Human reasoning and description logics : applying psychological theory to understand and improve the usability of description logics

• Main Author: Warren, Paul
• Published: Open University 2017
• Subjects: 006.3
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.724879

Description Logics (DLs) are now the most commonly used ontology languages, in part because of the development of the Web Ontology Language (OWL) standards. Yet it is accepted that DLs are difficult to comprehend and work with, particularly for ontology users who are not computer scientists. The Manchester OWL Syntax (MOS) was developed to make DLs more accessible, by using English keywords in place of logic symbols or formal language. Nevertheless, DLs continue to present difficulties, even when represented in MOS. There has been some investigation of what features cause difficulties, specifically in the context of understanding how an entailment (i.e. an inference) follows from a justification (i.e. a minimal subset of the ontology that is sufficient for the entailment to hold), as is required when debugging an ontology. However, there has been little attempt to relate these difficulties to how people naturally reason and use language. This dissertation draws on theories of reasoning from cognitive psychology, and also insights from the philosophy of language, to understand the difficulties experienced with DLs and to make suggestions to mitigate those difficulties. The language features investigated were those known to be commonly used, both on the basis of analyses reported in the literature and after a survey of ontology users. Two experimental studies investigated participants’ ability to reason with DL statements. These studies demonstrate that insights from psychology and the philosophy of language can be used both to understand the difficulties experienced and to make proposals to mitigate those difficulties. The studies suggest that people reason using both the manipulation of syntax and the representation of semantics with mental models; both approaches can lead to errors. Particular difficulties were associated with: functional object properties; negated conjunction; the interaction of negation and the existential or universal restrictions; and nested restrictions. Proposals to mitigate these difficulties include the adoption of new language keywords; tool enhancement, e.g. to provide syntactically alternative expressions; and the introduction during training both of De Morgan’s Laws for conjunction and disjunction, and their analogues for existential and universal restrictions. A third study then investigated the effectiveness of the proposed new keywords; finding that these keywords could mitigate some of the difficulties experienced. Apart from the immediate applicability of these results to DLs, the approach taken in this dissertation could be extended widely to computer languages, including languages for interacting with databases and with Linked Data. Additionally, based on the experience of the three studies, the dissertation makes some methodological recommendations which are relevant to a range of human-computer interaction studies.