One of the main benefits of working with OWL is the availability of powerful automated reasoners. Reasoning involves two steps: logical validation (described in detail below) and automatic classification. Automatic classification involves asserting all inferred superclasses.
There are several reasoners available, and each has different capabilities and characteristics. For this example we’ll be using ELK, a very fast reasoner that supports the EL subset of OWL 2.
robot reason --reasoner ELK \ --input ribosome.owl \ --output results/reasoned.owl
It’s also possible to place the new inferences in their own ontology with
robot reason --reasoner ELK \ --create-new-ontology true \ --input ribosome.owl \ --output results/new_axioms.owl
--create-new-ontology-with-annotations true will place the inferences in their own ontology, along with their annotations (e.g. labels, definitions, etc.).
Sometimes it is important to know which axioms were inferred (designated by an axiom annotation “is_inferred true”):
robot reason --reasoner ELK \ --annotate-inferred-axioms true \ --input ribosome.owl \ --output results/reasoned.owl
reason includes two more options to help clean the reasoned output:
--exclude-duplicate-axioms: if set to true, the axioms will not be added to the output if they exist in an import (default
--remove-redundant-subclass-axioms: if set to false, redundant axioms (those that have been asserted and were also inferred) are included in the output (default
--reasoner is provided, ROBOT will default to ELK. The following other reasoner options are supported:
emr- Expression Materializing Reasoner
structural- Structural Reasoner
ROBOT will always perform a logical validation check prior to automatic classification. Formally, this is known as testing for incoherency, i.e. the presence of either a logical inconsistency or unsatisfiable classes. If either of these hold true, the reason operation will fail and robot will exit with a non-zero code, after reporting the problematic classes.
You can perform detailed debugging using an environment like Protege - load the ontology, switch on the reasoner and use the explain feature. For example, if you have unsatisfiable classes, find one of them (they should show up red) and click on the
? where it says
If you are working on a large complex ontology with multiple imports and you encounter unsatisfiable classes during the release, you can make a minimal ontology for debugging purposes using the
--dump-unsatisfiable) option folled by an output file path. This will find all unsatisfiable classes and use the extract operation to create a debug module.
robot reason --reasoner ELK \ --input incoherent-tbox.owl \ -D results/debug.owl
If the input module contains at least one import, axioms in the debug module will be tagged with the source ontology, to assist in debugging.
By default, ROBOT ignores one-to-one equivalent classes (e.g.
A EquivalentTo B). But, in many cases, inferring equivalence between two classes is usually a sign that something has gone wrong.
Sometimes we want to avoid equivalence between named classes at all (in the case of OBO, where we strive for orthogonality). This can be done through
--equivalent-classes-allowed <arg> (shorthand
<arg> options are:
all: always allow (default behavior)
none: never allow
asserted-only: allow asserted equivalence, but throw an error if equivalence is inferred
A tautology is an axiom that would be true in any ontology, i.e., it’s just a given due to the semantics of OWL. For example, every class is a subclass
owl:Thing; it is not generally useful for an ontology to explicitly state this. Some reasoners may however include these kinds of statements as part
of their output. The ROBOT
reason command provides an option for filtering out tautologies generated by the reasoning process:
The options for
false(default): allow any generated tautologies in the output.
all(recommended): use the HermiT reasoner to exclude any inferred axioms that would be entailed by an empty ontology.
structural(fast): exclude axioms matching a hard-coded set of tautological patterns (e.g.,
X SubClassOf owl:Thing,
owl:Nothing SubClassOf X,
X SubClassOf X). Much faster than
By default, the
reason operation will only assert inferred subclass axioms. This can be configured with the
--axiom-generators option. OWLAPI provides the following inferred axiom generators:
One or more of these axiom generators can be passed into the option (separated by spaces). For example, to generate both subclass and disjoint class axioms:
robot reason --input unreasoned.owl --axiom-generators "SubClass DisjointClasses" --output reasoned.owl
If you are only passing one axiom generator, it does not need to be surrounded by double quotes.
The axioms that are generated for subclasses, class assertions, and sub-object properties can be either direct or indirect. By default, the operation will only generate direct axioms (
--include-indirect false). It is recommended to pipe to
reduce after to remove any redundant axioms. For example, to generate all direct and indirect superclass expressions:
robot reason --input unreasoned.owl --reasoner emr --include-indirect true reduce --output reasoned.owl
You must select between
-n). Both cannot be passed in as
true to one reason command, as they have opposite results.
The input for the
--axiom-generators option must be one or more space-separated valid axiom generators, listed above.
ROBOT has been configured to not allow one-to-one equivalent classes (
--equivalent-classes-allowed none or
--equivalent-classes-allowed asserted-only) and one or more equivalency has been detected. Either remove the axiom(s) causing the equivalence axiom(s), or run
--equivalent-classes-allowed none. More details here.