is an overview of all errors currently documented: { Instance
InstanceBlankNode
NotAnElement: NotRestrictionElement, ObjectInvalidAtClass, EdgeOrphanInstance, DataInvalidAtDatatype (NotBaseTypeElement) InstanceProperty: NotPropertyDomain, InvalidEdge, NotFunctionalProperty, LocalOrphanProperty, NotInverseFunctionalProperty { Schema
ClassViolation: NotUniqueClassLabel, NotUniqueClassName, NotDomainClass, ClassCycle, NoImmediateClass (NotSuperClassOfClass), OrphanClass (NotIntersectionOfClass, NoSubclassOfClass, NotUnionOfClass) PropertyViolation: PropertyCycle, NotUniquePropertyName, SchemaBlankNode, PropertyTypeOverload, PropertyAnnotationOverload, OrphanProperty (NotSubpropertyOfProperty), PropertyDomain (InvalidDomain, DomainNotSubsumed, NoExplicitDomain), PropertyRange (InvalidRange, RangeNotSubsumed, NotExplicitRange) 3.3
In our example a reasoning error is asserted rst in JSON as raw data and then converted to RDF triples using RVO in order to be consumed in Dacura. The example shows a ClassViolation which is a SchemaViolation and more speci cally a ClassCycleViolation. Such speci c violation detection results make it possible to provide exact suggestions to ontology developers or repair agents and trigger ontology improvements. 1 { " rdf : type ": " ClassCycleViolation ", 2 " bestPractice ": { " type ": " xsd : boolean ", 3 " data ": " false "} , 4 " message ": " Class UnitOfSocialOrganisation has a class cycle with path : [
TemporalEntity , UnitOfSocialOrganisation ]" , " path ": [ " seshat : TemporalEntity ", " dacura : UnitOfSocialOrganisation " ], " class ": " seshat : UnitOfSocialOrganisation "
The response RDF graph provides a much better way to interpret the results: The instance is called example1 and is a ClassCycleViolation. bestpractice is false, so it is an error rather than a warning. The message provides a summary of the cause of the violation, but the important parts are the next two properties. The path property marks all classes which were involved in the cycle and the class property marks the class where the cycle has been detected. Another important feature is that RVO provides us direct links to the Seshat ontology (an ontology which models the Seshat Global History Databank6) and hence to the OWL classes from the external ontology which were involved in the violation process. 4
The Dacura Toolchain Case Study covers toolchaining and reporting to users. At this point we want to have a closer look into supervised ontology repair and publishing.
In the use case scenario in Figure 2, the Dacura Shema Service gets an external ontology which is loaded by the Dacura Schema Manager and needs to be validated into the knowledge base (step 1). The ontology is then send to DQS and checked for errors by the DQS reasoner (step 2), whose rules comply with the violation classes of RVO. The detected violations are sent back to the client (step 3). The results are classi ed and assigned classes from RVO (step 4). Dacura then integrates the report in the UI and presents it to the user (step 5). Finally, the user can repair the ontology in an editor and republish it to the Web (step 6).
So how did RVO help in this speci c scenario? After the Data Quality Service checked the external ontology for errors, it used the RVO structure to provide information about a speci c violation by creating individuals. This provided us with a classi cation of DQS' results. RVO has been used by the client for classi cation of the errors and for providing relevant information about violations to the user. An additional bene t would be to archive the history of errors in a knowledge base and be able to query for certain occurrences of violations for an ontology.
Table1 shows the validation of ALIGNED ontologies7 which have all been developed by using di erent approaches (Protege, RDF2RDF, human checks, etc.). We have used the Data Quality Service and RVO to validate several project ontologies and report the errors and warnings found in a rst run. Although, this is only a case study and especially the validation and correction e orts
are estimated, the table signalises the potential for improvement of existing ontologies. 5
In this paper we have shown that a dedicated reasoning error ontology improves error reporting with structured data, and integration of the ontology in the Dacura toolchain case study. The Reasoning Violations Ontology not only bene ts the interpretation and further processing of reasoning errors in tools, platforms, and Web UIs which present results of the reasoning process or ontology validation, but can also be used as a common format to represent violations found in ontologies during the whole software toolchain process. We have shown an example of supervised ontology repair use case and explained the advantages of our approach. Furthermore, we have given some examples for reasoning violations and constructed RDF graphs to present the results. Our future work will continue with the integration of the ontology in the ALIGNED toolchain and linking of the ontology to SHACL constraints as well as using it together with RDFUnit. Finally, we plan to evaluate the bene ts in a case study with ontology engineers and investigate their work with the ontology and our tools in order to improve the quality of their ontology or repair it.
This research has received funding from the European Unions Horizon 2020 research and innovation programme under grant agreement No 644055, the ALIGNED project (www.aligned-project.eu) and from the ADAPT Centre for Digital Content Technology, funded under the SFI Research Centres Programme (Grant 13/RC/2106) and co-funded by the European Regional Development Fund. References