3.1. Rule base for mobile phones

The framework presented in the last section was tested with an example of a rule base for mobile phone designs. A set of Nokia phone designs, selected on the basis of stylistic and typological similarities, was used for the creation of rules. The “candy bar” phone with an alphanumeric keypad was considered apt for this study because it has a number of design elements on its face. Current smartphone designs, in contrast, are dominated by a large screen and do not have such an interplay of design elements.

Fig. 3.

Experiment with designers.

The design of the selected type of mobile phone was decomposed into primary elements and detail elements. The primary elements were the principal design elements that impacted the overall appearance of the design. These were the body contour, body frame, display screen, and keypad. The remaining elements were classified as “detail elements.” Consequently, rules that governed the placement and location of primary design elements were marked as “primary” (rule sets A, B, C, and G), whereas the remaining rule sets were classified as “detail” rule sets (Fig. 4).

Fig. 4.

Components of mobile phone design.

The rule base had 142 rules, organized into 15 function-based rule sets. Three initial shapes defining the body contour of the mobile phone were given in the rule base (Fig. 5). The large number of rules as well as the additional hypothetical rules ensured that there was diversity in the rule base. For instance, rules had descriptors that ranged from “very balanced” to “very unbalanced” and “very basic” to “very derived.” The rule base could therefore be used for assembling grammars that generate extant designs, as well as novel designs. Figure 6 shows examples from each rule set. The complete rule base is detailed in our study (Ahmad, Reference Ahmad2009).

Fig. 5.

Initial shapes with adjectival descriptors.

Fig. 6.

Examples of rules with description (a) rule sets A-H and (b) rule sets I-O. Adjectival description of the right-hand side of the rule is detailed.

3.2. Style description scheme

A small study was conducted to select the adjectives appropriate for the descriptions of primitives, spatial relations, and designs. A set of 30 adjectival pairs was identified from the literature (Dondis, Reference Dondis1975; Holgate, Reference Holgate1992; Ching, Reference Ching1996; Chen & Owen, Reference Chen and Owen1997; Ngo et al., Reference Ngo, Samsudin and Abdullah2000). Based on the authors’ perception and judgement, 10 more adjectival pairs were added. These 40 adjectival pairs were narrowed down in an informal study conducted with three domain experts, selected as a convenience sample. The study involved showing the experts samples of primitives, spatial relations, and designs, and asking them to select the most appropriate adjectives for description. Based on their input, a set of 12 descriptors was selected (Appendix A, Tables A.1, A.2, and A.3).

A 5-rank scale corresponding to numeric values ranging from –2 to 2 specified primitives and spatial relations. Logical conditions based on the geometric properties of the primitives and spatial relations were utilized to assign adjectival descriptor ranks (Appendix A, Tables A.1, A.2). For example, properties such as relationships between corresponding segments, axes, and centroids of constituent primitives were used to assign adjectival descriptor ranks to spatial relations. Bounding boxes of constituent primitives were used to determine proportions. For instance, if the primary axes of two primitives were coincident and the centroids of both primitives were coaxial, then the spatial relation was ranked as “very axial.” If the corresponding axes of the two primitives were inclined to each other, the spatial relation was ranked as “very nonaxial.” The middle rank, “partly axial and partly nonaxial,” was used for the description of spatial relations that were ambiguous and had both these opposing qualities present in them.

The computation of design descriptors was based on rules present in design derivations, and their primitive and spatial relation descriptor ranks (Appendix A, Table A.3). Because these descriptors describe designs that were computed from grammars, they were more complex to predict. Hence, only a 3-rank scale corresponding to numeric values ranging from –1 to 1 was used for specifying design descriptor pairs. An additional descriptor “dominance” was introduced to describe the most frequently occurring primitive or spatial relation in the design. Adjectival descriptors were computed for all primitives and spatial relations and were added to the rules in the rule base (Fig. 6a, b).

3.3. Original grammar

We assembled an original grammar “O” from the rule base. The grammar generated existing Nokia designs. It had 8 primary rules and 35 detail rules, making a total of 43 rules. Initial shape 1 was selected for the grammar.

A typical design derivation involved the creation of the layout first with the application of primary rules from the rule sets A, B, and C. The design was then refined by adding details and substituting markers with terminal primitive elements (Fig. 7). Four designs were generated from grammar O, and their design descriptors were computed (Table 1) based on the ranking conditions given in Appendix A, Table A.3. Using the descriptor ranks that are common to designs, a number of styles could be defined in the design language of a grammar. Based on Eq. (1), we defined a style S ₁ as

$${S_1} = \{ {D_{x\comma\,y}}:x \supseteq {{\rm\delta _p}}\comma\, y \supseteq {{\rm\delta _d}}\}\comma $$

where S ₁ is the set of designs D with description x, y; x is the set of design descriptors for primary elements, and y is the set of design descriptors for detail elements; δ_p is the set of design descriptors for primary elements that are common to a set of designs; and δ_d is the set of design descriptors for detail elements that are common to a set of designs.

Fig. 7.

Derivation from original grammar O.

Table 1.

Designs generated by the grammar O

Note: Descriptors that are common to all designs have dark shading, and descriptors that are common to three designs have light shading.

Based on the analysis in Table 1, we instantiate common adjectival descriptors as

• δ_p = {simple, partly unified and partly diversified, balanced, dominance of vertical primitives, dominance of monolithic relationships, dominance of axial relationships}

• δ_d = {simple, partly unified and partly diversified, balanced, dominance of horizontal primitives, dominance of monolithic relationships}

Designs O-I, O-II, and O-III fit into this style description.

A related style S ₂ could be defined by instantiating

• δ_p = {partly unified and partly diversified, balanced, dominance of monolithic relationships, dominance of axial relationships}

• δ_d = {simple, dominance of horizontal primitives}

Hence,

$${S_2} = {\{} O\hbox{-} I\comma\,O\hbox{-} II\comma \,O\hbox{-} III\comma\,O\hbox{-} IV{\}\comma} $$

where S ₁ is a substyle of S ₂.

Based on Equation (2), we used the commonalities in rule applications in designs (Table 1), to define a style S ₃ as

$${S_3} = { \{} {D_d}:d \supseteq {\rm\rho} {\}}\comma $$

by instantiating

$${\rm \rho} = (B\,1.1,\,C\,1.1,\,D\, 1.2).$$

Designs O-I, O-III, and O-IV belong to this set, and S ₃ is also a substyle of S ₂.

3.4. Experiment with designers

The original grammar assembled was tested for strategic style change in an experiment involving two designers at the University of Strathclyde. Due to the small number of designers who were familiar with grammars, this experiment was carried out with two designers only. Designer A had worked as an architectural assistant for 1 year whereas Designer B had worked as an architect for 3 years. Both designers were familiar with techniques of grammars and grammar transformations, and had used them previously at the university.

The design task was a paper-and-pencil exercise framed in two 1-h sessions. Prior to the first session, the designers were provided with the original grammar, designs that were generated from it, and the rule base, along with its style description scheme, in order to familiarize them with the exercise. In the first session, designers were asked to transform the primitives in the original grammar based on given style goals. Designers could add, delete, or modify the rules in the original grammar with rules from the rule base. In the second session, designers were asked to transform the spatial relations in the original grammar based on given style goals. Designers were asked to generate designs from the transformed grammars. The grammars authored by Designer A were termed T2 and T3, whereas the grammars authored by Designer B were termed T1 and T4. The generation of correct mobile phone designs required the application of rules from each rule set. In addition to correct designs, the designers were asked to generate an incomplete design as well. This was to test whether inconsistency in the application of rules in design derivations produced variation in adjectival descriptor ranks of designs. Finally, the designers were provided with feedback forms to elicit their views on transforming grammars with the help of the style description scheme.

The experiment was designed to accommodate the complexity of the design problem. First, we developed a rule base with wide-ranging adjectival descriptor ranks, so that the designers had choice in rule selection. Second, goals involved the manipulation of only one adjectival descriptor at a time that facilitated the evaluation of grammars and designs. Different goals were given out to each designer in order to generate diversity in grammars and designs.

We used a feedback form to gauge the views of the designers regarding the grammar transformations using the style description scheme. The feedback form was designed using the questionnaire design methods described by Oppenheim (Reference Oppenheim1992). Open-ended questions were prepared, so that designers could freely express themselves. Both designers filled out and returned the feedback forms.

3.5. Analysis: Transformed grammars and designs

Transformed grammar T1 is elaborated here. The goal given to the designer was to constrain the descriptors to the range “partly rectilinear and partly curvilinear” (0) to “very curvilinear” (+2). The designer could search the rule base and based on their judgment decide which rules to include in the grammar. The designer selected a total of 52 rules that included 11 primary rules and 41 detail rules. Initial shape 3 was selected in the grammar.

Design descriptor ranks were computed for the designs generated from the grammar (Table 2) based on the ranking conditions given in Table A.3 in Appendix A. Based on Eq. (1), we defined a style S ₄ as

$${S_4} = \{ D_{x,y}:x \supseteq {\rm \delta}_{p},\,y \supseteq {\rm\delta}_{d}\}.$$

Here,

• δ_p = {simple, balanced, dominance of curvilinear primitives}

• δ_d = {simple, balanced, dominance of curvilinear primitives, dominance of monolithic relationships, dominance of axial relationships}

• S ₄ = {T1-I, T1-II, T1-III}

Based on Eq. (2), we defined a style S ₅ using the following commonalities in rule applications (Table 2):

• ρ = (C 1.2, D 1.2, G 1.3, J 1.1)

• S ₅ = {T1-I, T1-III, T1-IV}

A commonality in the designs generated by transformed grammar T1 is the dominance of the constrained adjectival descriptor rank “curvilinear” in primary elements as well as details.

Table 2.

Designs generated by transformed grammar T1

Note: Descriptors that are common to all designs have dark shading, and descriptors that are common to three designs have light shading.

We compared the adjectival descriptor ranks present in the rule base, the original grammar O, and the transformed grammar T1 using the measures of style range and style mode (Fig. 8). With the exception of two descriptors in primary rules, the rule base had rules with descriptor ranks that covered the complete range. While the original grammar O also had a complete range of the descriptor “very rectilinear”—“very curvilinear” for both primary and detail rules, grammar T1 was constrained to the range between the median and “very curvilinear.” This also impacted the modal values of T1 for the aforementioned descriptor.

Fig. 8.

Style ranges and style modes of the rule base, original grammar O and transformed grammar T1.

Transformed grammars T2, T3, and T4 involved constraining descriptors from the median position to “very rectilinear,” “very fragmentary, and “very monolithic,” respectively. Designs generated from these grammars were analyzed (Table 3). Designs generated by grammar T2 had a large number of commonalities in descriptor ranks. Hence, definitions of style from grammar T2 were closely related to one another. Using our techniques of style definition, it is also possible to categorize a style with designs from different grammars. For instance, T4-I, T4-II, O-II, and O-III can be categorized as a style.

Table 3.

Examples of designs generated by grammars T2, T3, and T4 with adjectival descriptors

Note: Descriptors that are common to all designs have dark shading.

4.1. Framework for strategic style change

We assessed the use of our framework for strategic style change with the feedback given by the designers on the feedback form. Descriptors aided designers in searching through the large number of rules and selecting appropriate ones. Designer B found the adjectival descriptors useful in assembling the grammar. Designer A, in contrast, found the descriptors easy to use but commented on the task being simple and tedious at times. However, both designers agreed that the generated designs conformed to the style goals. As Designer A commented,

Designer A had an ambiguous response regarding the ease of use of the overall framework. A concern voiced by the designer was the issue of subjectivity in defining the descriptors and specifying their limits.

An unexpected advantage of the framework was pointed out by Designer B, who felt that the rule base and its descriptors helped in the design of a product domain with which the designer was previously unfamiliar.

4.2. Techniques of definition of style

Our examples illustrated the proposition that the design language of a grammar may host a number of styles or substyles. We found the first technique of defining style, which relied on commonalities in adjectival descriptor ranks of designs, to be more pertinent than the second technique, which relied on the commonalities in rule application in design derivations. This may be attributed to the fact that different rules that added similar subshapes in rule applications yielded the same adjectival descriptor ranks. Hence, there were more commonalities in adjectival descriptor ranks of designs than in their rule applications (Tables 2 and 3). Both techniques allowed incomplete designs, as well as designs from different grammars, to be included in style definitions. Inconsistency in the application of rules in derivations of incomplete designs did not result in a significant variation in adjectival descriptor ranks.

4.3. Relationship between grammar and style(s) of designs

We studied the relationship between the style range of a grammar and the style(s) of designs generated by a grammar. The style range showed how varied the rules were in a grammar. From the examples, we deduced that if a grammar had a large range, then a number of diverse styles could be defined in its design language. For instance, assuming the rule base to be one large grammar, diverse styles such as S ₁ and S ₅ could be defined in its design language. Grammars with narrow ranges hosted fewer styles that were similar to one another. Representation of the style range of the transformed grammars also made evident the change in adjectival descriptor ranks in these grammars from the original grammar.

We also investigated the relationship between the normalized modal descriptor ranks of grammars and the dominant descriptor ranks of designs generated by them (Fig. 9). Although the adjectival descriptor ranks of designs mostly conformed to modal ranks of grammars, it was not so in all cases. Hence, we deduced that modal descriptor ranks of a grammar could only be moderately indicative of the adjectival descriptor ranks of designs generated by it. This is in keeping with Knight's (Reference Knight1999) observation that the greater the complexity of a grammar, the more difficult it is to predict with precision the nature of designs generated by it. Our analysis revealed that the descriptors constrained in the grammar were present in all designs as dominant descriptor ranks. This is attributed to the repetition of the constrained style descriptor in design derivations, resulting in the dominance of those adjectival descriptors in designs. Thus, we deduced that constraining rules in grammars was a valid technique for ensuring dominance of adjectival descriptors in designs and thus aiding in strategic style change.

Fig. 9.

Comparison of dominant descriptor ranks of designs with normalized modal descriptor ranks of grammars.