In architectural design words are an unemployed source of information. Through a series of case studies we deduced a design annotation data model. All entities is tThis model can be captured from the design draft, and includes additional informationexcept one: the word-relation. Therefore a system was developed that can generate words graphword graphs using single words as input. The system searches for semantic relations between words and for new in-between words that can connect two existing words. Filters were developed to filter out only those graphs that were considered interesting by the designers. The envisioned applications of word graphs in the context of Computer Architectural Aided Design, are the reduction of fixation during the design process and enhancing creativity by inferring novelty.