Abstract The representation of design information through ontologies has proven to be effective in fostering creative ideation within product design. Consequently, researchers have developed databases comprising models of engineering and biological systems by leveraging ontologies. However, the manual construction of a large number of models from technical documents is an effort-intensive task that demands specialized expertise. To address this challenge, researchers have investigated automatic information extraction methods utilizing data-intensive machine-learning models. However, previous research has not fully documented the end-to-end process of information extraction and representation and has not reported the end-to-end accuracy. This study introduces a novel method for automatic information extraction pertinent to the State change-Action-Part-Phenomenon-Input-oRgan-Effect (SAPPhIRE) model of causality alternative to creating data-intensive machine-learning models. This method employs the dependency parsing technique of natural language processing, along with rules supported by a lexical database, to extract words relevant to the SAPPhIRE model. Unlike previous approaches that rely on supervised learning methods, this new technique does not require extensive datasets for the training and validation of machine-learning models. Furthermore, it reports the end-to-end accuracy of information extraction, rather than focusing solely on the word classification task, which is preceded by manual pre-processing in prior research. The results of this newly developed method have been validated against SAPPhIRE models reported in the literature and through input provided by SAPPhIRE specialists and design researchers.