DSNL IN ARCHITECTURE- A DEEP LEARNING APPROACH TO DECIPHERING ARCHITECTURAL SKETCHES AND FACILITATING HUMAN-AI INTERACTION

The language of interaction between architects and machines has been evolving towards a more user-friendly paradigm. As the capabilities of machines and artificial intelligence have advanced, it has become increasingly feasible for architects to communicate with machines using their customary expressive methods. Consequently, this has led to the development of Domain-Specific Natural Language (DSNL), which, unlike traditional Domain-Specific Language (DSL), places greater emphasis on naturalness. While this naturalness enhances usability for architects, it also presents challenges in machine comprehension. To address this issue, we propose a data-driven approach that utilizes domain-specific data for model training or fine-tuning through unsupervised or weakly supervised methods. Our study, which focuses on teaching AI to learn architectural sketching from architects, demonstrates that our proposed method captures the characteristics of human architectural sketching more effectively than traditional approaches.