Automatically extract common patterns from an OWL Ontology

Creating large ontologies or terminologies is a resource-intensive task, with the possibility to introduce human error. Use of description logic languages such as OWL-DL accompanied by reasoning can automate some parts of ontology construction and validation, but adding the appropriate OWL axioms can itself be a time consuming process requiring expert ontologists.

the OWL patternizer looks for stereotypical patterns in ontologies and vocabularies (both logical and lexical), and generalizes these to DOSDPs

For example, if an ontology has lots of terms like:

• abnormal limb morphology
• abnormal head morphology
• abnormal epiphysis morphology
• increased limb size
• decreased limb size
• increased head size
• decreased head size

With logical definitions, then it will detect the OWL patterns for these. If there are no logical definitions present, it will do a best effort at inducing logical definitions from lexical patterns.

The approach is intended to be knowledge-free. Minimal underlying assumptions about the ontology. As such, the generated yaml will be more generic than hand-crafted. The intent is that it is used to bootstrap and analyze an ontology.

Additionally, if your starting point is a thesaurus-like ontology (no logical defs), there is a step for generating candidates.

Given as input PATO, an ontology which contains implicit patterns of the form {increased, descreased} X, DPs such as the following are generated:

pattern_name: X increased_in_magnitude_relative_to normal
pattern_iri: http://purl.obolibrary.org/obo/pato/X_increased_in_magnitude_relative_to_normal

description: >-
  This is auto-generated. Add your description here

  Examples: [increased quality](http://purl.obolibrary.org/obo/PATO_0002300), [increased speed](http://purl.obolibrary.org/obo/PATO_0000303), [increased age](http://purl.obolibrary.org/obo/PATO_0001764) (105 total)

classes: 
  quality: "PATO:0000001"
  normal: "PATO:0000461"

relations: 
  increased_in_magnitude_relative_to: "http://purl.obolibrary.org/obo/pato#increased_in_magnitude_relative_to"

vars:
  v0: "'quality'"

name:
  # Induced, frequency=0.5619047619047619, p=http://www.w3.org/2000/01/rdf-schema#label 
  text: "increased %s"
  vars:
    - v0

def:
  # Induced, frequency=0.24761904761904763, p=http://purl.obolibrary.org/obo/IAO_0000115 
  text: "A %s which is relatively high."
  vars:
    - v0

annotationProperties:
  exact_synonym: "oio:hasExactSynonym"
  related_synonym: "oio:hasRelatedSynonym"

annotations:
  - annotationProperty: exact_synonym
    # Induced p=exact_synonym 
    text: high %s
    vars:
      - v0

equivalentTo:
  text: "%s and ('increased_in_magnitude_relative_to' some 'normal')"
  vars:
    - v0

The patterns are annotated with comments indicating provenance and confidence.

Currently this only looks for equivalence axioms between a named class and a class expression (aka logical definitions), where the class expression uses the following constructs (arbitrarily nested):

• some
• only
• and
• or
• named classes
• named object properties

There are two ways to do this

1. via Docker
2. direct via swipl

For 1, no additional installation required. TODO: document this

For 2, you will need to install sparqlprog

TODO

For purely auto-generated examples, see:

• examples/

(may be moved to other repo)

• range validation gets thrown off if there are dangling classes

• google doc