Procedural level generation for games is an active field of research with successful applications. However, how to generate content that embodies design intent is still an open research question. Level designers lack abstractions and tools for authoring generated artifacts for affecting emotion. We propose a novel pattern language for generative level design inspired by Appraisal Theory. Its patterns enable designers to add meaning, depth, and cohesiveness to the resulting content, and modify artifacts to make the content more engaging. We illustrate how these patterns can be implemented in a generative grammar for level generation for an adventure game. Formative evaluation of generated level content demonstrates the feasibility of the approach and suggests points for further improvements. Future work could focus on other elements which seem important for affecting emotions, including pacing, perception, and expectation.
Procedural level generation is a kind of Procedural Content
Generation (PCG) that helps level designers author game
levels
This paper assesses the feasibility of formulating a design
pattern language for affecting emotions in level design
and the implementation of those patterns in generative
grammars for procedural level generation in order to help
achieve this type of direction. Our approach is based on
Appraisal Theory
The assumption is that patterns for affecting emotions are able to make the content more engaging and contribute to a more coherent player experience. The research is carried out in the context of a commercial adventure game currently in development, which relies on advanced content generation. The approach presented here is meant to be general enough to be applied to level generation for other adventure games, in particular methods that incorporate generative grammars. Ultimately, the goal is to enable the discipline of level director who orchestrates the level generation process and affects related dramatic aspects for improving the overall perceived quality and player experience.
Generating content that embodies game design intent is a key open research challenge. Here we relate our work to conceptual frameworks, techniques for PCG, design pattern languages, and emotional modeling.
Proceduralist readings is a framework that connects game
elements, mechanisms and dynamics, and explicitly
formulates rhetorial arguments in a thematic and cultural setting
from which meaning is derived
Procedural content generation can be employed for various
reasons, including generating variations for increased
replayability, saving time for the designer by automating the
design process, gaining a better understanding of the
creative processes in design, etc.
Although we understand that each of these goals may be
important in specific game development or research
contexts, we believe that for successful game development, the
quality and coherence of the generated results is critical.
PCG allows game developers to create an experience where
players cannot rely on previous experience with the same
content, creating an experience that is different from games
where all levels are pre-designed. But this works best when
certain standards are met in the generated levels
Theories on emotional process include categorical
Appraisal Theory states that “…people’s emotions arise
from their perceptions of their circumstances – immediate,
imagined, or remembered”
Design patterns were first introduced by architect
Our pattern language represents a new category that introduces applied Appraisal Theory to create high-level emotional patterns and low-level appraisal patterns that steer the content generation towards structures we relate to desired emotional effects.
Following the approach of design patterns, combining
elements from different design pattern templates and
approaches
Appraisal patterns support the emotion patterns by describing how to use composition, structure and meaningful objects in abstract solutions to realize the relevant appraisals. Patterns currently formulated within the language are Centering, Pointing Out, Appearance of Objects, Frame by Surrounding, Content Density, Symmetry, Just Not Symmetric, Rewards, Rare Sightseeing and Conflict. The Centering pattern is presented in Table 3 as an example.
Based on Appraisal Theory, specific emotions arise from a specific set of appraisals. Therefore, by combining multiple patterns for appraisal, a pattern for emotion can be applied. For example, in Table 4, the design pattern for the emotion Wonder is presented. The solution part of this pattern, depicted in Figure 1, shows the required appraisals (Novelty, Pleasantness and the negation !Urgency), together with potential ways to realize these appraisals in a particular form. The appraisal Novelty can be realized, for example, by Centering or Pointing Out the novel object, or by Frame by Surrounding.
Name. Centering Problem. How to make an element of interest stand out in a level. Forces. The element that should capture the attention has a particular importance or amount of focus, such that it stands out while all other objects are not completely overlooked. The more diegetic the way of drawing attention is, the better it fits the experience. The size of the element is important, since it can be hard to make small elements stand out. The surroundings influence the ease at which something can be made stand out and its resulting effect. Patterns triggering novelty inhibit emotion patterns that require the absence novelty.
Context. A particular object, group of objects or other applied pattern should capture the attention. A separated area pattern has been applied to the level, and the resulting area is large enough to consist of the object to capture attention and center things inside. Importance: the amount of focus this pattern can provide is medium to high, depending on the size and the surroundings. Diegetic: the implementation is diegetic but can be influenced by the surroundings. In some situations, it is not natural to center a specific element. Size: the size of the element of interest is at least medium and can be greater in relation to the other objects in the surroundings. Surroundings: the surroundings of the application of this pattern allow for centering by consisting of an area that stands out in such a way that it can be recognized as a separated area. All other forces are influenced by the surroundings.
Participants. Object of focus: the object, or group of objects, that captures the attention. Area: the available space, defined the applied separated area pattern, to place the object.
Affected appraisals. Novelty: medium – high Solution. Place the object, or group of objects, to stand out in the center of the area created by the applied separated area pattern.
Relations. Requires a separated area pattern to create an area to focus in. Reinforces other novel objects patterns and rare sightseeing since applying those together results in a higher novelty. Extends symmetry, just not symmetric and rewarding resources since it adds novelty to the effect of those patterns.
Example. In the game currently under development, scrying points are a way of exploring the surrounding locations quickly. Such a point is indicated by a pillar-like object placed somewhere in the level. To make this object capture more attention and stand out, it can be centered within a frame. A variety of options can be used to create the frame: the area itself can be formed in a symmetric, basic shape, or a circle can be created with flowers.
Variations. Place the object on the edge of the area. Note that this is
only possible if you can center on the form of the given area.
Inspirations. Metropolis – Ralph Koltai: The elevator – which rises
into the city above in act I – is centered in the middle of the ground
Name. Wonder Problem. How to evoke wonder in a given part of a level.
Core relational theme. Interest in the unusual or extraordinary. Forces. Emotions can vary in intensity. A higher intensity will not only cause the emotion itself to be stronger but will also increase the number of different players actually experiencing the desired emotion. Some emotions lose their impact when they are experienced more often. They are also experienced more fully when they are alternated with other emotions. One cannot experience joy forever for it would become the norm. Therefore, the amount of usages of patterns of emotions is correlated with the impact of it.
Context. The whole level or a particular part should give the player the sensation of wonder. Intensity: the intensity heavily depends on the selected low-level patterns and the application of those low-levels. To increase intensity further, low-level patterns affecting the same appraisal can be combined. Amount of usages: since wonder captures a feeling of surprise, an unusual feeling, the amount of usages should be limited and alternated with other emotions.
Participants. The selected appraisal patterns and their participants. Required appraisals. Novelty: medium – high.
Pleasantness: medium – high.
Urgency: none – low.
Solution. See Figure 1.
Example. A moment of wonder can be created in the game under development by triggering all the listed appraisals. The player walks through the mountains when suddenly the pebble path changes in an old road with tile patterns, triggering novelty by appearance of objects. The surroundings are flattened and surrounded by pillars, which increases novelty with the use of the frame by surrounding pattern. Large statues of faces are artifacts of an ancient civilization and, although they spawn a number of times in a world, they are quite rare. One is placed in the center of the frame, which further increases novelty with the centering pattern, and adds pleasantness using the rare sightseeing pattern. The appraisal Pleasantness can be realized using Symmetry, or by presenting a Rare Sightseeing, or by Rewarding Resources. To give an example of another emotion that has been formulated in our pattern language, Fear requires a different set of appraisals, including relatively high Novelty, Urgency, Outcome probability, Conduciveness, but no or low Pleasantness, Adjustment and Power, and Agency associated with other actors than the player.
As a proof-of-concept, a design pattern for emotional effect
and its associated appraisal patterns have been implemented
to generate appealing content in the form of levels for the
game currently in development (The Sequel to Unexplored),
using generative grammars in Ludoscope
C. After some intermediate steps, all non-terminal symbols are resolved to terminal symbols.
D. Unity interprets the level and places all objects accordingly, resulting in a graphical scene to be presented in the game.
Since the application of the Wonder pattern is a matter of combining low-level patterns that affect those appraisals, with only a few different low-level pattern implementations variations can easily be realized. For example, in Figure 2D, Wonder was realized by Frame by Surrounding (circle of objects) with Rewarding Resources (experience points) and Centering (scrying point in the middle of the frame), but the Centering and the Frame by Surrounding patterns could have been instantiated using other game objects, and instead of Rewarding Resources, other realizations of the Pleasantness appraisal could have been selected.
Some patterns are placed with one rule, such as the Centering pattern which only marks the center tile. Other patterns are placed with a sequence of rules that build up the pattern in a stepwise manner, such as the Frame by Surrounding pattern. After a pattern is placed, interesting locations remain marked so participants can be placed later on.
Our approach is generic, but allows for domain- (i.e., game) specific extensions. The design patterns for emotions and appraisals can also be applied in procedural content generation for other games, or even other multimedia applications.
To cater for specific uses, users need to adapt the rules that specify how appraisals can be realized using elements of that particular game or application. When the elements are similar between games, the rules will be similar too. If other emotional effects are desired, new patterns can be added for those, without the need to change others. The appraisals for many emotions partly overlap, which means that the work done for these appraisals can be reused.
A content generation system that is able to achieve many different emotional effects, will need many rules, which increases its complexity and introduces new development and maintenance challenges. However, the specification of design intent in the design patterns makes the approach more modular and scalable than is often the case in procedural content generation systems; many rules will be variant implementations of basic appraisals, which in turn can be easily assembled into higher-level patterns. The modularity of the approach keeps the complexity under control, even when working with a large number of implemented patterns.
In addition, the design patterns facilitate a separation of concerns for the different responsibilities of a sophisticated content generator. The proposed patterns direct the implementation of high level design goals for individual levels or level sections. The responsibility for what those design goals might be can be placed elsewhere, either with a different expert algorithm, a human designer, or something as simple as randomly selected design goals.
In our approach we use design patterns as a means to capture and express design intent explicitly. However, rather than specifying what needs to be generated, we view design intent as setting goals and necessary constraints, leaving ample room for unexpected variation in content and form.
We consider our approach analogous to the guidance given by a director in improvisational theater sports. In such a context, directions are typically not expressing what needs to be said or done directly, but rather the effect to be reached, or constraints to be taken into account. This leaves the result underspecified, which often leads to surprising outcomes. We view procedural content generation in a similar vein, not specifying directly what needs to be generated, but rather the effect to be achieved, and multiple methods for achieving it, so that the results may still surprise the designer too.
To assess whether the approach is feasible, the generated levels were evaluated in a meeting with three professionals from the game industry. The experts confirmed the value of the insights gained in this study, and that the implementation approach of patterns in generative grammars is valuable. Nevertheless, the implementation should be more prominent. Several promising improvements were suggested, e.g., applying multiple appraisal patterns per required appraisal or creating anticipations and expectations.
Additionally, the experts suggested that the current pattern language should be supplemented by taking into account elements that contribute to creating overarching experiences. Interesting elements to explore are the surroundings and the context of a pattern, the expectations and the anticipations of the player, pacing between emotion patterns and the perception of the game world. More thorough validation, and evaluation through playtesting is part of future work.
Applying and extending and implementing the pattern language is work in progress. We identify the following bottlenecks and actionable steps: • Recognizing shapes with variable sizes and consistency. • Obtaining information about objects in the surroundings, such as the average color or size. • Influencing the surroundings of a pattern. • Providing sufficient variations and possible combinations of patterns to prevent recognizable repetition.
Further work on the pattern language includes the addition of more emotion and appraisal patterns, and improving the implementation of patterns in generative grammars by addressing the limitations mentioned above. When the main implications have been identified and addressed, validation of the pattern language with more game design experts from research and industry is desired, as well as playtests with users to analyze the actual emotion effects, to further improve the language and the implementation of patterns.
Future research could focus on other elements that seem important for affecting emotions within the overarching experience of playing a game, such as pacing, perception, anticipation and expectation.
This study explored the feasibility of formulating design patterns for emotion in level design and the implementation of the resulting patterns in generative grammars for level generation. Using Appraisal Theory as a theoretical foundation, design patterns were presented for emotions such as fear and wonder, and their required appraisals, which describe how to use composition, structure and meaningful objects in abstract solutions.
According to Appraisal Theory, appraisals are made regarding the relationship between the event and the person involved and, together, they support the occurrence of an emotion. This theory depicts the structure of the presented pattern language in which appraisal patterns (triggering appraisals) are combined to apply emotion patterns (affecting emotions). An approach was proposed to implement such design patterns, expressing how rewrite rules can be created by increasingly adding features to and removing abstraction from the level, resulting in an applied pattern.
To prove the feasibility of implementing such patterns, a procedural level generator was implemented for a game currently in development, using an approach based on generative grammars. Preliminary evaluation with three experienced game designers has indicated that it is possible to create moments in which a given emotion is suggested. From a technical point of view, the use of the design patterns structures the content generator in such way that different concerns are handled by different components. This results in an architecture that is modular and scales well enough to accommodate the amount of content that is required for a commercial game.
The main contributions of this paper are (1) a new framework for capturing design intent as direction in grammarbased procedural level generation (2) the proposed pattern language for design patterns for emotions and the required appraisals, and (3) the implementation of a level generator that allows the game designer to play the role of a level director wishing to achieve emotional effects. These contributions are considered valuable for the research field of procedural generation of content for games and other interactive media. This is especially true when an approach is taken that involves design intentions, or when emotional effects are desired. The theoretical foundation of Appraisal Theory allows for logical design choices that allow extending the pattern language with other emotions.
Recommendations for future work include expanding the current set of patterns, addressing the limitations that were identified for the pattern implementation, validating the pattern language within a wider community of practitioners and researchers, and exploring the aspects in games that influence overarching experiences of emotional dynamics.
This research has been carried out in the context of the ‘Live Game Design’ project, which is subsidized by the Dutch funding body for applied research NWO-SIA. In addition, we thank indie-game studio Ludomotion for supporting this research. Gameplay terns.org