Design as a sequential process of description and redescription in the development of designs is presented. Rerepresentational theory (RRH) is introduced and the cognitive mechanisms which enable re-representation in design are explicated.
Design as a sequential process of description and redescription in the development of designs is presented. Rerepresentational theory (RRH) is introduced and the cognitive mechanisms which enable re-representation in design are explicated.
Design as a sequential process of description and redescription in the development of designs is presented. Rerepresentational theory (RRH) is introduced and the cognitive mechanisms which enable re-representation in design are explicated.
Rivka Oxman, Faculty of Architecture and Town Planning, Technion, Haifa, Israel The conception of design as a sequential process of description and re- description in the development of designs is presented. Re- representational theory (RRH) is introduced and the cognitive mechanisms which enable re-representation in design are explicated. A theory of re-representation and multiple representations in design is proposed. The concept of adaptation developed in CBR is exploited to formalize transformations in the re-representation process. An empirical research was constructed to study the cognitive abilities which underlie the creative phenomena of re-representation in design adaptation. On the basis of the findings of the experiment, a model of design re- representation is developed and presented. The main concepts of the model, multiple representation and re-representation, are shown to provide a powerful basis f or the understanding of creative behavior in design. 1997 Elsevier Science Ltd. Keywords: drawings, re-representation, creativity, design precedents, design process 1 SchOn, D A The Reflective Practitioner: How Professionals Think in Action I=~sic Books, New York (1983) 2 Lawson, B How Designers Think The Archite(~ral Press, London (1980) 3 Akin, O Psychology of Archi- tectural Design P~on, London (1988) ELSEVIER T he design sketch can be considered as the basis of a visual and mental transaction between the designer and the representation, which evokes a discrete graphical response. According to Schrn, the designer is engaged in ' graphical conversation with the design ' t or according to Lawson, ' t he designer has a conversation with the drawing '2. It is these transactions with the external representation which illuminate the visual-mental processes of designers. Where graphic media such as design sketches are the medi um whereby the design is evolved, design moves are ' a series of actions' of the designer which result in transform- ations of a representation 2,3. The term ' move' is frequently empl oyed in contemporary design research to describe the discrete increments of change in the evolution of a design. This elucidates one of the distinctive characteristics of the graphical design process, particularly the process of conceptual design, namely, its develop- 0142-694X/97 $17.00 Design Studies 18 (1997) 329-347 PII: S0142-694X(97)00005-7 1997 Elsevier Science Lid All rights reserved Printed in Great Britain 329 40x man, R M 'The reflective e y e : visual reasoning in design' in A Koutamenls, H T l m m e r - m a n e n d I V e r m e u l e n ( e d s ) Vis- ua/ data bases in architecture Averbury, Aldershot, U.K. (1995) pp 89-112. 5 Arnhelm, R Visual Thinking University of California Press, Berkeley and Los Angeles ( 1 9 6 9 ) ment t hr ough a sequence o f pr ogr es s i ve st ages ove r t i me. The sequence o f sket ches can act as a r ecor d o f r eas oni ng pr ocesses whi ch can be i nf er r ed f r om a transition of states f r om one r epr esent at i on t o a subsequent r ep- resent at i on. Typi cal pr oces s es can be i nt er pr et ed and cat egor i zed as cl asses and t ypes o f modi f i cat i on o f the r epr esent at i on 4. Accor di ng t o Ar nhe i m 5, cogni t i ve oper at i ons in per cept i on in desi gn i ncl ude di st i ngui shi ng st ruct ural rel at i on- ships o f t he i mages in the desi gn r epr esent at i on. Thi s i ncl udes t he i nt erpret - at i on and concept i on o f such st ruct ures o f r el at i onshi ps ( pr ot ot ypes) . Spa- tial gest al t and t he st ruct ural qual i t i es in conf i gur at i ons ar e e xt r e me l y si gni fi cant in concept ual desi gn. The st ruct uri ng and r est r uct ur i ng o f shapes whi ch ma y be obs er vabl e i n sequences o f sket ch desi gn r epr esent at i ons evi dence the cogni t i ve abi l i t i es o f t he desi gner such as t he abi l i t y t o e voke pr ot ot ypes. The wa y cr eat i ve desi gner s fit and adapt gr aphi cal r epr esent at i ons t hr ough dr awi ng and r edr awi ng i s one o f t he l east under st ood phe nome na in desi gn. Des pi t e cur r ent wor k in under st andi ng gr aphi cal r eas oni ng in desi gn, t he wa y per cept ual and concept ual mechani s ms oper at e i n desi gn dr awi ng are still unf or mul at ed. The wa y knowl e dge is e mpl oye d and sequent i al clarifi- cat i on and concept ual enr i chment ar e suppor t ed in desi gn r equi r es st udy, under st andi ng and f or mal i zat i on. The s e ar e a mong t he obj ect i ves o f our work. For mal i zi ng t he pr ocedur al aspect s o f desi gn mo v e s ma y cont r i but e t o our abi l i t y t o model vi sual r eas oni ng i n desi gn. Thi s des cr i pt i on- r e- des cr i pt i on pr ocess is one wa y in whi ch desi gner s cogni t i vel y expl oi t gr aphi cal r ep- r esent at i ons 4 and is, t herefore, r el evant t o t he model i ng o f desi gn. Sequen- tial modi f i cat i on in desi gn dr awi ng is a me d i u m t hr ough whi ch desi gn pr oces s es oper at e and are made mani f est . 1 Towards a re-representation model of visual reasoning in design drawings I n t hi s paper , we pr opos e a mode l o f the cogni t i ve mechani s ms and abi l i t i es whi ch under l i e and enabl e sequent i al st at es o f gr aphi cal r e- r epr esent at i on in desi gn. Wi t h r ef er ence t o cer t ai n r el evant ps ychol ogi cal t heor i es ci t ed bel ow, we r ef er t o this mode l as a re-representational model of design reasoning. The model i ng is f ounded upon an empi r i cal st udy o f des i gner s ' be ha vi or in t he gr aphi cal adapt at i on o f a desi gn solution. The obj ect i ve o f the exper i ment wa s t o st udy t he wa y desi gner s mani pul at e gr aphi cal s ym- bol s i n desi gn dr awi ngs, and t o i dent i f y t he knowl e dge and cogni t i ve capa- bi l i t y r equi r ed t o enabl e r e- descr i pt i on in desi gn. 330 Des i gn St udi es Vol 18 No 4 Oct ober 1997 We ha ve mode l e d the knowl e dge e mpl oye d in vi sual r eas oni ng in t he i nt er- act i on o f gr aphi c desi gn pr esent at i on. I t i s t hese t r ansact i ons in a gr aphi cal me d i u m whi ch we at t empt t o st udy, t o mode l and t o f or mal i ze. As we demons t r at e t hr ough ps ychol ogi cal studies, this l evel o f knowl e dge under - lies t he cogni t i ve abi l i t y f or st ruct uri ng, or gener al i zat i on, o f the r epr es en- tation. We have pr opos e d a r e- r epr esent at i onal model whi ch at t empt s t o i dent i f y and f or mal i ze t he l evel o f knowl e dge whi ch i s r equi r ed f or maki ng t r ans f or mat i ons o f r e- r epr esent at i on i n the i nt er act i on wi t h t he gr aphi cal medi um. As a result, we i dent i f y and char act er i ze the t r ansf or mat i onal pr o- cesses i nvol ved in i ncr ement s o f st at e t ransact i on. Fur t her mor e, we expl i - cat e t he expl i ci t sets o f knowl e dge st ruct ures whi ch enabl e t he m t o occur. Thi s exper i ment t hr ows l i ght on t he di f f er ences in t he cogni t i ve abi l i t y o f i ndi vi dual desi gners. I t al so pr ovi des i mpor t ant i nsi ght s on knowl edge and r e- r epr esent at i on as phe nome na o f creat i vi t y. Thi s wor k emphas i zes the f ol l owi ng poi nt s: (1) conceptual structures of knowledge in visual reasoning in design Gr aphi cal r eas oni ng in desi gn is vi ewed her e as a t r ansact i on bet ween concept ual knowl e dge st ruct ures and desi gn r epr esent at i on t hr ough a vi sual r epr esent at i onal me di um; (2) a re-representation theory for visual reasoning in design Ba s e d upon empi r i cal r esear ch, and in or der t o expl ai n t he cogni t i ve abi l i - t i es enabl i ng the sequent i al evol ut i on o f gr aphi cal r epr esent at i ons, we pr o- pos e a r e- r epr esent at i onal mode l o f vi sual r eas oni ng in desi gn; (3) a re-representational model in design adaptation Thi s mode l i s hi ghl y r el evant t o cogni t i ve and comput at i onal appr oaches t o desi gn such as Cas e- Bas ed Des i gn ~5. One o f the pr i or i t y r es ear ch i ssues in CBD i s the f or mul at i on o f appr oaches and pr oces s es t o t he ' a da pt a t i on' o f desi gns, or t he modi f i cat i on o f pas t desi gn r epr esent at i ons t o fit a cur r ent desi gn situation. Wi t hi n t he cont ext o f CBD, we pr opos e t he r el evance o f our r e- r epr esent at i onal mode l as a basi s f or f or mal i zi ng adapt at i on; (4) re-representational theory as a foundation of studies of design crea- tivity Model i ng t he i nt er act i on wi t h vi sual r epr esent at i on me di a pot ent i al l y con- t r i but es t o t he under st andi ng o f t he r el at i onshi p bet ween concept s, knowl - edge st ruct ures and cr eat i vi t y i n vi sual r easoni ng. The mode l o f r e- r ep- r esent at i on is, t her ef or e, hi ghl y r el evant t o t heor i es o f cr eat i vi t y i n desi gn. Des i gn b y r e- r epr esent at i on 331 6 K a r r n i l o f f - S m l t h , A, ' Con- straints o n representational c h a n g e : e v i d e n c e f r o m c hi l dr en' s drawing' Cognition Vo l 3 4 ( 1993) p p 5 7 - 8 3 , 7 B o d e n , M A The Creative Mind, Myths and Mechanisms C o x a n d Wei man, London ( 1995) 8 Da r t n a l l , T ' AI , creativity, r e p - resentational redescdption, intentionality, ment al life: a n e me r g i n g pi ct ur e' Ar t i f i ci al Intelli- g e n c e a n d Creativity, technical report, AAAI Pr es s ( 1993) 2 The theoretical relevance of the re-representation hypothesis Cognitive and computational researchers are currently developing theoreti- cal models which can assist in the analysis of sequences of representations and in the formalization of models. Particularly relevant to our research have been the psychological theories of re-representation with graphical re-representation. In modeling re-representation in design, we have exploited the Re-Representation Hypothesis (RRH), a psychological theory of creativity 6. The RRH is a cognitive approach which explains creativity as an act of ' conceptual exploration through re-representation '6-8. RRH theory has been relevant to our research in design, since it provides a plausible model of the way in which human beings explore new modifi- cations through the externalization of knowledge structures in represen- tations. The human capability to ' transform knowledge into representational struc- tures' underlies the ability to make novel modifications and changes in those representations. Researchers in RRH theory have demonstrated that representations of drawings which are procedural are not flexible for innov- ative manipulations until they are turned into explicit declarative re-rep- resentational f orm. Karmilloff-Smith 6 demonstrates this cognitive process by the example of children' s cognitive capability in drawing in the ' funny- man pictures study' . Re-representation of a form is possible onl y after the underlying representational structure of that form is externalized and made explicit in the drawing. In the case of the ' f unny- man' experiments, an externalization of the man- body- schema is prerequisite to the distortions of the schema which are ' funny-men' . In the development of explicit rep- resentations of form, humans are able to transform implicit knowledge to explicit representational structure. This appears to enable novelty through modification and change which transcend, contradict, or depart from the generic representation. We will demonstrate the direct relevance of these findings to design. Re- representation provides a changing pattern in the graphical medium. It is this pattern which is explicated in our research. I f we are to model design as a process of re-representation, we require the ability to model design through, and relative to, its declarative abstractions. Following Karmilloff- Smith, we assume that the process of transforming graphical represen- tations can be supported only according to the declarative knowledge struc- tures of design representation. In design, these representational structures are frequently domain specific. An example would be the abstract formal structure of a particular design. This formal structure must first be made explicit in order that specific re-representational operations can be achieved 332 Design Studies Vol 18 No 4 October 1997 9 Simon, H 'Keyr~ote Lecture, AID '92' First International Con- ference a t C a r n e g i e Mellon Uni- versity, Pittsburgh, Pennsyl- vania, USA (1992) on that structure. That is, in order to reformulate a design representation, an understanding of its structural content in terms of the underlying schema must be externalized. One of the contributions of our research has been to demonstrate the cognitive ability of the human designer to recognize and externalize the underlying schema and structural content of designs, and then to work relative to the declarative knowledge which has been exter- nalized. Design creativity may be seen in this sense as the ability to innov- atively re-represent the schema or the particular structural content of the externalized representation. In the medium of sequential sketching we can observe the abilities enabling change in a representational pattern which is supported by Knowledge structures. Do there exist domain specific knowledge structures? Accordi ng to Simon 9, drawings are empl oyed for the externalization and the ' hol di ng' of the various design representations which allow the designer to ' reason' with the design. Once externalized within the representation, the designer can focus upon aspects of the representation as a part of the design prob- lem. For example, measure, or spatial order, are knowledge structures in plan development in architecture. Each knowledge structure may also be associated with its own representational medium. As we have observed in our experiments in architecture, formal and compositional knowledge such as modulation can be activated by making explicit a grid system in the graphic representation, and then modifying the representation according to the explicit ordering schema. Our research has identified such underlying domain specific schema and knowledge structures, and their role in design. A further contribution of the research has been to demonstrate the signifi- cance in design of the concept of multiple representations. In design a single representation generally has embedded various underlying multiple representational structures. For example, various underlying ordering media such as grids or zonal structures, or functional orders and relation- ships are frequently implicitly present in design representations~L From our research, the human designer appears to be capable of exploiting sim- ultaneously these various underlying representational structures and of mitigating between them. This differs from the traditional RRH model in proposing that multiple representational structures are implicit in design representations. The designer may elect to make explicit the various implied schema as the design evolves. Thus, in design, the concept of multiple representation has a unique significance. In any design domain, there probably exist multiple underlying representational structures which may, in fact, be inherent in classes of design problems. What is significant in our research is the Design by re-representation 333 10 Kolodner, d L Case-based Reasoning Morgan Kaufmann (1994) 11 Hlnrlhs, T R Problem Solving in Open Worlds: A Case Study in Design Erlbaurn, Northvale, NJ (1991) demonstration that designers possess the cognitive ability to recognize, explicate and modify these structures, and to deal with these various rep- resentations simultaneously. 3 The methodological relevance of the concept of adaptation in Case-Based Reasoning Though diverse models of design relate to the developmental processes, the way design knowledge is structured and empl oyed in such processes is, as yet, poorly formulated. The related research area of computational models of design cognition provides certain relevant concepts. One of the objectives of this field is to build the theoretical foundations of design cognition through computational modeling of design processes and reason- ing. Work in the field generally exploits theoretical, experimental and com- putational research. Work in Case-Based Reasoning 1 and its application to design is directly relevant to the formulation of developmental processes in design. In the CBR view, adaptation is empl oyed as a means of fitting an old solution to a new one, or evolving a new design by modifying an existing solution representation. In this work, the concept of adaptation developed in CBR is exploited to formalize transformations in the re-rep- resentation process. The process of adaptation involves the re-use through modification (' adapt at i on' ) of the prior existing representational content of a design sol- ution. Research on case adaptation in design 11 provides a perspective in the study and the modeling of cognitive sequences in the modification of case representations. We employ the term, design adaptation, in the follow- ing experiment description in order to distinguish and identify the methodo- logical content of the requisite knowledge and formalized steps of reason- ing in a model of graphical re-representation. In our research we exploit certain concepts derived from adaptation research in CBR in order to for- mally model the sequential transformations of graphical representations in design. 4 A formal model of re-representation The theoretical assumptions developed in the first phase of the research were empl oyed to construct an empirical study of designers' behavior in adaptive tasks employing design sketches and drawings. In the research we have attempted to explicate the multiple structures of knowledge under- lying graphical representations and to demonstrate how they support modi- fication in designs. The proposed approach to adaptation is based upon the seminal concept of multiple-representation. The modeling of adaptation as a process of 334 Design Studies Vol 18 No 4 October 1997 r e- r epr esent at i on wi t hi n t he f r a me wor k o f this concept f ocus ed on a desi gn t as k o f t he a c hi e ve me nt o f uni que desi gns t hr ough adapt i ve pr ocesses. Sub- j ect s wer e as ked t o const r uct expl i ci t r epr esent at i onal st ruct ures i mpl i ci t i n a par t i cul ar desi gn f or m, in our case, an ar chi t ect ur al pl an. Adapt at i on pr oces s es i n gr aphi c r e- r epr esent at i on wer e document ed t hr ough pr ot ocol anal ysi s. Empl oyi ng t he fi ndi ngs der i ved f r om an anal ysi s o f t he pr ot ocol s, a gener al mode l o f adapt at i on o f r epr esent at i ons was devel oped. The r esul t i ng gener al mode l is consi der ed t o be val i d f or i nnovat i ve and cr e- at i ve desi gn as wel l as f or r out i ne desi gn. We descr i be t he exper i ment , its obj ect i ves, r es ear ch const r uct i on and find- ings. The fi ndi ngs wer e subsequent l y expl oi t ed in t he const r uct i on o f a gener al r e- r epr esent at i on mode l o f desi gn. 4.1 Objectives and methods The exper i ment was const r uct ed t o mode l desi gn adapt at i on in t he ar chi t ec- tural domai n whi ch is st r ongl y char act er i zed by expl oi t at i on o f gr aphi cal r epr esent at i ons. The mai n goal o f the exper i ment wa s t o obs e r ve and r ecor d how desi gner s act ual l y mani pul at e gi ven desi gn r epr esent at i ons in r e- r ep- r esent at i onal pr ocesses o f adapt at i on. Adapt at i on was def i ned as t he t as k o f t r ans f or mi ng a gi ven desi gn ( r epr esent at i on) t o me e t changi ng con- straints. The chosen t as k domai n was spat i al conf i gur at i on i n pl an desi gn. The c ompl e x nat ur e o f i nf or mat i on in t he gr aphi cal r epr esent at i on o f desi gns r ai sed sever al r esear ch issues. An addi t i onal obj ect i ve o f t he exper i ment was t o st udy the wa y desi gner s deal wi t h this r i chness o f i nf or - mat i on in gr aphi c r epr esent at i ons, and t he wa y t hey mani pul at e gr aphi cal s ymbol s as abst r act knowl e dge dur i ng t he adapt at i on o f desi gn dr awi ngs. The over al l obj ect i ve was t o pr ovi de an account o f t he r epr esent at i ons e mpl oye d, t he oper at i ons used and t he wa y t hey wer e cont r ol l ed in t he adapt at i on pr ocess. The fi ndi ngs wer e subsequent l y anal yzed and cat egor - i zed f or ki nds o f r epr esent at i onal modi f i cat i ons, t he st rat egi es and oper - at i ons e mpl oye d and the wa y t hey i nt er act in t he pr ocess. Fi nal l y, it was possi bl e t o post ul at e t he par t i cul ar st ruct ures o f r epr esent at i on whi ch ma ke gr aphi c r e- r epr esent at i on pos s i bl e in t hese t asks. Each o f t he desi gner s in the exper i ment was r equest ed t o f or mul at e and const r uct expl i ci t r epr esent at i onal st ruct ures i mpl i ci t in t he r epr esent at i on whi ch, f or hi m/ her , enabl ed t he ma ki ng o f changes. Subj ect s wer e al so asked t o r ecor d each o f t hei r gr aphi cal mani pul at i ons and rat i onal e. Thus, bot h the r epr esent at i onal st ruct ures e mpl oye d in r eas oni ng by t he desi gner s wer e i dent i fi ed and the r esul t i ng pr ocedur al mani pul at i ons wer e r ecor ded. Des i gn b y r e- r epr esent at i on 335 12 Baharat, D, Schmitt, G, Faltlngs, B end Smith, I 'Case- based design in architecture' in J Gero end F Sudweeks (eds) AID 94 Kluwer, Dordracht (1994) pp 145-162 13 Hue, K, Smith, I, Faltlngs, B, Shill, S and Schmltt, G 'Adaptation of spati al design cases' in J S Gero (ed) AI in Design "92 Kluwer, Dordracht, (1992) 14 Oxmen, R E 'Design shells: a formalism for prototype refinement in knowledge-based design systems' Artificial Intelli. gence in Engineering Vol" 519 (1990) pp 2-8. 15 Oxman, R E 'Design case- bases: graphic knowledge bases for design worksbaca' in M Tan (ed) CAAD Futures '95 Com- puter Aided Architectural Design Singapore (1995) The subjects were advanced students in architecture. They were required to adapt a given design on the basis of changed requirements. The required changes were purposely kept minimal in order to enhance understanding of the complex behavior in adaptation. The problem was to reduce the area of a given plan drawing by 20%. In such a limited problem statement students could easily explore different solution spaces and accurately rec- ord their decision-making rationale. The subjects were asked to provide self-protocols recording explicit steps, actions and reasoning. The protocols were later formally recorded in a CAD graphics system to provide a sequence of re-representations sup- plemented by annotated textual explanations. Multiple representations were made explicit in the sequence. At a later stage, the group of participants contributed in the development of a joint model. 4.2 Assumptions An underlying theoretical assumption is that the form of the graphic rep- resentation of designs and their amenability to adaptation are related. The form of representation encodes information and knowledge which support reasoning during manipulation. Therefore, in order to assist processes of reasoning in the adaptation of design drawings, and in order to be respon- sive to manipulations, graphic representations must be explicitly structured. Accordi ng to these main theoretical assumptions, the following concepts are proposed as the theoretical basis of the research. 4.2.1 Multiple representations as underlying abstraction levels One definition of adaptation is that the current graphic state of the represen- tation is manipulated until a fit is achieved with the new requirements. In order to support operations of re-representation we propose that graphic representations must be explicitly structured. The concept of multiple rep- resentations as underlying abstraction levels is the concept which we pro- pose for accommodating semantic knowledge in graphics. In the particular design task we can observe a structure of multiple underly- ing, or background, representations. These can be different kinds of generic abstractions of the design representation. Various works have begun to address the role of such abstractions in design and to demonstrate their utility in adaptation. For example, such descriptions include geometry- based descriptions of spatial aggregations, topological descriptions of spa- tial relations 12, grammar-based descriptions ~3 and typological descriptions such as housing schema ~4 etc. Other work on adaptation in CBR H has 336 Design Studies Vol 18 No 4 October 1997 poi nt ed out t he si gni fi cance and r ol e o f gener al knowl e dge o f adapt at i on pr ocedur es i n a t ask. I n this vi ew, adapt i ve knowl e dge can be r egar ded as t he abi l i t y t o mani pul at e hol i st i c desi gns whi ch under l i e the uni que a r r a nge me nt and r el at i onshi ps o f component s . The r ange o f pr oces s es o f adapt at i on i n desi gn shoul d pr es er ve and expl oi t t he si gni fi cant aspect s o f a desi gn by mai nt ai ni ng t he i nt ernal consi st ency o f its concept ual sol ut i on. I n this exper i ment we ha ve suggest ed t o de c om- pos e a desi gn dr awi ng t o its st ruct ures. The decompos i t i on o f desi gn dr aw- i ng i nt o di f f er ent st ruct ures o f knowl e dge is, in fact , decompos i t i on o f di f f er ent t ypes o f knowl edge. Des i gn dr awi ngs are usual l y not r eadi l y de c ompos a bl e i nt o separ at e r epr esent at i ons, or structures. I n or der t o i so- l at e i mpl i ci t r epr esent at i onal st ruct ures it is i mpor t ant t o i dent i f y t he expl i ci t r epr esent at i ons whi ch suppor t vi sual r easoni ng. Si nce dr awi ngs are not expl i ci t l y st ruct ured accor di ng t o t hose r epr esent at i ons, i t is si gni fi cant t o ext er nal i ze t he m i n r e- r epr esent at i on. I n or der t o assi st pr oces s es o f r eas oni ng in t he adapt at i on o f desi gns we have exper i ment ed wi t h t he expl i cat i on o f under l yi ng s chema in gr aphi c r epr esent at i ons b y r equest i ng t hat t he subj ect s ma ke t he m ( each abst r act i on l evel ) i ndi vi dual l y expl i ci t . I n the final CAD f or ma l r econst r uct i on o f t he pr oces s t hese mul t i pl e r ep- r esent at i ons wer e act ual l y pl aced on separ at e l ayers. We ha ve al so at t empt ed t o gener al i ze r egar di ng t hose st ruct ured r epr esent at i ons whi ch r ecur r ed wi t hi n t he gr oup o f subj ect s, and ma y be consi der ed r el evant as domai n knowl edge. 4. 2. 2 Internal consistency Dur i ng adapt at i on, t he desi gn mus t r emai n i nt ernal l y consi st ent i n or der t o sat i sf y var i ous const rai nt s. Const r ai nt s i n evol ut i on ma y be der i ved f r om di f f er ent sources. We ha ve i dent i fi ed t wo ki nds o f adapt at i on const rai nt s: ext er nal gi ven const rai nt s, and i nt ernal der i ved const rai nt s. - - ext er nal const rai nt s are i mpos e d f r om t he p r o b l e m defi ni t i on, and are expl i ci t l y r epr es ent ed in t he p r o b l e m st at ement . For exampl e, con- straints bet ween desi gn and desi gn envi r onment , such as t he si t e in ar chi t ect ur e 12. - - i nt ernal const r ai nt s are der i ved as a cons equence o f an ear l i er desi gn mani pul at i on. For exampl e, each change in component s and t hei r r el at i onshi ps can i mpos e new const rai nt s. I nt er nal const r ai nt s are i mpl i ci t const r ai nt s t hat are bas ed on domai n knowl edge, t hei r r ol e is t o keep t he cons i s t ency a mong t he mul t i pl e i nt ernal r epr esent at i ons. I nt er nal const r ai nt s are usual l y pr opagat ed b y ot her const r ai nt s whi ch e me r ge dur i ng t he pr ocess. The abi l i t y t o mani pul at e const r ai nt s can be r egar ded as a si gni fi cant ki nd o f adapt at i on knowl edge. The Des i gn b y r e- r epr esent at i on 337 designer must identify those reasoning processes that maintain the consistency of the internal representations. I f a design is to remain consistent, it should satisfy both external and internal constraints. It is the ability to handle consistency by the modifi- cation of structures, components and constraints at any time that charac- terizes the dynamic nature of the process of adaptation. Without controlling the impact of internal constraints between the different multiple represen- tations, internal consistency cannot be maintained. Wi t h each re-representation new internal constraints emerge. It may also be necessary to violate the constraints which are imposed by the new rep- resentation. In the experiment one goal was to understand the role of design constraints in re-representation and how the internal consistency and integ- rity of the design is maintained during its evolution. It is this ability to handle consistency by modifications under conditions of dynamic con- straint propagation that characterizes one aspect of creative design. 4.2.3 Modification strategies and operations Another research issue was to characterize modification strategies employed, or preferred by the subjects. Can these strategies provide a tax- onomy for design adaptation? We believe that it is possible to define classes of strategies and operations. These are based upon types of modifi- cation relative to the underlying representational schema and constraints. In addition to modification of structures of a design, the ability to manipulate constraints can also be regarded as a significant kind of adaptation knowl- edge. For example, adaptation may be applied to constraints by weakening preferences and priorities. We have been able to draw only limited con- clusions on this subject with our relatively small population of student designers. However, it appears to us that the cognitive mastery of graphical adaptation operations and schema modification strategies is one of the forms of design expertise. 4.2.4 The dynamics of re-representation Each new adaptation opens the possibility for further interactions and new changes. In such an approach, diversity arises when opportunities for new interactions that can be exploited are recognized by the designer as an ever-changing pattern of possibilities. 5 Re-Representing the new from the known: a case study in architectural re-representation In order to test and validate our assumptions, an empirical research was constructed. In the experiment, domain knowl edge made manifest in 338 Design Studies Vol 18 No 4 October 1997 graphical operations in a design task was studied. Representations in archi- tecture tend to be task specific. Each task has its own representational system which supports reasoning processes within the task domain. The case study that was chosen was a house plan, the complexity of which was easily manageable by the subjects. The task domain was spatial configur- ation, or the formal and functional modification of the design representation of the plan. The subjects, who were 15 advanced level architectural stu- dents, were requested to do the following assignment. Given an existing design representation (plan drawing): (1) To construct a set of underlying multiple representational abstraction levels for the given design. Each abstraction level suggested design operations associated with the type of knowledge or design principle. For example, an underlying abstraction such as an axis which is asso- ciated with symmet ry may also be associated with operative knowl- edge such as morphological transformations relative to axial order. In the case study, certain classes of abstraction common in the domain were suggested as examples, while the subjects were requested to formulate other schema or abstraction levels which were relevant to their own re-representational operations. Certain typical schema or abstraction levels in the domain are (Figure 1): underlying abstractions of order: for example: main rectangular axis, sec- ondary diagonal axis system, main grid system, etc. Fi gure I Abstra~:tion l evel s as re-represent at i onal sys- t ems in desi gn comers di agonal axi s or t hogonal axi s system system formal Myers emphasized main grid system axis Design by re-representation 339 typological schema: (typologies in housing design): for example, the knowledge of minimal and maximal dimensions of functional elements such as bedroom, bathroom, stairs etc. underlying abstractions of order in schema: for example, principles of zon- ing, functional relations such as linkages between rooms, etc. general design principles related to the task: for example, the reduction of comers, layering systems, emphasizing the main axis, etc. (2) Given an external requirement, the subject was requested to identify design constraints and to select abstraction levels for re-representation which were related to these constraints. In this case, the external requirement was to decrease the area of the plan design by 20%. The subjects identified the constraints and most decided to preserve the topological relations of the design while effecting the re-represen- tations. (3) After first re-representations were effected, the designs were evalu- ated and faults were identified. For example, the functional types may have ' failed' functionally or formal principles such as the dimensions of grid system may have been ' violated' . (4) At this stage the emerging internal constraints were identified and new re-representations were undertaken. In this way a cyclical process of changes propagated new constraints, which elicited subsequent modifications activated by the various knowledge sources. I : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : o r i g i na l d e s i g n initial state adapted d e s i g n f'mal state Figure 2 Initial and adapted final state 340 Design Studies Vol 18 No 4 October 1997 Figure 2 illustrates the initial state and the final state of this subject. The re-representation sequence of the subject' s example illustrated in this paper, included some twenty-eight recorded iterations. The first six re-rep- resentations are illustrated in Figure 3 and described below. Each re-rep- resentation generates a sequence of internal constraints which must be sat- isfied in the next re-representation and this continues in a cycle of sketches until all constraints are satisfied including the initial constraint. - - r e- r epr es ent at i on 0 t o 1: (components affect structural schema) Initial constraint is a reduction of size. The initial strategy was to start the reduction process on the holistic structure. The inner layer, one of a set of formal layers which contain functional zones (court) was reduced to a mini- mal functional size. This operation disturbed the consistency of the grid system; - - r e- r epr es ent at i on 1 t o 2: (structural operation affects component integrity) In order to reconstruct the regulation system the rest of the functional layers were shifted along the diagonal axial system. As a result some typological constraints emerged by changing the dimensions of functional spaces (entrance corridor) and architectural elements (fireplace); - - r e- r epr es ent at i on 2 t o 3: (parametric modification) The extension of a functional space (entrance corridor) was made in corre- lation with the diagonal system and axis; - - r e- r epr es ent at i on 3 t o 4: (parametric modification of component) Architectural element (fireplace) was transformed to original size in relation to central axis; - - r e- r epr es ent at i on 4 to 5: (topological and functional relationships) The enlargement of one element (fireplace) has affected relations between architectural elements (column and fireplace), and required a subsequent change in related elements; - - r e- r epr es ent at i on 5 t o 6: (formal system and parametric reduction) The columns were shifted symmetrically and the t wo adjacent spaces were reduced. Design by re-representation 341 . . . . . . ' ' ' - ' l ' . . . . . . i ni t i al s t a t e r ~ _ _ r ~ l ! ' I i:l , Lr! q no.1 " " ~" ' ! ~ . ' U ~ ' ~ ' ' 1 no. 2 I - "% ! I t - 71 ~ \ , ; . , 'm ! n o . 3 I no. 4 n o . 5 Figure 3 Re-representation on transformations (analysis done by Amos Wachman, Faculty of Architecture and TP, Technion) 342 Design Studies Vol 18 No 4 October 1997 5. 1 Findings Human designers were found to be able to extract various abstracted explicit representations of the existing design based on their own domain knowledge. These abstractions included typological schema, organizational principles, morphological principles, functional relationships, etc. Subjects were also able to externalize and categorize their graphic manipulations conceptually despite the fact that drawings tend not to provide explicit representational information to support such decomposition. Furthermore, the graphical re-representation of a design exploits this same knowledge in creating new representations. From our research it appears that human designers utilize a personal hier- archical order of underlying conceptual structures such as grids, construc- tion lines etc. These multiple schematic representations serve the designer as supportive representations for design manipulation. It is the structuring quality of these multiple representations and the explication of their indi- vidual internal constraints (the internal logic of the schema) which appear to support design manipulation. The adaptation, or modification, of representations also appears to be dependent upon generalized domain knowledge which is learned through design experience. In the particular design to be adapted, this generalized knowledge enabled the externalization/construction of the underlying schema of the design. These (various) schema constitute the multiple underlying representations of the design. According to our findings, each design must have such multiple representations, i f it is to support the manipulation of designs. This abstract knowledge may be considered an inter-related system of domain specific representational structures. This finding is most relevant to the problem of appropriate representations for designs. Another of the most interesting observations was the contingent quality of the adaptation process. It was possible to observe how each new (modified) representation of the design introduced new modification strategies and operations, and even new structuring concepts. Recognition and maintenance of the internal consistency of the system of abstractions was found to be a characteristic of designers' cognitive ability. The ability to relate to this schema as a reference and to use or depart from it is a quality of the creative designer. 6 Towards a general model of reasoning in re- representation In order to develop a general model we recorded the states of evolution of the design representations as well as the protocols of interactions Design by re.-representation 343 between the designer and the semantic representations. The model is iterat- ive and consists of the following main stages: ( 1 ) (2) External Constraints: identification of external constraints. Extemal constraints are generally derived from the problem definition or re- definition. Initial Strategy: selection of re-representation strategy, including among others, structural modification of the schema, or component re-representation within the schema. In design there is always some form of hierarchical relationship between structure and component. Thus the selection of working within the abstraction levels of the schema structure, or modi fyi ng the schema depends upon the design- er' s evaluation of the external constraint. It may also be determined by personal design preferences. It is the ability to handle consistency by the modification of structures, components and design constraints which is characteristic of re-represen- tation. The strategy must be determined by the order in which to achieve modification. A constraint may be violated by adapting a value or by mod- ifying the structure and it may also be possible to simplify a problem by relaxing a constraint. (3) Representational Abstraction Layer: selection of layer of represen- tation of schema, abstraction level, or individual component. In gen- eral, the constraints specify properties of the design that must be transformed. Such properties may be restrictions on values, or relationships between different components, and help to determine upon which representational layer the re-re-representation can be effected. (4) Operational Method: selection of operational method associated with the representational layer. The constraint(s) guide the operations on a representation during the modification process. For example, typo- logical substitution retrieves refinement methods; t opol ogy retrieves topological operations such as delete, insert and parametric oper- ations; change in a formal system retrieves its own formal represen- tational system. (5) Evaluation: evaluation of internal consistency with respect to rep- resentational integrity, for example, in the formal structure of a design. (6) Internal Constraints: identification of internal constraints which ensure internal integrity. The nature of a constraint affects the way re-representation may be performed. A re-representation strategy may also be invoked in response to the violations of the constraint. 344 Design Studies Vol 18 No 4 October 1997 (7) Selection: selection of active constraints according to priority criteria. (8) Continue cycle. Figure 4 describes some of the abstraction levels and their associated oper- ational methods. The general model is schematically illustrated by one cycle of re-representation in Figure 5. 7 Summary and conclusions: the creativity in re- representation Our work has explored the scientific means through which our own intuition about design can be transformed into the development of a theor- etical approach. We have presented the theoretical and empirical foun- dations for re-representation models of adaptive design. In the empirical study an approach to the modeling of the design reasoning underlying re- representational processes through sequential drawing and sketching of re- representations has been described. The empirical study has led to the construction of a general model of re- representation exploiting multiple representations of the abstraction levels in design. The model is based on the transformation of chunks of knowl- edge in design into explicit re-representational structures which can support the creative act of re-representation in a graphic environment. some abstraction levels and schema which support multiple representational systems t y p o l o o c a l schema bathroom bedroom corridor t o p o l o g y f o r m a l system relations 1[ size ] [ c o n c e p t I inside larger central, linear, axial outside smaller refmement specialization gener'ali7~ion " o n l l i n s e r t i o n ] I o p e r a t i o n s scaling operations and parametric adjustment formal syst em symmetry operations Figure 4 Abstraction levels and their associated operational methods Design by re--representation 345 s e l e c t i n i t i a l c o n s t r a i n ! s e l e c t i n i t i a l s t r a t e g y [ s c h e m a [ I c o m p o n e n t ] assign priodty Figure 5 General model of re-representation select first representational abstraction level t ypol ogy [ t opol ogy ] I formal ] s e l e c t a s s o c i a t e d operation specialization/ generalization I i n s mi o n / ddi f i on I parametric [ evaluate c o n s t r a i n t s The sequent i al mani pul at i ons o f graphical represent at i ons and t he ability t o re-descri be must be consi der ed one model o f t he creat i ve act. Invent i on in desi gn is, among ot her cogni t i ve capabilities, consi der ed t o be i nfl uenced by t he behavi oral mast er y o f this process. Mani pul at i on o f represent at i ons t hrough sequences o f mul t i pl e re-represent at i ons is a keyst one o f t he cre- at i ve process whi ch can resul t in desi gns o f unexpect ed di versi t y and nov- el t y. It is par t o f t he par adox o f creat i vi t y that t hrough re-represent at i on, t he new can be made f r om t he existing. Our re-represent at i onal model o f re-represent at i on in desi gn appears t o pro- vi de one basis f or a f or mal t heor y o f creat i vi t y. The f ol l owi ng concept s, whi ch have emer ged f r om our empi ri cal studies, are r el evant t o t he f or mal model i ng o f creat i vi t y. - - t he human ability t o t r ansf or m i mpl i ci t knowl edge t o represen- t at i onal st ruct ure enabl es modi fi cat i on and change; - - t he capabi l i t y t o t r ansf or m knowl edge i nt o represent at i onal struc- tures underl i es t he ability t o make novel modi fi cat i ons and changes within, or t hrough, t hose represent at i ons; - - t he est abl i shment o f the ri ght represent at i on may be consi der ed t o be a creat i ve act; Creat i vi t y can be expl ai ned in such an approach by demonst rat i ng how desi gns can be accessed and t r ansf or med i n novel ways. Fur t her mor e, cer- 346 Desi gn St udi es Vol 18 No 4 Oct ober 1997 tain internal constraints of a re-representation in our model may necessitate the accessing of a new layer in another relevant design. This is one of many forms of ' explorative processes' which are common in creativity in design. These phenomena can be formally modeled within a multi-rep- resentational scheme of design as re-representation strategy. Given that graphic re-representation in design is such an important and all pervasive phenomenon, we require more empirical and theoretical work in order to develop the basic concepts such as a t axonomy of types of knowl- edge structures, constraints or classes of strategies and modification oper- ations in design. Lacking this level of basic concepts, even the description and analysis of the design protocols has been a difficult task. It is antici- pated that these results will lead to a better understanding of human design- ers and ultimately strengthen our ability to development tools and methods to support them. Acknowledgments This research was supported by the Technion VPR Fund. I would like to express my appreciation of the stimulating work of the students who participated in the graduate course in ' AI in Desi gn' . I would like to parti- cularly thank Architect Amos Wachman who provided his work to illus- trate our empirical study in this paper. The assistance of R. Sharvit and H. Casakin is also acknowledged. Design by re-representation 347