Verbalizing but Not Just Verbatim Translations of Ontology Axioms

In this paper, we propose an inference-based technique to remove redundancy from natural language (NL) descriptions of Web Ontology Language (OWL) entities. The existing ontology verbalization approaches generate NL text segments that are closer to their counterpart statements in the ontology. Some of these approaches also perform grouping and aggregating of the text segments, aiming at a more fluent and comprehensive representation. However, we observed that the human-understandability of such descriptions is affected by the presence of repetitions and redundancies, and our studies show that such issues can be removed easily at the semantic level than at the NL level. We propose a novel technique called semantic-level refinement (or simply, semantic-refinement) for this purpose. Our approach aims at transforming the knowledge that is represented as a combination of less expressive (and not specific) logic-based expressions into the ones with high expressivity and specificity. This technique utilizes a predefined set of rules which are applied repeatedly on the restrictions associated with the individuals (and the concepts) to obtain a refined set of restrictions, guaranteed to be semantically equivalent to the original representation. Such refined sets of restrictions can then be verbalized to get concise descriptions of the ontology entities. Our experiments on ontologies from two different domains show that the proposed approach could significantly improve the readability of the NL texts when compared to the texts generated without a semantic-level refinement.