Extracting Information for Creating SAPPhIRE Model of Causality from Natural Language Descriptions

Structured representations from natural language descriptions of biological and engineered systems are a good source of inspiration in analogical design. Researchers proposed methods for developing knowledge representations from such documents, to make them conducive for use as a source of analogy. Ontology-based representations, such as FBS, SBF, SAPPhIRE, etc. are effective in analogical design, but manually creating accurate descriptions using these models is both time- and resource-intensive. Hence, methods to automatically create ontology-based representations are useful for developing a repository of biological and engineered systems. However, such methods are partially automated, with major human decision-making touch-points. Before standardizing and automating the process, it is important to understand it end-to-end for accuracy and variability. This paper reports results from a detailed study on manual information extraction from systems description texts, using the SAPPhIRE model. A new process is proposed that aims to reduce variability in the extracted information across subjects, with preliminary results that show significant promise.