TiQi: answering unstructured natural language trace queries

Software traceability is a required element in the development and certification of safety-critical software systems. However, trace links, which are created at significant cost and effort, are often underutilized in practice due primarily to the fact that project stakeholders often lack the skills needed to formulate complex trace queries. To mitigate this problem, we present a solution which transforms spoken or written natural language queries into structured query language (SQL). TiQi includes a general database query mechanism and a domain-specific model populated with trace query concepts, project-specific terminology, token disambiguators, and query transformation rules. We report results from four different experiments exploring user preferences for natural language queries, accuracy of the generated trace queries, efficacy of the underlying disambiguators, and stability of the trace query concepts. Experiments are conducted against two different datasets and show that users have a preference for written NL queries. Queries were transformed at accuracy rates ranging from 47 to 93 %.