2.1. Representation

A representation is a physical or mental construct that stands for some other physical or mental construct. Analyses of the concept of representation suggest that there are four necessary parts to a mental representation: the physical or mental construct serving as the representation, the domain being represented, rules (usually implicit) that map parts of the representation onto the item represented, and a set of processes that makes use of the information in the representation (Markman, Reference Markman1999). Understanding the design process requires understanding both the internal mental representations of designers as well as the external representations (e.g., sketches, function, and flow basis diagrams) that are used during the design process.

The study of mental representations makes clear that people represent relationships among items, and that these relationships play an important role in analogical reasoning. Theories of analogy often posit that mental representations have a structure akin to that of predicate–argument structures used in logic, artificial intelligence, and linguistics (Gentner, Reference Gentner1983; Holyoak & Thagard, Reference Holyoak and Thagard1989). Using this representational notation, a predicate is a statement that is asserted of a subject or subjects, and arguments are the subjects of which predicates are asserted. For example, Brown(x) is a predicate capable of representing the property that some object x is brown. The variable x serves as an argument to this predicate and delimits the scope of the predicate. Thus, the proposition Brown(boot) is a statement that has the gloss “The boot is brown.”

By convention, a predicate (like Brown[boot]) that takes one argument is called an attribute. Attributes are typically used to describe objects in a domain. Predicates that take two or more arguments are called relations. For example, Larger_than(x, y) takes two arguments and represents the relation that some object x is larger than some other object y. This distinction is important, because analogies typically involve similarities between two domains in the set of relations that describe them (see Falkenhainer et al., Reference Falkenhainer, Forbus and Gentner1989). We discuss analogical reasoning in more detail below.

2.2. Cognitive memory representation

Cognitive models of memory propose that there are many different modalities of representation that play an important role in cognitive processing. One distinction of interest is between perceptual (i.e., nonverbal) representations and verbal representations (Loftus & Kallman, Reference Loftus and Kallman1979; Barrlett et al., Reference Barrlett, Till and Leavy1980; Paivio, Reference Paivio1986). The distinction between perceptual and verbal representation is supported by findings such as the verbal overshadowing effect in which talking about perceptual information can interfere with the later retrieval of that information from memory (Schooler et al., Reference Schooler, Fiore, Brandimonte and Medin1997). One implication of these kinds of verbal overshadowing effects is that verbal idea generation techniques may suppress or interfere with perceptual information in memory that may be the source of important analogies. Thus, sketching techniques may be particularly useful for supporting the retrieval of perceptual information. Finally, although perceptual and verbal representations appear to be psychologically distinct, there is good reason to believe that there are relational structures of the sort described in Section 2.1 in both perceptual and verbal modalities (Barsalou, Reference Barsalou1999).

2.3. Cognitive process model for design by analogy

We know that analogies are important in the design process, because designers frequently report using analogies when generating novel solutions to design problems (Basalla, Reference Basalla1988; Dunbar, Reference Dunbar, Ward, Smith and Vaid1997; Christensen & Schunn, Reference Christensen and Schunn2007). Thus, it is important to describe what is known about analogical reasoning processes in more detail. The consensus view of analogical reasoning in cognitive science is that analogy involves the mapping of relational knowledge from one situation to another (Gentner, Reference Gentner1983; Falkenhainer et al., Reference Falkenhainer, Forbus and Gentner1989; Holyoak & Thagard, Reference Holyoak and Thagard1989; Chiu, Reference Chiu2003). The problem domain is typically called the target of the analogy. A domain of prior knowledge that provides a potential solution to the problem is called the base of the analogy. Research on analogy suggests that people first find a mapping between the relations in the base and the target. On the basis of this mapping, aspects of the target may be rerepresented to make them more similar to the base. Furthermore, inferences about the target (such as potential solutions) may be made based on the similarity of the target to the base. The potential for creative problem solving is clearest when the two domains being compared are very different on the surface, although the same process of comparison can also be used for domains that share significant surface similarity (Gentner & Markman, Reference Gentner and Markman1997).

Research has been carried out in the field of psychology to understand the cognitive processes people use to create and understand analogies (Falkenhainer et al., Reference Falkenhainer, Forbus and Gentner1989; Gentner & Markman, Reference Gentner and Markman1997; Hummel & Holyoak, Reference Hummel and Holyoak1997; Gentner, Holyoak & Kokinov, Reference Gentner, Holyoak and Kokinov2001; Blanchette & Dunbar, Reference Blanchette and Dunbar2001). Figure 2 shows the basic process steps involved in reasoning by analogy, the most cognitively challenging step, and the design methods that are available to support each step.

Fig. 2

The steps in human reasoning by analogy and the current methods available to support those processes. [A color version of this figure can be viewed online at www.journals.cambridge.org]

Analogy has traditionally been viewed as a comparison between two items in which their relational, or causal structure, but not the superficial attributes, match (Gentner, Reference Gentner1983; Gentner & Markman, Reference Gentner and Markman1997). For example, an airplane wing and a hydrofoil can be viewed as analogous because both generate lift using flow over their surfaces. The fact that airplane wings involve air flow and hydrofoils involve water flow does not affect the analogy (nor does other potential surface detail such as the colors that the items are painted).

In the psychology literature, there has been a great deal of interest in the roles of analogy and expertise in problem solving when working with undergraduate students who have no specialized domain knowledge. A classical finding is that analogies are helpful in solving insight problems, but they are difficult to retrieve from memory (Gick & Holyoak, Reference Gick and Holyoak1980). Conversely, naturalistic research with experts typically finds that analogies are often used (e.g., Dunbar, Reference Dunbar, Ward, Smith and Vaid1997; Casakin & Goldschmidt, Reference Casakin and Goldschmidt1999; Leclercq & Heylighen, Reference Leclercq, Heylighen and Gero2002). This dichotomy may reflect that experts can see the deeper, logical structure of situations, whereas those without domain expertise are mainly aware of only the superficial features (cf. Chi et al., Reference Chi, Feltovich and Glaser1981; Gentner & Landers, Reference Gentner and Landers1985; Novick, Reference Novick1988).

To clarify and more fundamentally understand these issues, laboratory research, which affords good experimental control, needs to be conducted with burgeoning domain experts. Such individuals are capable of recognizing the causal structure of products, but may also be distracted by superficial features. These characteristics make them an appropriate test bed for determining the role of base representation in analogical reminding. Moreover, it has been suggested that implicit processes could mediate analogical problem solving (Schunn & Dunbar, Reference Schunn and Dunbar1996). That is, problem solving may occur based on analogy even when the problem solver is not aware that the analogy is being used. Therefore, in studies of analogical reasoning, it is important to look separately at when a solution based on a prior analogy is found and when an individual because aware of the analogy between two domains.

2.4. Retrieving analogies

Thus far we have discussed cognitive processes that allow a problem domain to be augmented by analogy to some base domain. A central problem in developing innovative solutions to problems, however, is that domains that are analogous to the problem are difficult to retrieve, particularly before the designer recognizes that the base domain is relevant for solving a problem.

The core principle of human memory retrieval is encoding specificity (Tulving & Thompson, Reference Tulving and Thomson1973). In essence, this principle states that a memory will be retrieved to the extent that the context at retrieval is similar to the context at encoding. The context consists of the representation of information at the time of retrieval as well as other factors like emotional state and physical location. Much research in cognitive psychology suggests that people tend to retrieve information based on attribute similarities between domains (e.g., Holyoak & Koh, Reference Holyoak and Koh1987; Gentner et al., Reference Gentner, Rattermann and Forbus1993; Catrambone, Reference Catrambone2002). Good analogies are ones that have primarily relational similarities. Paradoxically, then, people find good analogies useful, but they have difficulty retrieving them when they need them. On the encoding specificity view, this difficulty in retrieving analogies occurs, because people are typically focused on the specific situation they are in at the time of encoding. That is, representations of specific situations have a lot of attribute information in them. Consequently, they tend to be reminded of those situations only in new contexts that also share those attributes (see Forbus et al., Reference Forbus, Gentner and Law1995, for a computational model of analogical retrieval).

What would this view of analogical retrieval suggest if we wanted to improve people's ability to retrieve known situations that could be used to solve a new problem? One clear prediction is that, for any given target domain, a relationally similar base domain is more likely to be retrieved if it has few attributes than if it has many, because those attributes can only interfere with relational retrieval. In addition, this view predicts that a base domain will generally be easier to retrieve when it is represented using general relational terms (e.g., fill or travel) than when it is represented using specific relational terms (e.g., inflate or walk). When a domain is represented using specific relational terms, it will only be similar to other domains that also use related relational terms. In contrast, a domain that is represented using general relational terms, will be similar to problems expressed with a wider variety of more specific relational terms. For example, a domain represented using the relation walk will only be similar to domains that use some kind of locomotion, but a domain represented using the more general relation move will also be similar to relations like drive or fly.

It is less clear how design by analogy should be affected by the specificity of the problem representation. On the one hand, a general representation of a problem will minimize the attributes in the description and will create a description focused on relations. On the other hand, a problem domain does not contain any relations that are part of the solution to the problem (otherwise it would not be a problem). Thus, it may actually be better to have a specific representation of the problem being solved, because this representation will contain much of the detail that will be necessary for constraining the solution to the problem. The study we present here will examine the influence of the level of specificity of the base and problems domains on the retrieval and use of analogies.

2.5. Formal design by analogy methods

A few formal methods have been developed to support design by analogy such as Synectics, French's work on inspiration from nature (French, Reference French1988, Reference French1996), biomimetic concept generation and analogous design through the usage of the function and flow basis. Synectics is a group idea generation method that uses four types of analogies to solve problems: personal (be the problem), direct (functional or natural), symbolic, and fantasy (Gordon, 1961). Synectics gives little guidance on finding successful analogies. Other methods also base analogies on the natural world. French (Reference French1988, Reference French1996), highlights the powerful examples nature provides for design. Biomimetic concept generation provides a systematic tool to index biological phenomena (Hacco & Shu, Reference Hacco and Shu2002; Tinsley et al., Reference Tinsley, Midha, Nagel, McAdams, Stone and Shu2007; Vakili et al., Reference Vakili, Chiu, Shu, McAdams and Stone2007). From the functional requirements of the problem, key words are derived. The key words are then referenced to an introductory college textbook and relevant entries can be further researched.

Analogous concepts can be identified by creating abstracted functional models of concepts and comparing the similarities between their functionalities. Analogous and nonobvious products can be explored using the functional and flow basis (McAdams & Wood, Reference McAdams and Wood2000). A case study of a pickup winder for an electric guitar developed using this approach is shown in Figure 3. A guitar pickup is an electromagnetic device with thousands of small-gauge wire windings used to electrically transmit the vibration from the strings. Obvious analogies for the pickup winder include a fishing reel and a bobbin winder on a sewing machine. In addition to the obvious analogies, the abstracted functional model for the pickup winder identifies the similarity to the vegetable peeler. The analogy to a vegetable peeler leads to an innovative design (prototype shown in Fig. 3). Developing a systematic approach to search for and evaluating the utility of functionally similar concepts is critical to the successful implementation of design by analogy as is enhancing natural human capability.

Fig. 3

An innovative analogy that was discovered based on a functional model and using the representation of the function and flow basis. [A color version of this figure can be viewed online at www.journals.cambridge.org]

Other design by analogy methods have been recently developed, including both electronic tools and sketching-based approaches. A representative example of such recent tools is the work by Chakrabarti et al. (Reference Chakrabarti, Sarkar, Leelavathamma and Nataraju2005a, Reference Chakrabarti, Sarkar, Leelavathamma and Nataraju2005b). In this case, an automated tool exists to provide inspiration to designers as part of ideation process. Chakrabarti has tested the automation tool and its analogy representations with student participants as part of university design courses.

2.6. Previous research on design by analogy

Human-based design methods require a deep understanding of the processes people use and the areas in which guidance or assistance could improve the process. This knowledge is gained largely through experimental work. Even though design by analogy is a well-recognized method for design, few human experiments have been done that focus on the role of analogy in design. Important work in this line has been done by Casakin and Goldschmidt (Reference Casakin and Goldschmidt1999), Ball et al. (Reference Ball, Ormerod and Morley2004), Kolodner (Reference Kolodner1997), and Kryssanov et al. (Reference Kryssanov, Tamaki and Kitamura2001). Casakin and Goldschmidt (Reference Casakin and Goldschmidt1999) found that visual analogies can improve design problem solving by both novice and expert architects. Visual analogy had a greater impact for novices compared to experts. Ball et al. (Reference Ball, Ormerod and Morley2004) investigated the spontaneous use of analogy with engineers. They found that experts use significantly more analogies than novices do. The type of analogies used by experts was significantly different from the type used by novices. Novices tended to use more case-driven analogies (analogies where a specific concrete example was used to develop a new solution) rather than schema-driven analogies (more general design solution derived from a number of examples). This difference likely reflects that novices have more difficulty retrieving relevant information when needed and have more difficulty mapping concepts from disparate domains because of a lack of experience (Kolodner, Reference Kolodner1997).

A structured design by analogy methodology would be useful for minimizing the effects of experience and for enhancing experts' abilities. The cognitive analogical process is based on the representation and processing of information, and therefore can be implemented systematically given appropriate conceptual representations and information processing tools (Goldschmidt & Weil, Reference Goldschmidt and Weil1998; Kryssanov et al., Reference Kryssanov, Tamaki and Kitamura2001).

Prior research in analogical reasoning found the encoded representation of a source analogy (the analogous product) can ease retrieval if it is entered into memory in such a way that the key relationships apply in both the source and target problem domains (Clement, Reference Clement1994; Clement et al., Reference Clement, Mawby and Giles1994). This work shows that the internal representations in memory play a key role in retrieval. The analogies and problems used in these experiments were not specific to any domain of expertise and used fantasy problems relying only on linguistic descriptions.

Little work has been carried out based on a strong psychological understanding of analogical reasoning combined with the design knowledge of analogies for high-quality designs. This paper takes a distinctive interdisciplinary route to combine these threads of research to develop a more complete understanding of the use of analogy in engineering design and to provide the basis for formal method development. Designers rely on both internal mental representations and numerous external representations ranging from sketches to specialized diagrams such as black box models. The use of various representations and modalities in the design process warrants further understanding. The following experiment further investigates visual and semantic representation effects on design by analogy, and lead to a deeper understanding of how to enhance the design by analogy process.

3.1. Overview of the experiment

This experiment controls the way in which a designer learns about an analogous product (represents it in memory) and also how a design problem is stated. This setup allows the effects of representation in memory and of the design problem to be observed. Throughout the experiment, participants used the modality of sketching and words to both reason and document their ideas. These participants were made up of senior-level mechanical engineering students. These students ranged in age from early 20s to early 30s, and experience level from minimal industrial experience, to internship and coop experiences, to multiple years of experience obtained before returning for a higher education degree.

The choice of participants is appropriate for this study for a number of reasons. A key characteristic of the experiment concerns the use of domain knowledge for multimodal reasoning with different types of representations. The choice of experimental subjects clearly meets this characteristic. In addition, the use of college student participants allowed us to gather a sample of engineers with a range of demographic backgrounds without being affected by the scheduling constraints involved in running engineers from industry. Finally, our chosen participant group provides the opportunity to explore the effect of ideation methods as part of a higher education curriculum.

The experiment consists of two tasks: memorize the analogous products and solve the design problems with a week in between for most participants. Normally, when faced with a design problem, a useful analogous product has not been seen immediately beforehand, but the analogous product is stored in a person's long-term memory. A week was chosen as a relevant time period for the experiment because any analogies retrieved will clearly be taken from long-term memory. This time frame has been used in previous experiments (Thompson et al., Reference Thompson, Gentner and Loewenstein2000). Results from the first task were matched to the second task. Participants were senior mechanical engineers with instruction in design methodology including idea generation. Multiple solutions were encouraged for all phases. Participants were told that the experiment evaluated various skills used in the design process. The effects of the design problem and the analogous product representation were evaluated. A 2 × 2 factorial experiment design was employed which resulted in four different experimental groups (Table 1). For both the analogous product and the problem description, two levels of participants were compared, a domain specific description group and a general description group. In each task, participants received linguistic representations using either very domain specific wording or in more general terms (Table 2).

Table 1.

Overview of the factorial experiment design

Table 2.

An example of the domain-specific and general device descriptions given to participants for task 1

Sentences are general (G) or domain specific (D).

Table 3.

Domain-specific and general problem statements

3.2. Procedure

For the first task, memorize the analogous products, participants were given five short functional–textual descriptions of products along with a picture (Fig. 4) and were asked to spend 30 min memorizing the descriptions. The products were functionally described in a few short sentences either with a more general description that applied in both the source analogy and target design problem domains or with a domain-specific description. An example of the descriptions used for the flour duster device is shown in Table 3. The product descriptions and the design problems included meaningful pictures. The semantic descriptions of the devices were varied, but the pictures were identical for both conditions. The focus of these experiments was on the linguistic representations of the devices, but visual information was also present.

Fig. 4

Analogous products and solutions based on the analogies. [A color version of this figure can be viewed online at www.journals.cambridge.org]

Both groups were then given up to 15 min to answer a quiz, requiring them to write out the memorized descriptions. Finally, the groups spent up to 10 min to evaluate their results. Two of the products acted as source analogies for the design problems in the second task, solve the design problems, and three were distracter products that shared surface similarities with the design problems. An example of the descriptions used for the air mattress is shown in Table 2.

All time limitations throughout this experiment were based on a pilot experiment with graduate students in mechanical engineering in which they were given no time limits. Time limits were set to be longer than the amount of time required by most participants in the pilot experiment. For certain tasks and phases, it was clear participants were not spending enough time on the task, so the time limits were actually extended well beyond the time required for the participants in the pilot experiment.

In the second task, solve the design problems, participants were given two design problems to solve. Each design problem was staged in the following seven phases:

• phase 1: open-ended design problems, few constraints

• phase 2: highly constrained design problems

• phase 3: identify analogies and try using analogies

• phase 4: continue using analogies

• phase 5: try to use a function structure to help you find a solution (Fig. 5)

• phase 6: informed task 1 products are analogous

• phase 7: target analogous product is given

Fig. 5

The functional model for design problem 2: flour sifter.

Phases 1 and 2 were completed for the two design problems followed by phases 3–6. Throughout all phases, participants were given the general idea generation guidelines to generate as many solutions as possible with a high quality and large variety and to write down everything even if it did not meet the constraints of the problem including technically infeasible and radical ideas. Participants were also instructed to use words and/or sketches to describe their ideas. They were asked not to discuss the experiments with their classmates until all the experiments were completed.

In phase 1 the problems were initially presented with few constraints. Participants were given 11 min to generate ideas for the open-ended design problems and then given an additional 11 min to create more solutions to the same problem with additional constraints. The additional constraints limited the design space, thus increasing the chance the participants would retrieve the desired source analogy. Next they had a 5-min break.

In phase 3 participants spent 10 min listing any analogies they had used and also using analogies to develop additional solutions. An open question from one of our prior experiments (Linsey et al., Reference Linsey, Murphy, Wood, Markman and Kurtoglu2006) was whether participants would be more likely to find the source analogy from task 1 if they were given more time to use analogies. Therefore, following the initial phase using analogies, participants were given an additional 10 min to continue to use analogies to create solutions.

Next, participants were shown a series of six function structures and asked to develop more solutions to the constrained design problem. This phase provided a foundation for evaluating the effectiveness of function structures for generating novel design solutions. Function structures are representations used in engineering design (Stone & Wood, Reference Stone and Wood2000; Otto & Wood, Reference Otto and Wood2001; Kurfman et al., Reference Kurfman, Stock, Stone, Rajan and Wood2003; Hirtz et al., Reference Hirtz, Stone, McAdams, Szykman and Wood2002). They are a particular form of functional representations, where a number of such representations have been studied as part of the design process (Chandrasekaran et al., Reference Chandrasekaran, Goel and Iwasaki1993; Qian & Gero, Reference Qian and Gero1996; Goel, Reference Goel1997; Umeda & Tomiyama, Reference Umeda and Tomiyama1997; Balazs & Brown, Reference Balazs, Brown, Cugini and Wozny1998, Reference Balazs and Brown2002; Kitamura et al., Reference Kitamura, Sano, Namba and Mizoguchi2002; Gero & Kannengiesser, Reference Gero and Kannengiesser2003; Chandrasekaran, Reference Chandrasekaran2005; Stone & Chakrabarti, Reference Stone and Chakrabarti2005). When function structures are created for novel design problems, process choices must be made. Process choices include using human energy to actuate the device as opposed to a battery and electric motor or a gasoline engine. The process choices for the function structures were made to be consistent with the solution based on the analogous product and were expected to improve participants' ability to generate a solution. This phase of the experiment addresses whether an appropriate functional representation will assist participants in solving a difficult design problem. This experiment does not address how these particular functional representations with appropriate process choices can be developed by participants.

In phase 6 the participants were told that products from the first task were analogous, and were asked to mark their solutions that used the analogy and to generate additional solutions using analogies. Finally, participants were given the target analogy for each problem, and were asked to place a check where they had used it and to generate more ideas if they had not used the described analogy. These final two phases serve as a control to verify that the analogies being used are sensible, are useful for these particular design problems, and facilitate data evaluation. At each phase the participants used a different color of pen, which made it easier for the experimenters to identify the phases of the study at which information was added. A short survey at the conclusion of the experiment evaluated English language skills, work experience, if the participant had heard about the experiment ahead of time, functional modeling experience, if they felt they had enough time, and prior exposure to the design problem solutions. During one of session of task 2, a fire alarm occurred during phase 2. This caused a break in the middle of the experiment. The data were reviewed, and little impact was observed. These four participants are included in the results. The entire experiment required about 2 h.

3.3. Metrics for evaluation

Each analogy produces a set of solutions, not a single solution. Participants also created a large number of solutions that were not based on the analogies provided. We were primarily interested in the phase of the study at which participants produced a solution to the constrained design problem based on the targeted analogy and also the phase at which they identified the analogy that they used. As we will see, people often show evidence of being influenced by an analogous product without explicitly recognizing where the idea came from. Two evaluators judged the data independently, recording when the analogous solution was found. Initial agreement was approximately 80% across the experiments, and disagreements were readily resolved through discussion. The most common reason for the initial differences was the participant referenced solutions that appeared on different pages of the results.

6.1. Research question 1

General linguistic representations, which apply both in the analogous product and design problem domain, increase the success rate more than domain-specific representations. General linguistic representations are more likely to be retrieved from memory. If a designer retrieves analogous products from memory with more general representations, then they are more likely to later use these analogies to solve novel design problems (Figs. 6a and 7a). This result has very important implications for the way we should teach designers to think about and remember design solutions they encounter. If they seek representations that apply across more domains and in more general forms, they will be much more likely to be able to use the design in the future. For example, framing an air mattress as “a device that uses a substance from the environment it is used in,” rather than “a device that is filled with air” makes it much more likely to be used in future design problems that seek innovative solutions.

6.2. Research question 2

The representation of design problems clearly influences a designer's ability to generate analogous solutions (Figs. 6a and 7a). The representation that will give the designer the highest probability of exposing or generating an analogous solution depends on how the analogous solution is stored in memory. This experiment evaluated cross-domain analogies; the products and the design problems were not in the same domain. Retrieving solutions to a design problem within a domain is much easier than cross-domain analogies but results in less novel solutions presumably because both the product and design problem are represented in the same domain specific form. For the case of cross-domain analogies, if the analogous product is stored in a general form, then a domain specific representation is the most efficient means to retrieve it. For products that are committed to memory in more domain-specific terms, it is unclear what representation is best. Generally, it is not known in advance what representation is most likely to retrieve the desired information. This means that the best approach for seeking analogous solutions is to use multiple representations that vary across the range of domain specific to domain general.

6.3. Research question 3

There is a clear increase in the number of participants who found a solution based on the analogy during phase 5, when participants used the function structures to assist in generating solutions. This result is exciting and a validation of anecdotal claims about an important role of functional modeling in design. Function structures are another potential representation that will enhance the design process and should be included in the search for analogous solutions. It is important for us to point out, however, that participants were given function structures with process choices that were consistent with the analogous solutions we hoped that they would find. These function structure also included linguistic functional descriptions that were different from the given problem statements. This experiment does not address the way participants would go about developing these particular function structures on their own. Instead, it suggests that if designers create an appropriate function structure, it will increase the likelihood that they will generate the analogous solution. Further research must explore the kinds of function structures that designers generate spontaneously and the influence of these function structures on the analogies retrieved. However, clear implications from this work is that functional representations are important, and in turn, verb constructs (active functions) from the English language should be exploited to assist in the retrieval or search for analogies.

7.1. Analogy identification and implications for naturalistic analogy research and evaluation of automated tools that provide analogous solutions

Designers frequently use analogies to solve design problems without realizing the source of the idea. The participants used analogies to solve the design problems, but did not mention that they were using analogies and/or did not realize that their solutions were analogous to previously experienced products until a later phase (Figs. 6 and 7). Instructing subjects to use analogies and list the analogies they had used caused little effect. Our findings replicate the work of Schunn and Dunbar (Reference Schunn and Dunbar1996), except for an independent data set and in the engineering domain. Schunn and Dunbar found that participants often used analogies to solve difficult insight problems, but the subjects did not realize they were doing this. One implication of this result is that analogies play an important role in problem solving, but they do so, at least in part, outside of awareness. Another implication is that, in naturalistic observation studies or when evaluating an automatic design tool that facilitates analogies, simply recording how often people say they are basing their solutions on analogies is likely to underestimate their true frequency. For example, imagine an investigator who seeks to determine how important analogies are in generating new designs. This researcher decides to observe expert designers at their workplace generating novel designs and counts the number of times the experts say “this is just like (some other product).” Intuitively, this procedure seems reasonable, but our data suggest that it will underestimate the role of analogies. These results also indicate that designers frequently use analogy without recognizing it. This implies that design by analogy has an even greater impact on the design process than what is currently indicated by the anecdotal evidence.

7.2. Implications for automated or semiautomated design tools

Automatic tools have great potential to support and enhance conceptual design and design by analogy. Designers need more tools that assist in searching and retrieving analogous design solutions, especially far-field solutions. Some tools have been and are currently being developed to assist designers in finding analogies. Chakrabarti et al. (Reference Chakrabarti, Sarkar, Leelavathamma and Nataraju2005a, Reference Chakrabarti, Sarkar, Leelavathamma and Nataraju2005b) have created a tool that searches a biological database and retrieves possible solutions. Hacco and Shu (Reference Hacco and Shu2002) created a tool that cross-references a functional description in engineering terms to the related biological phenomena thereby retrieving possible solutions. Computational tools need to be able to search other representations (shape, form, dynamic motion, etc.) other than linguistic (Yaner & Goel, Reference Yaner, Goel, Barker-Plummer, Cox and Swoboda2006a, Reference Yaner, Goel and Gero2006b, Reference Yaner and Goel2007). Computational tools can also support engineering design by creating multiple function structures with different processes choices. It would be useful for automatic tools to a transition from one representation (functional model to problem statement) and to present information in multiple representations.