Talking Rabbit: A User Evaluation of Sentence Production

Ordnance Survey, the national mapping agency of Britain is researching the development of topographic ontologies to describe our spatial data, motivated by the business need of improving ease of understanding and reuse of our data, as well as assisting its integration with third party datasets. Through our experience of authoring topographic ontologies, we have found that the family of Web Ontology Languages (OWL), the W3C standard for ontology authoring, are difficult for domain experts to master. This insight led to the development of a controlled natural language, Rabbit [1], which is designed to make it easier for domain experts – people with expertise in a particular subject area, but not versed in knowledge modeling or Description Logics – to understand and author their own ontologies. With the possible exception of Sydney Syntax (SOS) [2], Rabbit differs from other controlled natural languages such as ACE [3] and CloNE [4] by limiting sentence complexity to ensure that sentences are simple and describe single concepts. Further more its development differs from other similar CNLs in two main ways. Firstly, its development has been informed by the building of several medium-scale (approximately 500-concept) ontologies, with updates to Rabbit being based on the requirements that have arisen from these realworld knowledge modeling needs. Secondly, we have modified Rabbit based on user evaluations: testing domain experts‘ understanding and authoring of Rabbit. Rabbit comprehension tests were reported in [5], and this paper now describes the results of our user evaluations of Rabbit authoring.