Rules for Composing Graphics

A third class of rules concerns the composition of graphics. Senay and Ignatius, who extended Mackinlay’s approach to automate the design of visualisation (APT) [Mac86a] for their visual- isation system Vista [SI94], introduced a set of rules on composition to complement rules on expressiveness and effectiveness. These composition rules define constraints on data charac- teristics and apply to the five composition techniques defined by the authors. For each kind of composition, Senay and Ignatius describe rules and associated conditions that concern the »compatibility of component visualisation techniques«, the »visibility of each component upon

CHAPTER 5. A VISUALISATION ONTOLOGY – VISO

composition« and the »distinguishability of components in the composite design«. Although, unlike expressiveness and effectiveness, the authors do not subsume composition rules as »rules of visual perception«, many rules concern the perception of the final graphic (e. g., »fields should mostly be visible through the marks of the other component«). These rules are often vague and difficult to formalise.

While we do not discuss graphic composition here in detail (we return to composition in Sect. 6.5, 6.6 and 8.2), the important thing to note is that the composition of graphics requires an additional set of facts and rules in order to respect syntactical and perceptual aspects. Based on Senay and Ignatius’ composition rules as well as on statements of Wilkinson and Mackinlay on »bundling« and »interaction« of attributes, we propose the following two types of relations in which two graphic relations may take part: Dependency and Interaction.

Dependency

Dependency applies only to visual attributes and can directly be explained by the multi- dimensionality of graphic spaces [vE02]. The problem of dependent attributes needs to be considered, wherever combinations of the dimensions that span a graphic space are used for visual mapping. Two attributes are independent, if it is possible to set them to any value without changing the values of other variables, e. g., colour and shape can be varied independently. Visual attributes are dependent, if one variable is composed from the other, for example area = height ∗ width. That means area and height are dependent as well as area and width. As a consequence, it is not possible to use all three involved attributes at the same time (two of them can still be used, as long as the third is not encoded). Dependency is closely related to the observations of Wilkinson, who observes that attributes can be bundled. For example, he describes colour as a bundle of hue, saturation and brightness [Wil05]. For independent attributes, Wilkinson also uses the term orthogonal. Similarly, Engelhardt calls size a »versatile« attribute and notes that variations of size can either be homogeneous or »restricted to height, length or width of an object« [vE02].

depends on

http://purl.org/viso/facts/depends_on

States whether a graphic attribute depends on another graphic relation, i. e., whether changing the graphic values for the one attribute will always change the values for the other attribute. Type: owl:ObjectProperty

Interaction

By interaction we refer to a relationship between graphic relations having the potential to influence the composition of graphic relations – at least for certain values. Usually, this influence manifests itself in a negative way. Wilkinson [Wil05] remarks that »orthogonalisation in design means making every dimension of variation that is available to one object available to another «. That is, some graphic relations are not orthogonal or independent per se, but have to be made independent by putting appropriate constraints. Further, he states that »how these variations are perceived is another matter «. Accordingly, we subdivide interactions and the corresponding constraints into two groups: Syntactic interactions and constraints and perceptual interactions and constraints. We differentiate, whether an interaction concerns a fundamental syntactical issue – which makes it impossible to construct or decode a certain composed visualisation – or a problem with human perception. Unlike for dependency, interactions do not only apply to the composition of graphic attributes, but also to the composition of graphic object-to-object relations, as it will become clear from the examples below.

5.7. VISO/FACTS MODULE – FACTS FOR VIS. CONSTRAINTS AND RULES

Syntactic interactions and constraints Although some visual attributes such as shape and colour are clearly orthogonal and can be varied independently, others such as shape and size are not independent per se. This is again supported by Wilkinson, who states that »shape must vary without affecting size, rotation and other attributes«. That means, two shapes are different and distinguishable, only if one cannot be created from the other by rotation or scaling. Hence, when both attributes shall be used, constraints have to be put on the shapes allowed for encoding (Fig. 5.13, top; the same problem arises, when no orientation exists in a graphic, e. g., at table-displays where users sit around the presentation). The second and third row of Fig. 5.13 show two examples of syntactic interactions involving object-to-object-relations: Superimposition can only be recognised as such, if the objects are filled, otherwise it cannot be distinguished from (overlapping) containment. Similarly, transparency can only be recognised as such, when the transparent objects do partially overlap (again superimposition), otherwise it cannot be distinguished from variations of colour parameters such as saturation and lightness. Syntactic constraints between object-to-object-relations also apply to ordered line-up and linking, when we try to avoid crossing connectors (Fig. 5.13, bottom). The composition with an ordered line-up makes this harder than the problem of planarity alone, because the position of nodes is not arbitrary anymore.

interacts with (syntactically)

http://purl.org/viso/facts/interacts_with_syntactically

States whether a graphic relation syntactically interacts with another graphic relation. Two graphic relations syntactically interact, if data encoded by these relations cannot be decoded in a unique way, leading to ambigious interpretations. We also speak of syntactic interaction, when the composition only works under additional constraints.

Superproperties: viso-facts:interacts_with Type: owl:ObjectProperty

Perceptual interactions and constraints Perceptual interaction between two graphic relations may cause difficulties in interpreting when an encoded value takes certain values, such as extreme ones. The limitation is given by the human eye (e. g., resolution) and brain (e. g., distinction of colours). This is different from two graphic relations being in syntactic conflict, as described in the last paragraph. As an example of interaction, Mackinlay [Mac86a] describes the perceptual problem that occurs if shape and size of a visual object both encode information and size takes very low values (illustrated in Fig. 5.14, top row). Mackinlay already used the term interaction, however, he only refers to visual attributes, not graphic relations in general. But also the composition of object-to-object relations may be under perceptual constraints. For an example of how lineup and separation by a separator interfere with shape (second row) respectively clustering (last row), please refer to the detailed description of Fig. 5.14.

interacts with (perceptually)

http://purl.org/viso/facts/interacts_with_perceptually

States whether a graphic relation perceptually interacts with another graphic relation, i. e., whether the visual perception and interpretation by humans becomes difficult. In contrast to syntactic interaction, the interpretation of the composed graphic is still (theoretically) possible though.

Superproperties: viso-facts:interacts_with Type: owl:ObjectProperty

CHAPTER 5. A VISUALISATION ONTOLOGY – VISO

Figure 5.13: Examples of syntactic interaction between different graphic relations causing problems during composition. Top row: If both shape and rotation shall be used, constraints have to be put on the allowed shapes, otherwise rotating an object cannot be distinguished from changing an object’s shape. Second and third row: Further examples of syntactic interactions: Superimposition is only perceptible if objects are filled, otherwise it could be confused with (overlapping) containment . Also transparency is only perceptible, if the transparent objects partially overlap, otherwise it could be confused with variations of colour parameters such as saturation and lightness. Bottom row: Syntactic constraints between ordered line-up and linking , when we try to avoid crossing connectors while keeping the position of the nodes.

5.7. VISO/FACTS MODULE – FACTS FOR VIS. CONSTRAINTS AND RULES

Figure 5.14: Examples of perceptual interaction between different graphic relations causing problems during composition. The example in the first row is taken from a figure of Mackinlay [Mac86a], titled the »Interaction between Visual Variables«, which shows how size and shape interact for small values. Other examples of such interactions are the interaction between shape and the shape used for separators, which must clearly be distinguishable from other shapes in the graphic (second row) or the interaction of clustering and lineup, since a lineup can only be perceived as such as long as the clustering does not too much distort its appearance (last row).

CHAPTER 5. A VISUALISATION ONTOLOGY – VISO

Constraints on the Usage of Single Graphic Relations

While, so far, we only discussed constraints for composing different graphic relations, rules exist as well for composing graphic objects by means of a single graphic relation. Again, we can differentiate between constraints that apply due to syntactic problems and those that apply due to perceptional problems. In some cases, syntactic constraints apply to this kind of compositions such as the problem of non-crossing connectors (planarity; Fig. 5.15, first row). Moreover, not every graphic relation supports every possible graph structure (e. g., containment cannot represent cyclic structures; Fig. 5.15, third row). But also perceptual constraints apply. For example, the number of graphic objects often needs to be constrained to produce a perceptually effective graphic (Fig. 5.15, bottom row).

Implementation in VISO

Currently, VISO is able to describe which graphic relations depend on each other or interact, but not how they do exactly. Constraints on the usage of single graphic relations cannot be specified so far.