Knowledge of Technological Artifacts: Investigating the Linguistic and Structural Foundations

Design and innovation processes critically depend on the synthesis of knowledge embedded in existing technological artifacts. As generative models, such as large language models (LLMs), increasingly support knowledge-intensive tasks, understanding how artifact-level knowledge is represented becomes essential. In this study, we analyze a large, stratified sample of 33,881 patented artifact descriptions across the total technology space. We populate knowledge graphs of these descriptions by extracting factual triplets (entity :: relationship :: entity) at the sentence level. By studying the knowledge structures explicated by these graphs, we uncover the linguistic and structural foundations of technological artifacts. Linguistically, we identify syntactic patterns that explain how entities and relationships are constructed at the term level. Structurally, we identify motifs, including dominant 3-node and 4-node subgraph patterns, that reveal how entities and relationships are combined locally in artifact descriptions. Delving into these motifs reveals that natural language artifact descriptions primarily capture the design hierarchy of artifacts. At a local level of artifact descriptions, the motif analyses reveal that only abstract technical knowledge is captured, informing potential limitations for AI-driven innovation research and practice that heavily rely on text data for knowledge-intensive tasks.