Verbalizing OWL in (controlled) English

1 Introduction

We describe how an ontology that is written in OWL RDF/XML is verbalized in Attempto Controlled English (ACE). This conversion is designed to be reversible, i.e. one can convert the ACE representation back into OWL RDF/XML so that no loss in meaning occurs.

Note that at the moment we do not support data properties. Also, very complex class descriptions that involve coordination are not supported. An example of such a complex description is e.g. (∃ p1 (∃ p2 C1)) ∩ (∃ p3 (∃ p4 C2)). Note also that only the logical content of the OWL file is verbalized, and not the information about annotations, versioning, import-structure, etc. Also, most of the URIs are truncated to preserve only the part that comes after the #-sign. As this can cause ambiguity in the names, some names are preserved as URIs.

2 Try it out!

3 How is the verbalization done?

We first rewrite the class descriptions and axioms via a limited number of more basic class descriptions and axioms. The motivation behind this step is to remove axioms like

ObjectPropertyDomain(R C)

which have very little syntactic structure and contain logic-specific words like `domain'.

The remaining class descriptions are mapped to ACE noun phrases and the remaining axioms to ACE sentences. OWL properties are ACE verbs and OWL individuals are ACE proper names. The following table shows an example how the OWL 2 constructs are mapped to ACE constructs.

Note that in the beginning of a sentence the complex noun phrases start with `Everything that' or `Nothing that'.

The next step is to make sure that we can verbalize more complex class descriptions which are built from the basic ones, e.g.

ObjectSomeValuesFrom(eat ObjectComplementOf(ice-cream))

ACE allows relative clause coordination and embedding of relative clauses into other relative clauses. This gives us quite a bit of flexibility, e.g. the above class description example would be verbalized as an ACE noun phrase

something that eats something that is not an ice-cream

Still, verbalizing very complex class descriptions would require a support for parentheses which ACE does not offer. We do not see it as a shortcoming, rather it is a natural restriction on the complexity of class expressions (which should be readable for others to understand).

Finally, we apply some simple modifications to make the sentences more readable, e.g. we try to remove negations and reorder the elements in coordination.

4 Examples

These example ontologies mostly originate from the Protege Ontology Library. Note that some ontologies import other ontologies. The ACE file is also "visualized" in HTML as an index: all names occurring in the ontology have been ordered alphabetically and each name is "explained" by listing all the sentences that contain the name.

The quality of the verbalization depends somewhat on the morphologic and orthographic nature of the names used for individuals, classes and properties in the input ontology. The verbalization assumes that all individuals are denoted by singular propernames (preferably capitalized), classes by singular common nouns, and properties by transitive verbs in their lemma form (i.e. infinitive form). Examples of individual names are `John', `London'; examples of class names are `man', `carnivore'; examples of property names are `own', `drink', `look-at'.

Those restrictions are needed because the names will be used in certain syntactic constructions or will undergo certain morphological changes.

• Individuals (propernames) are used in the subject and object positions without a determiner, e.g. "Every man knows John.", "John is a man."
• Classes (common nouns) are used in the subject and object positions with determiners every, no and a, e.g. "Every man knows something.", "No woman likes a spider.".
• Properties (transitive verbs) are often used in singular, e.g. "Every boy owns a bike.", but under negation (i.e. ObjectComplementOf) will stay in infinitive, e.g. "Every one-year-old is a child that does not own a bike.". In some cases, most often when verbalizing the ObjectPropertyRange-construct, the verb is turned into a past participle in order to construct a passive sentence, e.g. "Everything that is owned by something is a possession."

5 Implementation

The conversion is done in 3 steps.

1. From OWL RDF/XML to OWL Abstract Syntax, by Thea — An OWL library for [SWI] Prolog
2. From OWL Abstract Syntax to Attempto DRS, by a simple SWI-Prolog program
3. From DRS to ACE, by a DRS verbalizer described in http://rewerse.net/deliverables/m30/i2-d9.pdf

6 Known restrictions of the implementation

• The new features introduced in OWL 2 are not supported.
• Data properties are not supported.
• Very complex class descriptions are not verbalized correctly.
• ObjectAllValuesFrom, DisjointClasses, etc could be handled in a shorter way.
• Equivalent properties are mistaken for equivalent classes.
• Anonymous properties and individuals could be handled in a smarter way.
• Morphologic synthesis of 3rd person singular verbs (e.g. `owns') and past participles (e.g. `owned') is rather naive.