Petri Lankoski
Media Lab
University of Art and Design Helsinki

Staffan Björk
Interaction Design Collegium
Chalmers University of Technology & Göteborg University

In GDTW2007 Proceedings
Fifth International Game Design and Technology Workshop and Conference
Liverpool John Moores University, UK
14–15 November 2007

ABSTRACT
This paper explores how games can be designed to make the social networks of characters as part of the gameplay. We start with a premise that game characters and social relations between them are import in games. We examine several games and derive gameplay design patterns from those games. Models from social network analysis, actor-network theory and Egri’s model for dramatic conflict is used to focus the analysis. In addition to isolating design patterns from existing features of the games, we look situations where game structures do not support social networks or conflicts as proposed in above-mentioned theories. Patterns identified include Competing for Attention, Gain Allies, Social Dilemma, Internal Conflict, and Social Maintenance.

Keywords
Gameplay Design Patterns, Gameplay, Narration, Non-player Character, Computer Games, Gameplay Design

1. INTRODUCTION
As social creatures, humans easier to engage in a game and narration when characters portrayed in these have social relations to each other, or in other words that the relations between characters form a social network. This is common knowledge within scriptwriting theories for theatre and film (see, e.g., [6, 7, 17, 19]), and these theories are also applied to creating games. However, social relations in games are typically part of the storyline (see, e.g., Thief II: The Metal Age [34], Dead or Alive 3 [44], Silent Hill 3 [45], and Half-Life [48]) and games typically do not let players directly act to influences those relations, instead letting them be consequences of other (most commonly physical) actions that are shown through cut-scenes. One example of this can be found in Quake 4 [22] where the relation between the player character, Matthew Kane, and the other characters in the Rhino squad are only changed in the cut scenes. No possibilities to do so are available during gameplay, including making it impossible for the player to terminating the relationship by killing the other team members. When players are given direct choices to influence the relationship this is typically done as explicit choices between a limited set of alternatives, and the effects of these are localized and seldom have the complexity of nuances of real social relationships, including how one change in a relationship can propagate through a whole network. Although these limitations typically make sense from gameplay or storytelling point of views, we think that the above-mentioned ways limits the design space of games, and having further alternatives would expand the expressive design space of games.

By limiting how players can affect social relations within a game, game designers are making social networks primarily a thematic aspect of the game rather than a core gameplay feature. Conversely, this may be a reason why emotional engagement in game characters is more difficult to achieve than for other types of fictional characters. In this paper we focus upon how using social relations as gameplay feature may expand the possible gameplay space and put stronger focus upon social actions rather that physical (or violent) ones.

There are of course social networks in many games, and we differentiate between the social networks of characters in games compared to the social networks of players. This paper specifically looks at social networks of non-player characters (NPCs) with the possible inclusion of one player character (PC). This removes many type of games, such as multiplayer online worlds (e.g., World of Warcraft [10], Eve Online [11], and Guildwars [5]). The primary reason for this is that the social relations between players tend to overshadow social relations between characters in such environments ¹, and by avoiding these social networks of players the social networks of characters are more easily discernable. Thus, we will in the following primarily focus upon non-player characters.

The aim of this paper is to explore the gameplay design space of characters’ social networks in games. As professional game developers have reported a need to explicitly codify design knowledge within game design [12, 13], gameplay design patterns [9] have been selected for that purpose. The gameplay design patterns is one of the few conceptual framework developed to describe games and gameplay and is specifically developed to support design work. Other candidates (e.g., [18, 32, 50]) also provide concepts to describing the possibility space of gameplay mechanics, but are geared towards analytical studies and players’ perception of games respectively. Gameplay design patterns is a development of the design patterns approach within Architecture by Alexander et al. [3] modified to suit particular demands of gameplay design. Specifically, analytical and construction aspects are separated and the types of relations have been modified. Gameplay design patterns have previous been used to study mobile games [16], categorizing pervasive games [38], and we have earlier presented patterns for believable non-player characters [29]. Now we extend our work to cover social structures that can be used as basis for goals and conflicts in games.

Gameplay design patterns are design tools that preserve design knowledge to support reproduction of gameplay features. For the patterns described in this paper, the gameplay features are representations of social interaction and social networks. We stress that we are not concerned with providing true descriptions of everyday life structures, and indeed argue that is not needed; the patterns are concerned with providing enough correspondence between a game and daily life practices that players can perceive the social interaction as believable, or at least consistent within the system.

2. BACKGROUND
In this section we describe the models used to study the social relations of NPCs. After describing two models focusing on the group level, social network analysis and actor-network theory, we complement this with Egri’s model that focused more on the individual level, and conclude with some notes on orchestration.

2.1 Social Networks
Kathrine Isbister [24] points out that social networks are important in game character design. She suggests that one should focus not only on the characters themselves but also between the characters: what kinds of relationships character have between each other. However, she says very little how social relations can be implemented in the gameplay. In the following we present two descriptions of social network that provide specific concepts for parts of social networks. It should be noted that for both models some decision about what limits the extent of the network must be taken, and we assume that a game can contain a multitude of social networks with different limitations and that the networks may be overlapping. For example, although these models may normally be used to interrelate all participants in a conflict as being part of the same network, in the case of game design it may make sense to do that but also have a separate network for each side since this provides a means to control what actions different characters can or must take. Not only this, but different types of relations may be used simultaneously to create all interesting sides, e.g., to model treachery and traitors correctly.

2.1.1 Social Network Analysis
The interest of studying social networks mathematically have increased as the Internet has grown, most probably since this restricted form of social interaction provides quantifiable empirical data of large networks. These studies, called Social Network Analysis (SNA), are typically based upon graph models consisting of nodes representing people and lines representing relations between people, and then using various mathematical measures to identify specific characteristics (see [49]).

Example of these include N-cliques (groups where all people are connected to each other by traversing at most N lines), N-clans (N-cliques where one may only traverse lines to other members of the N-clique to determine membership), K-plexes (where a member must have connections to all but K nodes), and K-cores (where every member needs to have connections to at least K other members). Somewhat less rigorously defined concepts (see [26] include Singletons (people not participating in the network), Isolated Communities (groups that are connected to each other through one central member, creating a ‘star’ shape) and giant components (well-connected region persisting even when ‘stars’ are removed).

2.1.2 Actor-Network Theory
The network descriptions above typically have individuals as the nodes in the network, or rather their online profiles for specific applications such as Facebook [1] and Flickr [2]. However, in an effort to redefine social science proponents of actor-network theory (ANT) [33] argue that actors (nodes) in social networks should not only be humans; instead they are collections of heterogeneous entities consisting of humans, human-tool combinations, and non humans (e.g., technologies, machines, or materials). This expansion of actors is the effect of but one of the five areas of uncertainty ANT proponents argues the social science should embrace. The five areas regard the nature of groups, actors, actions, science, and textual accounts.

The ANT approach has been advocated within research on games as a way of more correctly looking at massively multiplayer online games [15] Notably, ANT argues that freedom to describe agency, as not only coming from humans when describing the social phenomena, so curses and rings can make their bearers perform actions, and love can make somebody survive ordeals. Thus, the theory supports that a game not only can explicitly use these abstract concepts in a model of a social network, but also should do so. Proponents of ANT argue that actors themselves should primarily do the descriptions of the component of the networks, or by as faithfully as possible observing these and documenting how they refer to their social relationships. This provides an obstacle to applying the theory directly in design processes since the designers typically want create the actors and the network simultaneously. Creating the actors first and then letting them create a network and report a network is a potentially interesting way of automatically generating social networks, but outside the scope of this paper.

Nevertheless, the actor-network theory can be useful for gameplay design since it highlights certain concepts and activities. As one example, goals in a game can be created from the view that groups are not stable entities in the ANT, but rather something that constantly needs maintenance, or in other words “if you stop making and remaking groups, you stop having groups” [33]. Another example, based on the ANT, is seeing a textual account as being objective only if there is a presence of objectors, which the author needs to acknowledge. This is an argument for having believable social networks, since the social network in a game naturally has one such objectors: the player character. Furthermore, if too many objectors are raised, the network and the gameplay will fall apart.

2.2 Characters
Looking from a perspective of dramatic writing, Lajos Egri [17] also argues for the importance of a social environment of characters: the social structures influence the character behavior and set-ups a field for conflict between the characters. Moreover, Egri has presented a method for orchestrating conflicts based on character traits (including social structures). Lankoski [28, 31] has affirmed that the method is usable to live-action and tabletop role-playing game design as well as computer game design. He has pointed out that especially the goals of the characters are a tool to create conflicts in a game; in a well-defined character-driven game the goals of the game for a player are derived from the goals of the player character are the same [28, 30]. We prefer Egri’s method as starting point of our inquiry as Egri’s focus is in designing better drama² thought social conflicts and believable characters, in contrast to above-mentioned approach by Isbister [24], which concentrates designing better characters for games. In fact, others have also based on their discussion on character design for games to Egri’s method [25, 41], but they mainly discuss techniques for narration.
Egri stresses that a believable conflict raises from the qualities of the character; goals or natures of antagonists are in collision, which put an action in motion. However, there can be many types of believable conflicts. Egri argues that writers create interesting conflicts based on well-defined characters. He sees that a conflict stays believable when transition and changes in the characters are justified by earlier progression or the qualities already exposed. Thus, events must be causally connected. The above-presented model has been modified for use in games [31] and was used to distill gameplay design patterns [29]. This model of believable character is the assumed basis for the design of individual aspect of NPCs in this paper, and to provide an insight to that basis the model is presented below.

Condensed Bone Structure of a Character

Physiology

• Sex
• Age
• Height, weight
• Hair, eyes and skin color
• Posture
• Appearance
• Defects
• Heredity features
• Physique

Sociology

• Class
• Occupation
• Education
• Family life
• Religion
• Race, nationality
• Social status
• Political views
• Hobbies

Psychology

• Moral standards
• Goals, ambitions
• Frustrations, disappointments
• Temperament
• Attitude toward life
• Obsessions
• Imagination
• Extrovertness
• Intelligence

Egri’s premise that believability of character action is compatible with the idea that understanding others is fundamental for predicting their actions based on the history of interactions and ones current knowledge of the other. Believable character action is behavior that does not violate our expectations drastically, i.e., an expectation that the other would not act that way in a given situation (for a more detailed discussion see [14, 27]).

Here we are interested traits that influence social dynamics between characters. We have earlier identified gameplay design patterns that relate to believability of NPCs. Some of those patterns are relevant to our current focus:

• Own Agenda and Goal-Driven Personal Development relate to Egri’s psychology: goals and ambitions.
• Awareness of Surroundings, Sense of Self, Emotional Attachment, and Initiative are prerequisites for social dynamics between characters. [29]

The structure above is person-centric in that it explains the causes of various social interaction of that character. However, it does not focus upon the social interactions themselves, which in a game is interesting since they are they actions human or AI players can perform. Based upon the hypothesis that the knowledge of creating believable premises for social networks of characters can be taken from other disciplines we focus upon the interaction instead, seeing these as independent elements of the games. This is necessary since both the traits that provide the rationale for social interaction and the social interaction acts themselves needs to be described as parts of a system if the social interaction is to be encoded into the gameplay rather than being a solely thematic aspect of the game.

2.3 Orchestration

When you are ready to select characters for your play, be careful to orchestrate them right. If all characters are the same type-for instance, if all them all bullies-it will be like an orchestra of nothing but drums. [17]

Although this paper does not go into detail on how to relate NPC design with overall orchestration in a game it is important to note that there is a direct relation. Thus the introduction of some concepts regarding this is necessary for grounding the continued discussion.

Egri’s premise relates to what game designer Richard Rouse calls a focus: it is a short description on the most import features of a game. [40] Egri has a narrower idea³ when he introduces the premise that is a proposition that a work tries to prove. The premise guides the writer in the design process. Egri gives following interpretations of premises of some classical plays [17]:

• Hamlet: “Great love defies even death”;
• King Lear: “Blind trust leads to destruction”;
• Dead End (S. Kingsley): “Poverty encourages crime”.

The orchestration, according to Egri, is about creating well defined and uncompromising characters in opposition whose actions in conflict will prove selected premise. Uncompromising means that antagonists in the work have so strong motives that finding the middle ground is impossible. Designer should be working to find most interesting course of action (or goal-conflict structure) that is believable. Believable conflict originates from the traits and qualities of the characters. [17]

Egri’s argument is valid in the context of games too. If motivations are weak, players might not perceive the goals and conflicts in a game as believable. This will affect the perception of other design decisions, e.g., players might start wonder why the player character cannot simply avoid the conflict, and why the player character is forced to do certain things.

3. METHOD
This paper focuses on the design possibilities of NPCs, but since this is done from a gameplay perspective, we also discuss player character goals as a means to set up conflicts and co-operation⁴ between the characters in a game. This view is somewhat complicated by the fact that games may let players directly control a character for a limited amount of time, and then transfer the point of control to another character. This typically means that the player does not perceive him- or herself as playing that character but rather using it as a resource in the game. A classic example of this is Chess, where players’ transfer their point-of-interaction between the different Chess pieces⁵. Games such as these are included in our study since these have clear game mechanical structures for social interaction between the characters.

The general approach in the study has been to positively identify patterns regarding social interaction and relations existing in game design rather than negatively identifying patterns that are lacking. However, when clear examples of how the presence of a pattern could create new gameplay possibilities the lack of patterns are noted. The method for identify patterns has been to play games and evaluating the gameplay in terms of the above-mentioned models. We provide several case studies, as our analysis has been iterative. We started with one already studied for a similar purpose and deduced patterns. Based on these findings, additional games were identified that would provide complementing examples and analyzed. The patterns identified in a game were successively searched for in the other games to provide validity of the pattern as well as provide additional insights to it.

Since the paper focuses upon design aspects, the use of the theoretical models and concepts are primarily to support the design. This means that how correctly the models represent reality or everyday life is less interesting to how well they provide a good structure for creating interesting designs, in this case help create interesting gameplay and, to a lesser degree, narration.
As in our previous paper [29], we do not describe complete design patterns, e.g., the relation between them, due to the space requirements needed. Instead we present the pattern names and the context in which they were identified. In addition to this, all new patterns are listed in an Appendix: Gameplay Design Patterns Introduced with short descriptions. In this fashion we used the patterns in an informal fashion, which sacrifices precision for accessibility, and do this with the aim of providing an overview of the area in this paper and mapping the field for future, more detailed, studies of individual patterns.

The games described have primarily been chosen due to having a focus upon clear gameplay interaction with, and between, NPCs and being (commercially) available. The latter reason is a (weak) indicator of that a game in question has been well received by players but also ensured availability for continued research. The games have intentionally been chosen so they represent different types of games allowing the analysis to look many different approaches to the design space.

4. CASE STUDIES
Next we will present case studies of various games and sketch finding as gameplay design patterns. We begin with The Elder Scrolls IV: Oblivion [8]-building on the findings and patterns collected in our previous inquiry [29]-and then continue with other cases. Previously identified patterns are differentiated from new patterns by having given references, and are in some cases generalized to be for both NPCs and a PC rather than just for the PC.

4.1 Oblivion
Characters in The Elder Scrolls IV: Oblivion [8] show a level of Initiative [29] to acknowledge the presence of other characters by greeting them or attacking them depending on the situation. However, mentions of characters both regarding discussion betweens NPCs and those between a NPC and the player character are based upon pre-scripted media snippets, meaning that the only cases where NPCs are part of conversation topics are when they are related to quests or services (such as training or trading). Thus NPCs do not have Context Dependent Reactions as they do not differentiate between which other NPC they are talking; as long as they are willing to talk to another NPC the things they say are independent of whom they talk to. The Gossip provided in this way has no meaning for Information Passing to other NPCs, rather their raison-de-être is to provide players’ with the possibility to gain information through Eavesdropping. The player can at any time interrupt these discussions, so the NPCs are not Competing for Attention.

The PC can join several different groups within the game, such as the fighter’s guild, the mage guild, and the thieves’ guild. Each of these is a Fraction, a specific social network where membership is defined by what actions are allowed, disallowed, and required. However, joining and advancing in these fractions are in most cases strictly controlled by requiring the player to complete quests given by a specific NPC, although which NPC gives quests changes as the player’s character advances. This method is probably implemented to give players an incentive to travel to the place where the quest-giving NPC is located, but reduces the player’s possibility to affect the social network severely. In SNA, the system would be described as a K-plex where K equals the number of people in the network minus one, and joining or advancing is equal to getting a relation. That is, the player only needs to have a relation with one NPC and that relation is the only one the player can have to members in the group. This NPC is a Social Gatekeeper, one that determines how and when new members can join the group. Loosening these requirements could provide several gameplay possibilities. Letting K be the number of people in the network minus two or more would require players to solve quests for more than one NPC before advancing. This could be used to require more travel in the game, and in effect having several Social Gatekeepers that have to be unanimous. Loosening the restriction that the number of NPCs with quests is equal to K would let players choose which subset of quests to complete from a larger set, and could let the player be enemies with some parts of the network while still belonging to it. This pattern could be described as Internal Rivalry.

Advancing within a guild makes the guild members react more positively towards the PC, so the game can be said to partially implement Actions have Social Consequences. However, once a character has reached a level in a fraction, that position is stable until one has advanced another level or performs actions that the fraction deem unacceptable and is rejected. The character does not need to perform any Social Maintenance, performing actions to redefine and reform the group, and for proponent of actor-network theory the fraction would thereby not be a group in a social network. Even if the unscripted redefining of social groups may be perceived by designers as permitting to unpredictable evolvement of the gameplay, requiring players to perform actions or complete quests to maintain their position in a fraction can provide additional Continuous Goals [9] to the gameplay.

The various guilds and other social networks in the game are in practice static with the exception of the player character and some effects of quests. This can be changed by providing gameplay mechanisms to recruiting new members to guilds (described as either expansions of the network or as tracings the network depending on which model is used), and a way to instantiate Gain Allies. It could also be a form of Social Maintenance if the guilds have recruitment as requirements or Social Norms.

4.2 Ico
Ico [42] have three main characters: Ico (the player character), princess Yorda, and the Queen. Ico is an Outcast⁶, a boy with horns and which is taken to a strange castle for sacrifice. In the first section of the game the player finds an imprisoned girl Yorda. Her mother, the Queen, tries to keep Yorda imprisoned by her control over shadowy figures (showing two forms of Hierarchical Fractions). The structure of the game can be described in the SNA as showing how dyad between the Queen and Ico-the smallest possible social network-is upset when the dyad becomes a triad by Ico and Yorda befriending each other.

Ico and Yorda are allied by the Mutual Goals [9], escape from the Queen. Ico needs Yorda’s help on opening magically locked doors and Yorda needs Ico to keep herself safe from shadows. Ico needs to guide and help Yorda through obstacles (Guide and Protect). This need of each other make them alternative doing actions beneficial to the other, which can be seen as creating the social network from an ANT perspective. If Yorda is captured by the shadowy figures the game ends, so she and Ico have Linked Destinies, since escape from the castle without her help is impossible.

The main predetermined change in the social network is when Ico finds a magical sword that can be used to open doors, and the Queen captures Yorda. However, after that Ico and Yorda still have Linked Destinies: to escape from the castle Ico must confront the Queen. In other words, the social network in Ico is predetermined and the only option to following a specific development is to end the game by failing. Possible redesigns that would imply different social network dynamics include enabling Ico to be able to save Yorda a number of times by pleading to the shadowy figures. The later can be described as a Favor, a promise of a future action against a Fraction one is member of due to social relations to someone not belonging to that Fraction.

4.3 UFO Afterlight
The resource management | tactical battle game UFO Afterlight ([4], one in a series) gives the controllable characters background history with stated family and friends relations. The wellbeing of friends and family is mechanized in the game as possible status changes. Although this can be seen as an instance of Others Fortune affects own Mood, but it is not possible to determine by playing if a character behaves the same way to all other characters or they have distinct Emotional Attachments [29] to the other characters.

Viewed from a social network analysis perspective, the network in the game are 1-cliques where the connection is of little interest, although the size of the clique changes over time, negatively due to losses in combats, retirements, and broken alliances; and positively due to new alliances, constructions of robots, and the coming of age of relatives to the characters. Characters in the game belong to one or two classes (scientist, technician, and soldier). These could be used to define other 1-cliques in the game since the classes restrict which actions can be performed together, e.g., being a soldier is required to go on a combat mission and being a technician is required to work in the workshop.

All members of teams sent to fight enemies can be said to indirectly redefine their social network as per the ANT, since they have to cooperate to win battles. However, scientists with medic training and technicians with suit training can perform more direct beneficial actions towards their fellow combatants by healing them or repairing their suits. Although this can be seen as a form of Social Maintenance, it is more of an effect on the dynamics level of the game than designed into the gameplay mechanics. The only distinction after a mission is whether a character survives or not, and all social actions can be reduced to this, including friendly fire. A possibility of introducing Social Maintenance would be to keep individual tabs on relations that changed do to direct interactions and common experiences, such as surviving a fire fight. The game keeps track of which characters were part of important missions, such as the opening of a teleportation device that let loose a new type of enemy, so the characters already has a form of Memory of Important Events. This could easily be incorporated into changing individual relations. However, since the possibilities of performing actions with these effects are unevenly distributed, addition actions and most likely a social game-mode during downtime would be needed to equalize the characters social capabilities. Having this gameplay functionality make characters dislike each other could be used to splinter the 1-cliques into smaller 1-cliques that require additional choices to be made, e.g., forcing the player to choose whether to bring two NPCs to a battle, despite them disliking each other and belonging to different 1-cliques⁷.

There is a second level of social networks in the game, that of relations between the Fractions (total of seven fractions in all including two human ones besides the player’s). The number of actors in the network changes over time, being introduced by game events and possibly disappearing by being exterminate, as so do their relations (between hostile, neutral, and allied), so the diplomatic social network is also dynamic. As new enemies appear, old ones are scripted to offer alliances, so a form of My Enemy’s Enemy is my Friend exists in the game, but alliance offers are not dynamically generated. Through this mechanic, Requesting Support is possible but it is not possible to predict the consequences of the request in advance.

The social network formed by the Fractions is linked to the individual social network as characters join or leave the player’s roaster as alliances are made or broken between the Fractions. This means that reaching a Gain Allies goal on the diplomatic level of the game can have as a direct effect that one succeeds in a Gain Allies goal on the character level. These new characters, which include aliens and robots, have thematically different descriptions but are treated as equal parts in the group as human characters. Social Maintenance is not a gameplay requirement for this social network but could easily be introduced, e.g., by actually requiring that some of the already occurring trade proposals are agreed for the relation to improve or at least not deteriorate. Another possibility, which would mirror the gaining of allies, could be to require characters to be sent to the other Fractions.

4.4 Crusader Kings
Crusader King [37] puts the player in control of a medieval dynasty in Europe that strives for power, but at any given point the player controls a ruler (a king, duke, or count). The power struggle is fought both through military and diplomatic means. The players has to consider papal support, having publicly acknowledged claims to provinces fought over, and having the loyalty of the dukes and counts under ones control since these most often provides the majority of the fighting forces in one’s Fraction. Each ruler has a court that provides the possibilities of assigning new dukes and counts, the positions of steward, spymaster, marshal, and chancellor, and being parts of political marriages. Given that the game stretches from 1066 to 1452 with each turn being one day, creating dynasties and dynastic alliances is of importance in the game. In fact, not having any blood-relatives to succeed the character one is currently playing when he dies results in the game ending. Every character, both rulers and court members, have traits (e.g., Lustful, Suspicious, Schizophrenia, Intricate Webweaver, and Scholarly Theologian) and primary stats (Martial, Diplomacy, Intrigue, Stewardship, and Loyalty) that influences relations with the PC and the characters.

The game clearly makes use of patterns of Loyalty and Hierarchical Fractions in a complex fashion: it is, for example, much easier to gain the loyalty of one dukes when they are your sons or brothers than when they are your uncles or grand-uncles. The importance of blood relations can appear to support Linked Destinies, but more often lead to Internal Rivalry since members of the same family have claims of rights to the same areas. The game instantiates Binding Promises, as rulers must support their lieges in war or risk losing their domains since refusal to help gives their liege a casus belli against their vassals. Binding Promises come into play in another way of the game: having Muslims or heathens as part of one’s court due to their skills can cause the pope to question one’s Loyalty to the Catholic Church and risk becoming an Outcast whom anybody can attack freely.

Expanding one’s social network can be said to be a core goal in the game since rising in a Hierarchical Fraction and gaining new subjects is one of the ways to determine success in the game. The social complexity of the game could be achieved by adding other social networks, such as Catharian Heretics, fighting order of the Knights Templars, or religious Orders like the Dominicans or the Franciscans. This could add Internal Rivalry, a need for Maintaining Lies, and also the possibility of False Accusations as political means. The theme of the game also provides a rich basis for introducing various needs of Social Maintenance, e.g., tournaments and pilgrimages.

4.5 Civilization IV
Although the anachronistic immortal leaders in Civilization IV [20] can be seen as player characters they are actually more often perceived as NPCs. This since they are typically controlled by AIs and, interesting for the topic of this paper, the game introduces new game mechanics specifically to handle social interaction with these. This diplomacy is primarily controlled through a value of how friendly or hostile the AI perceived the player to be. The value is influenced not only by the interactions with the AI but also with those towards other players, e.g., fighting the same enemy (indicated by the modifier “Our mutual military struggle brings us closer together”), trading with an enemy (“You have traded with our worst enemies!”), and not accepting a request to stop trading with another nation’s enemies (“You refused to stop trading with our worst enemy!”). These modifications, which can be seen as instantiating an Emotional Attachment [29], also express Memory of Important Events. Although these modifications do mainly relate to My Enemy’s Enemy is my Friend pacts, they do create patterns of Either You are with Me or against Me and Others Fortune affects own Mood.

From a perspective of the SNA, the game begins with the civilizations as lone actors and from which forms dyads, triads, and various 1-cliques as the game progresses, typically ending by merging into one single clique. Two of the game’s victory conditions can be directly linked to the social network: 1) destroying all other civilizations and thus the network leads to the conquest victory, and 2) sufficiently positive relations to other civilizations can result in diplomatic victory through being voted world leader. Various dyadic subgroups are possible in the game ( e.g., defense alliances), but also permanent alliances that create cliques that share victory conditions, in essence creating Linked Destinies. This use of Fractions could be expanded to multiple actors and different types of Fractions (so that phenomena such as military pacts, free trade agreements, international standards, IP rights, Olympic Games, etc., was treating with similar mechanisms), supporting gameplay to Gain Allies, to test one’s Loyalty between different alliances, and to fostering Internal Rivalry.

Civilization IV does not conform to the ANT in requiring individual actions to maintain social relations, although many such individual actions exists, e.g., demanding resources or territory and giving gifts in the form of technology or money. Cooling relations if no positive actions occur could be one way of requiring Social Maintenance, but other options include Gossip and Information Passing about other civilization’s status, Requesting Support for threats (which might require use of a Favor), or providing Outspoken Support for other’s threats. The events features adding in the Beyond the Sword expansion for the game could further be used to create additional, minor, diplomatic events, including False Accusations.

4.6 Canis Canem Edit (aka Bully)
In Canis Canem Edit [39] the player controls Jimmy, who is sent by his parents to the boarding school Bullworth Academy. Jimmy ends in the middle of a power play between different student Hierarchical Fractions (nerds, preps, bullies, jogs, greasers), and the gameplay focuses upon how the player can influence his relation to the Hierarchical Fractions. However, most of actions (like attacking people) do not influence how the NPCs react to Jimmy; only the missions seem to change respect values and behaviors. Each NPC has an Emotional Attachment to Jimmy’s actions, e.g., they can start a fight with somebody after seeing Jimmy kissing that person. Despite this reaction, the relation between Jimmy and the NPC doesn’t change. As a consequence, Jimmy can continue to date all of his girl- and boyfriends, so the NPCs cannot be said to have a Memory of Important Events. The social dynamics, thus, remains perceivably superficial. The NPCs seem to lack Goal-Driven Personal Development [29], or the game instantiates the pattern only partially: the only events that influence the personal development are the completions of the goals of PC. Hence, the social network evolves only in relation to the completed goals; Social Maintenance is not needed. This may seem odd because completing the main goal of the game requires that Jimmy wins the trust of the Hierarchical Fractions in the Bullworth.

One possibility to change the game would be to introduce more complex social network interactions, both in the SNA perspective of expanding networks or in the ANT perspective of redefining them. For the latter, the game could, for example, be based on Social Maintenance and finding ways to get the leaders of Hierarchical Fractions to like Jimmy. Each Hierarchical Fraction could have members, whom attitude toward Jimmy influence also indirectly to attitudes of others. Hierarchical Fractions also could have relations between each other. For the former, game dynamics could be based on pattern Gain Allies, and Competing for Attention. Besides the patterns mentioned above, this could require the use of patterns such as Actions Have Social Consequences (i.e., helping a Fraction member will help to get in good terms also with other members and attacking a member will be retaliated), Requesting Support, Outspoken Support, and make Loyalty, and Maintaining Lies as important goals of the game.

4.7 Splinter Cell: Double Agent
In Splinter Cell: Double Agent [47] the player controls Sam Fisher, which the NSA has an assignment by his employer the NSA to infiltrate a terrorist organization JBA. This forces the player to guide Sam into committing crimes. Completing goals increases the trust of the terrorists’ or NSA’s, but can at the same time reduce the trust of the other organization. The game uses Traitor pattern, as the game requires pretending to maintain an alliance while acting against it. However, the necessary Social Maintenance is per Fraction rather than per NPC.

In the game, the player is given moral dilemmas like if to shoot a prisoner to gain terrorists’ trust. The dilemmas are instantiated in the game by using Internal Conflict: Sam Fisher is given two Incompatible Goals [9], e.g., to kill the prisoner (and gain the trust of terrorists’) and to keep the prisoner alive (to avoid loosing the NSA’s trust). To generalize, the game builds these dilemmas on contrasting the need of Maintaining Lies to the terrorists, while providing one’s Loyalty to the NSA. As both patterns have with maintaining and developing Sam’s relations with the to group, and the game ends if either the NSA or terrorists’ trust reach zero, Social Maintenance has an important role in game.

The NPCs in the game have an Awareness of Surroundings [29]: they react to Sam Fisher and they know where they are. If the NPCs spots Sam doing something suspicious they comment on that and trust value lowers as long as a player chooses to continue the action. The reaction of NPCs also depends on the area (Context Dependant Reactions), e.g., some areas in the terrorist’s base are off-limits to Sam and trespassing quickly deteriorates the terrorist’s trust of Sam if the terrorists spot him.

Given the overarching goal of destroying the terrorist network, removing individual actors from it could be a possible expansion of the game. Besides stealthy assassination, this could has the social aspects of fostering Internal Rivalry by False Accusations, to developing relations to be able to ask for Favors against the terrorist organization, or even to Gain Allies by turning individual terrorists into Traitors against the organization. Focusing instead on establishing social relationships, the current game structure can be tuned to highlight social interaction by focusing on the pattern Traitor. By adding minor gameplay goals regarding Gossiping, Eavesdropping, and adhering to the Social Norms the player could have advantages in achieving goals of Maintaining Lies and providing the correct Outspoken Support. Depending on how much interaction this would require, maintaining the trust of the JBA members could be a secondary goal of Social Maintenance in addition to completing the missions given by terrorists and the NSA.

4.8 Façade
Façade [36] is a discussion-based game based on a dysfunctional marriage between two NPCs: Grace and Trip. Players influence the outcome visit to the couple through movement, simple interaction with objects, and free text input. Both NPCs in the game are Competing for Attention from the player to Gain Allies, providing the player with a social dilemma. This provides an implicit Internal Conflict to the player: how to be a friend to both.

Players can take sides by choosing whom to talk or by showing attitudes through modulating their personal distance, something the NPCs also do, but may also come with False Accusations to provoke responses. NPCs both directly involve players into the conflict by explicitly Requesting Support and indirectly when the player is Eavesdropping. NPCs react to comments made by the player to the other NPC, so NPCs are also Eavesdropping. In general, Grace and Trip have good Awareness of Surroundings. In addition successes of the other affects Grace and Trips moods, which leads to a pattern Others Fortune affect own Mood.

The initial setup of the game can give the impression that both Grace and Trip follow the pattern Either You are with Me or You are against Me. Hence, the player can try to escalate or calm down the conflict between Grace and Trip depending on the player’s goals of the experience. Trip and Grace react more and more negatively to prolonged silence from the player, so a form of Social Maintenance is required to not be thrown out of the apartment. The player can gain information of Grace and Trip’s internal states from their displays of Emotional Attachment [29], e.g., by the facial expression, personal distance, gestures, and comments. The NPCs have Context Depended Reactions and they expect the player to obey Social Norms like turn taking in discussion.

Façade is simultaneously the example focusing the most on social interaction and being the least game-like: the latter for not having clearly expressed goals and by effectively hiding the game state behind the performances of the NPCs. If is therefore difficult to state which patterns do not occur in the game as much lies in the eye of the beholder, e.g., the player can become an Outcast by being thrown out of the apartment after taking as a personal goal to be a Traitor within the game’s context. Unlike the other examples, increasing the gameplay aspects of social interaction in Façade does not require more gameplay mechanics, but rather making these more visible to players, something that seem to go against the design intensions of the designers as they “examine issues of procedural authorship using the interactive drama Façade” [36].

5. DISCUSSIONS
In this section we briefly discuss findings not directly linked to individual examples. These findings are groups into those pertaining to relations between identified patterns and already document pattern and structures, general design implications, and future work.

5.1 Patterns
Given the complexity of the game examples, and the difficulty of gaining information on the inner working of them, not all specific cases in each game may have been identified and thus the identified design patterns relate to the typical gameplay of each game. That being said, the identified patterns provide starting points for several ways of expanding gameplay generally regarding social networks.

Although most patterns have been positively identified in the examples-some were easy to hypothesize about, but more difficult to find. One such example is Information Passing, which would probably require modeling information as discrete entities in the system, possibly based upon the ideas of memes, to be part of structured gameplay. If games did model this, however, players would have not only to consider what goals they have with an ongoing conversation, but also the long-term effects of that the conversations can have. The gameplay effects of this in games such as Crusader King [37] and Splinter Cell: Double Agent [47] include possibilities to gain access to secrets through Gossip, being able to perform Brokering between hostile parties, finding ideal Match-Making solutions, Maintaining Lies to provide alternative identities, and causing conflicts through False Accusations.

Social Maintenance is another pattern primarily found as hypothetical redesigns of the examples. Two main requirements exist for this to be used in games: First, actions that can carry the role of maintain social networks exist in the game and that these actions are seen as meaningful for gameplay. This requires that players have goals attached to the effect of the actions. Second, that the actors in the social network are sufficiently complex in the social model that the social relations do not become trivial. If these both requirements are met by the design of the game system, gameplay can focus upon social interaction rather than physical or problem-solving activities, e.g., Triangular Drama or Threat from Outside, which have been posited as gameplay possibilities by Lankoski [28]. A side effect of this would be to enrich the interactions with Hierarchical Fraction, so that games such as Oblivion [8], Splinter Cell: Double Agent [47], and Canis Canem Edit [39] can support gameplay based upon Competition for Attention, Favor, Loyalty, and Internal Rivalry.

Several of the patterns identified, e.g., in Façade, have arguably more to do with the design of believable characters on the face-to-face level of social interaction that the large scale social network. Finding such patterns, which could be said to belong to our previous study on individual NPCs, are normal since changing the frame of references can highlight characteristics of a design that under previous circumstances have been so. Notably, the conflict in Façade [36] requires many patterns that relates to believability of NPCs to make gameplay possible and the player’s choices understandable-or patterns give means to players to predict NPCs reactions to their actions based on their previous (everyday) experiences.

5.2 Designing Social Networks
As for all design patterns, the patterns presented here should be seen as tools; their use does not suit to every design, as each game has their own gameplay goals. However, the patterns identified highlight some game structures and possible design solution that can be used in games in which the design goal has focus on social conflicts or social structures have an important role. Why these kinds of game structures have not been used more, or made into the core gameplay mechanics, is of course impossible to say with certainty, but a hypothesis is that focusing on social relations seems incompatible with simulations. Those focusing on depicting a developing social relation within a game may shun away from creating a simulation that carries the development of relations as this can cause unexpected, or simply boring, outcomes. Likewise, those focusing on simulation may shun away from social relations as they may seem simplistic or mechanistic compared to those describe through theme.

Looking at the gameplay structure underlying the examined games, most of the examples have models similar to those described in the SNA than those described in the ANT. This may be due to the ANT descriptions requiring a greater focus on the dynamics of social networks than the SNA, but this implies that an ANT-based approach could be more suitable for introducing more simulation into the social networks of NPCs.

5.3 Future Work
The gameplay design patterns identified above are not presented in such a detail as in the original collection [9]. This remains as a future work mainly for two reasons: 1) a desire to include more case studies in the paper required to reduce the details, and 2) more comprehensive studies of each individual pattern, including design experiments, is needed to describe aspects such as “implications of using a pattern” in detail. In addition, many more case studies should be added, with the Sims [35] or Japanese dating sim games as possible candidates.

Massively multiplayer games were excluded as player relations typically have more impact on the game experience of those games than character relations. There is, however, no reason to believe that the identified gameplay design patterns could not be applied to these games to strengthen the social network of the characters. In fact, some of the applications helping player guilds organize themselves in World of Warcraft [10] have been describe having the potential to “become powerful social actors” in the right contexts [43]. Furthermore, combining the functionality of such applications and NPCs could provide systems that could, like human players, participate in both social networks. A precursor of this phenomenon can be found in conversational agents developed for MUDs (e.g., Julia [46]), which was designed to support social networks through keeping track of player presences and passing messages [21]. Thus, a possible design challenge for creating social NPCs in massively multiplayer games may be to make them into functional guild members.

6. CONCLUSIONS
In this paper, we have described gameplay design patterns that offer design possibilities in games regarding social interactions between NPCs. The patterns have been identified from games through the lenses of scientific models of believable characters and social networks, and shown how these patterns provide new gameplay possibilities through hypothetical redesigns of the examined games. The patterns indicate that games are modeled more after the structure-oriented social interaction analysis than interaction-oriented actor-network theory. Therefore, looking at the latter could provide new avenues for novel gameplay designs.

Many new patterns identified seem to depend on the patterns we have presented while focusing non-player characters. This points towards a requirement, besides that of narrative believability, that both social conflicts and individual structures of a character must be believable in order for the gameplay experience regarding characters to be believable as a whole.

7. ENDNOTES

1. See for example [43, 46] for player-centered studies.
2. We do not imply that games are drama or stories, but simply that there are applicable similarities of designing conflicts for the both.
3. Egri is discussing only the play-writing, not the whole design process from the play-writing to staging, casting and directing.
4. Björk and Holopainen [9] point out that co-operation can be archieved by, e.g., using the pattern Mutual Goals.
5. Moreover, Chess pieces are not likely to be categorized as characters or persons, as the pieces often lack most qualities that will encourage that kind of conceptualization (such as human body, face, affective expressions, or intentional states).
6. In Ico, Outcast is mainly a device for narration, but idea is usable as a gameplay design pattern.
7. This explanation mixes concepts from the SNA and ANT, which would not be acceptable within social sciences. We motivate this in this context by the condensed description made possible.

8. REFERENCES
[1] Anonymous Facebook | Welcome to Facebook! 2007, 9/19/2007 (www.facebook.com).
[2] Anonymous Welcome to Flickr! 2007, 9/19/2007 (www.flickr.com).
[3] Alexander C., Ishikawa S., Silverstein M., Jacobson M., Fiksdahl-King I. and Angel S. A pattern language. New York (NY), Oxford University Press, 1977.
[4] ALTAR interactive. UFO Afterlife. Windows (2007).
[5] ArenaNet I. Guild Wars. ( 2005).
[6] Aristotle. Poetics. Penguin Books, London, 1996.
[7] Berman R. A. Fade in: The Screenwriting Process. Michael Wiese Productions, Studio City, 1997.
[8] Bethesda Softworks LLC. The Elder Scrolls IV: Oblivion. (2006).
[9] Björk S. & Holopainen, J. Patterns in Game Design. Charles River Media, Hingham, 2005.
[10] Blizzard Entertainment Inc. World of Warcraft. (2004).
[11] CCP Games. Eve Online. (2003).
[12] Church D. Gamasutra – Features “Formal Abstract Design Tools”. 2007, 9/17/2007 (www.gamasutra.com/features/ 19990716/design_tools_01.htm).
[13] Costikyan, G. I Have No Words & I Must Design. Interactive Fantacy, #2 (1994).
[14] Currie G. Image and Mind: Film, Philosophy, and Cognitive Science. Cambridge University Press, Cambridge, 1995.
[15] Cypher, M. & Richardson I. An actor-network approach to games and virtual environments. (2006), 254-259.
[16] Davidsson O., Peitz J. & Björk, S. Game Design Patterns for Mobile Games, 2004. Report to Nokia Research. Available at procyon.lunarpages.com/~gamed3/docs/Game_Design_Patterns_for_Mobile_Games.pdf.
[17] Egri, L. The Art of Dramatic Writing. Simon & Schuster, New York, 1960.
[18] Aarseth, E., Marie S. S. & Lise, S. A multidimensional typology of games. In Level Up Conference Proceedings (Uthrecht, November, 2003). Available at http://www.digra.org/dl/db/05163.52481.pdf.
[19] Field, S. Four Screenplays. Dell Publishing, New York, 1994.
[20] Firaxis Games. Civilization IV. Windows (2005).
[21] Foner, L. N. Entertaining agents: a sociological case study. (1997), 122-129.
[22] id Software Inc. Quake 4. (2005).
[23] Ion Storm. Thief Deadly Shadows. Xbox (2005).
[24] Isbister K. Better Game Characters by Design. Morgan Kaufmann Publishers, Amsterdam, 2006.
[25] Krawczyk M. & Novak, J. Game Story & Character Development. Thompson Delmar Learning, 2006.
[26] Kumar R., Novak J. and Tomkins A. Structure and evolution of online social networks. (2006), 611-617.
[27] Lankoski P. Goals, Affects, and Empathy in Games. In Philosophy of Computer Games, 2007.
[28] Lankoski P. Character Design Fundamentals for Role-Playing Games. In Montola M. and Stenros J. eds. Beyond Role and Play – tools, toys, and theory for harnessing the imagination. Ropecon ry, Helsinki, 2004, 139-148.
[29] Lankoski P. and Björk S. Gameplay Design Patterns for Believable Non-Player Characters. In Situated Play: Proceedings of DiGRA 2007 Conference, 2007. Available at http://www.digra.org/dl/db/07315.46085.pdf.
[30] Lankoski P. and Heliö S. Approaches to Computer Game Design: Characters and Conflict. In Mäyrä F. ed. CGDC Conference Proceedings. (Tampere, June 6-8). Tampere University Press, Tampere, 2002, 311-321. Available at http://www.digra.org/dl/db/05097.01201.pdf.
[31] Lankoski P., Heliö S. and Ekman I. Characters in Computer Games: Toward Understanding Interpretation and Design. In Level Up: Digital Games Research Conference, 2003. Available at http://www.digra.org/dl/db/05087.10012.pdf.
[32] Konzack, L. Computer Game Criticism: A Method for Computer Game Analysis. ( 2002). Available at http://www.digra.org/dl/db/05164.32231.pdf.
[33] Latour B. Reassembling the social an introduction to actor-network-theory. Oxford University Press, Oxford, 2005.
[34] Looking Glass Studios. Thief II: The Metal Age. PC (2000).
[35] Maxis. Sims. (2000).
[36] Mateas, M. & Stern, A. Procedural Authorship: A Case-Study Of the Interactive Drama Façade. Digital Arts and Culture (DAC), Copenhagen, November 2005
[37] Paradox Entertainment. Crusade King. Windows (2004).
[38] Peitz J. and Björk S. Understanding Pervasive Games through Gameplay Design Patterns. In Situated Play: Proceedings of DiGRA 2007 Conference, 2007. Available at http://www.digra.org/dl/db/07311.47471.pdf.
[39] Rockstar Vancouver. Canis Canem Edit (aka Bully). (2006).
[40] Rouse R. Game Design: Theory and Practice. Wordware, Plano, 2005.
[41] Sheldon L. Character Development and Storytelling for Games. Thomson Course Technology, Boston, 2004.
[42] Sony Computer Entertainment. Ico. PS 2 (2002).
[43] Taylor T. L. Does WoW Change Everything?: How a PvP Server, Multinational Player Base, and Surveillance Mod Scene Caused Me Pause. Games and Culture, 1, 4 (October 1, 2006 2006), 318-337.
[44] Team NINJA. Dead or Alive 3. (2001).
[45] Team Silent. Silent Hill 3. PS 2 (2003).
[46] Turkle S. Life on the Screen: Identity in the Age of the Internet. Touchstone, New York, 1997.
[47] Ubisoft Shanghai Studios. Tom Clancy’s Splinter Cell: Double Agent. PS 3 (2007).
[48] Valve Software. Half-Life. (2001).
[49] Wasserman S. & Faust K. Social Network Analysis. Cambridge, Cambridge University Press, 1994.
[50] Zagal J., Mateas M., Fernandez-Vara C., Hochhalter B. & Lichti N. Towards an Ontological Language for Game Analysis. (2005). Available at http://www.digra.org/dl/db/06276.09313.pdf.

APPENDIX: GAMEPLAY DESIGN PATTERNS INTRODUCED
Actions Have Social Consequences

Action by s a character influence on how others perceive and how they act towards the character.

Binding Promises

A binding agreement that is either a a goal (failure to keep a promise is penalized) or a rule (being impossible to break after commitment).

Brokering

Enabling interaction between to not directly connected actors in a social network by acting as an intermediate.

Competing for Attention

The competition between several characters to get attention of one character.

Context Dependent Reaction

NPC reacts to events and objects (including other characters) depends on the context (space, other objects in that space, and past events) in which character is.

Eavesdropping

A possibility to gain information through listening other characters talking.

Either You are with Me or against Me

A demand for a support where disagreeing threatens rejection from a network.

False Accusations

Untrue statements that can be made to affect the social network.

Favor

The promise of a future action against a Fraction one is member of due to social relations to someone not belonging to that Fraction.

Fraction

A specific social network where membership is defined by what actions are allowed, disallowed, and required.

Gain Allies

The goal to add new members to a social network defined as an alliance.

Gossip

Two characters passing information between each other mainly for informing a player about various things. The passed information does not have direct influence to gameplay.

Guide and Protect

A character needs to guide another character from place A to place B and protect that character during the journey.

Hierarchical Fraction

A group that have hierarchical power structure, e.g., a clan, family, or police

Information Passing

The passing, from a character to another, of information having influence on the gameplay.

Internal Conflict

Having a set of desirable goals where progress in one typically makes others more difficult.

Internal Rivalry

Being an enemy with a character within the same fraction

Linked Destinies

Two or more characters share the same persistent goal.

Loyalty

The continued goal of being part of a social network.

Maintaining Lies

The continuous goal of maintaining a condition in the social network created through a lie.

Match-Making

The goal of creating a relation between two characters.

Memory of Important Events

NPC keeps track of events that have impact to it and the event influence its behavior.

My Enemy’s Enemy is my Friend

A common enemy aligns two characters with each other.

Others Fortune Affects own Mood

Noticed events affecting parts of one’s social network cause secondary effects on oneself.

Outcast

A character that is thrown out from a group so that the group members, e.g., ignore or attack the outcast character if they meet.

Outspoken Support

Explicit declaration to support another actor with future actions.

Requesting Support

Being able to ask an actor in one’s social network for support based upon one’s relation.

Social Gatekeeper

The arbiter of membership for a specific social network.

Social Maintenance

Perform actions to redefine and refine the relation to a group.

Social Norm

A rule; breaking the rule will influence behavior of NPCs like changing attitude to more negative towards breaker of the rule.

Traitor

Traitor requires pretending to belong to a fraction while acting against it.