WHAT ARCHITECTS AND STUDENTS SEE IN ARCHITECTURAL DESIGN SKETCHES : A PROTOCOL ANALYSIS

The present research aims at examining what information architects think of and read off from their own freehand sketches, and at revealing how they perceptually interact with and benefit from sketches. We explored this in a protocol analysis of retrospective reports; each participant worked on an architectural design task while drawing freehand sketches, and later reported what she/he had been thinking of during the design task. This research lies within the scope of examinations of why freehand sketches as external representation are essential for crystallizing design ideas in early design processes. External representations such as diagrams, sketches, charts, graphs, and even hand-written memos not only serve as memory aids, but also facilitate and constrain inference, problem-solving and understanding. Geometry diagrams in theorem-proving tasks, for example, guide solvers to explore only visually plausible inference paths1, facilitate retrieval of perceptual-chunks that are useful for constructing efficient proofs2,3, and provide visual cues for extracting new chunks from the current problem and assimilating them for future use4. Petre5 showed that good use of graphical representations in programming environments, i.e. what she calls 'secondary notation' of graphics, prevents programmers from mis-cueing and mis-understanding. Larkin and Simon6 enumerated general features of diagrammatic representation by which human problem-solving is facilitated, providing a list of how facilitation tends to occur. Tversky discussed how people use space for conveying meanings and abstract concepts, drawing on examples from ancient depictions and children's early drawing7 as well as from comtemporary charts, graphs and diagrams8. Facilitation by external representation derives, not just from its external existence but from the interaction between the representation and the cognitive processes of interpreting it9. Architects' sketches are a tool for this sort of interaction as well10,11. Architects put ideas down on paper and inspect them. As they inspect their own sketches, they see unanticipated relations and features that suggest ways to refine and revise ideas. This cycle -sketch, inspect, revise -is like having a conversation with one's self12. Goldschmidt13 conjectured that sketches give access to various mental images, figural or conceptual, that may potentially trigger ideas in the current design problem. Further, she claimed that visual design thinking is a rational mode of reasoning as well, although it has been set aside behind the dominant paradigm of linguistic, logical reasoning in cognitive science14. This claim perfectly coincides with a growing enthusiasm for diagrammatic reasoning in cognitive science, especially groups of researchers who claim that visual information is valid for reasoning and that visual reasoning has its own sound logic15,16, just as conventional sentential reasoning does. Why are sketches a good medium for reflective conversation with one's own ideas and imagery? This general question can be reduced to more precise issues to be addressed. One is an issue of what aspects or features of sketches themselves as external representations allow for reflective conversation, an issue addressed by Goel17. He found that because freehand sketches in the early design process are "dense" and "ambiguous" in Goodman's18 sense, they work well for crystallizing design ideas. Another is an issue of what kinds of interaction architects have with their own sketches. This issue can, in turn, be reduced into a couple of issues; "how do they see sketches?", "what do they see in sketches?", and "how and what do they draw?". Goldschmidt's work13 pertains to the first category. She observed that there are two ways of inspecting sketches, i.e. "seeing as" and "seeing that", and that the former is an especially powerful means for what she calls interactive imagery. Van Sommers's work19 looked at graphic production from a developmental and cognitive perspective, and hence pertains to the third category. The present paper addresses the second issue, "what architects see", by focusing more precisely than the past work on the contents of information categories that architects "see" in their own sketches. The purpose of the present paper is to analyze how those different types of information intermingle with each other in their design thoughts, and to reveal how practicing architects differ from students in it. We brought these phenomena into the laboratory in a protocol analysis of retrospective reports of subjects' design thoughts. The most typical method for analyzing subjects' cognitive processes is, and has been, concurrent thinking-aloud verbal reports20,21. We did not employ it because talking aloud may adversely interfere with participants' perceptions during their sketching activities22. This issue is discussed in more detailed in Section 5.1. Another purpose of this study is to explore implications for ways that future design tools, especially sketching tools, assist designers/architects. Many researchers17,23,24 claim that the currently available computational tools do not support naive freehand sketching processes in the early design phases. This is because they are intended for visualizing, comparing, testing and implementing the design ideas that have been already obtained in earlier design processes, not for supporting the very processes in which design ideas occur. Several projects23,24,25 on pen-based sketch tools have recently addressed this problem. We believe that the present research will be able to provide important implications for more endeavors. This issue is discussed in Section 6.3. 1. Experimental Design The experiment consisted of two tasks, a design task and a report task. Two practicing architects and seven advanced students in an architectural department participated. In the design task, each participant worked on designing an art museum through successive sketches for 45 minutes. They were provided with a simple diagram representing an outline of the site, in which they were supposed to arrange not only a museum building but also a sculpture garden, a pond, a green area, and a parking lot. The building was required to have entrance(s), a ticket office(s), display rooms for about 100 paintings, a cafeteria, and a gift shop. Participants were supposed to use freehand sketches as a tool for designing. They were not asked to report concurrently what was going on in their minds, nor were they interrupted by the experimenter during the design task. Their sketching activity was videotaped. Following the design task was the report task. While watching their own videotape, participants were asked to remember and report what they were thinking as they drew each portion of each sketch. In case their reports lagged behind the videotape, they were allowed to stop the tape until reporting all that they remembered about the current topics. Therefore, the duration of the report task depended on the participant, varying from one hour to one hour and fifteen minutes. Participants were not interrupted with questions during the report, except for following cases; when they obviously skipped reporting about certain portions of their sketching activity, then they were requested by the experimenter to rewind the videotape and report those portions. We recorded the participants' voices as well as videotaped the screen itself on which not only their sketching activity in the design task but also their pointing gestures in the report task showed up. 2 Information Categories In interpreting the data, the first step was to determine a set of information categories into which the contents of participants' protocols could be fit. Table 1 shows the four major categories and their subclasses. We derived the four categories from theoretical discussions and historical evidence on how external representations convey meanings and concepts, from past literature on design processes that suggest what architects generally think of in design process, and from intensive study of the protocols. Many theorists like Larkin and Simon6 and Tversky7 have suggested that the pictorial devices for expressing meanings and concepts consist of (a) depicted elements, whether objects, spaces or icons, and (b) spatial arrangement of them. They have also suggested that spatial arrangements have the ability to express not only literal spatial relations but also abstract or conceptual relations. From these discussions, it would be safe to conjecture that at least three information categories should be proposed, i.e. depicted elements, spatial relations, and abstract relations. Depicted elements are sometimes intentionally drawn and thus possess explicit shapes and sizes, but sometimes they are embedded as partial elements or implicit objects and emerge to the viewer's eyes only when he/she discovers a new way of restructuring the whole configuration that includes those elements26. Larkin and Simon6 and Koedinger and Anderson3 referred to this property of diagrams as "emergent properties". Therefore, we chose to name the first category as "emergent properties", instead of just depicted elements. "Spatial relations" hold among these depicted elements, and are inherently visual features in the sense that architects/designers can see them in their own sketches, just as emergent properties are also visual. In the domain of architectural design, abstract relations typically correspond to "functional relations". Forms and functions are the two major concepts in the domain that are conceptually distinct and yet intertwined with each other27. Functional relations in this domain denote interactions among spaces, things, people visiting or using them, and/or environments. Unlike emergent properties and spatial relations, functional relations are inherently non-visual aspects of architectural designs. Table 1: Information categories and their subclasses Ma