=Paper=
{{Paper
|id=Vol-2282/EXAG_123
|storemode=property
|title=Addressing the Elephant in the Room: Opinionated Virtual Characters
|pdfUrl=https://ceur-ws.org/Vol-2282/EXAG_123.pdf
|volume=Vol-2282
|authors=Sasha Azad,Chris Martens
|dblpUrl=https://dblp.org/rec/conf/aiide/AzadM18
}}
==Addressing the Elephant in the Room: Opinionated Virtual Characters==
https://ceur-ws.org/Vol-2282/EXAG_123.pdf
Addressing the Elephant in the Room: Opinionated Virtual Characters
Sasha Azad, Chris Martens
Principles of Expressive Machines (POEM) Lab, North Carolina State University
{sasha.azad, crmarten} @ncsu.edu
Abstract stract knowledge base for the characters that groups various
objects of discussion under overarching topics, tracks the
A number of recent models for empirically-grounded social
simulation have emerged recently from games and interac- sources from which they originate with their inherent bias or
tive narrative research, generally exploring models of trust, ratings, and allow the non-player characters (NPCs) to form
emotion, and social graph changes that occur in the process opinions based on individual preferences or cultural norms.
of inter-character interactions. However, these models so far Our system can track the spread of influence (adverse or oth-
failed to provide realistic models of opinion change and pre- erwise) and change in the views of the participant NPCs.
disposition to new knowledge. Equipped with such a notion,
these emergent social simulations can express both real and Finally, we demonstrate our system with a case study that
fictionalized depictions of modern phenomena like adverse showcases a series of conversations where virtual charac-
media influence, the spread of “fake news,” and the polariza- ters discuss current political news from the U.S., exchange
tion of ideological sects. We present a preliminary computa- their views on individual news articles or issues of interest,
tional investigation into modeling opinion change in virtual and reevaluate their political ideologies and affiliations over
characters with this goal in mind. time. For instance, an NPC growing up in a more liberal so-
ciety may eschew conservative ideals, and have a low opin-
Introduction ion of the same. Our simulated NPCs are aware of the dif-
ference in their internal attitude on a topic of discourse as
Humans are rational and emotional beings. Their social sys- well as the public opinion shared by other NPCs during their
tems are complex and contextual. Understanding and simu- interactions. These differences can lead to the NPCs chang-
lating humans with virtual characters requires reasoning not ing their attitudes over time or expressing opinions different
just about observable social network graphs or social inter- from their attitudes to conform to the society they reside in
actions, but also about geography, economics, and increas- over time.
ingly, online participation and discourse. However, these
simulated models typically do not account for some of the We posit that the holding of these opinions on the various
most important features of social networks, namely that of topics could lead to the virtual character having access to
the social dynamics of opinion change. choices and interactions that would otherwise need to be au-
In this paper, we describe a simulation we have designed thored in the character’s preferences or goals. For instance,
for a society of virtual characters that can discuss and ex- an NPC growing up in a country with strict gun control,
change their views amongst one another. We define a char- or that holds an unfavorable opinion of gun ownership may
acter’s view on the topic as a combination of their internal, choose never to buy a gun.
personal attitude on a topic of discussion and their externally We imagine that in the future our opinion model could
expressed opinion. We allow for conversationalists to influ- be used to evaluate how a virtual society would integrate
ence each others’ opinions based on existing literature from and accept new additions with new members learning of
the social science regarding group conformity and accep- the views and opinions of the society while bringing with
tance, as well as by the strength of the overall public opin- them new ideas and concepts from their own culture. Simi-
ion. We describe a conversational model that allows virtual larly, opinion modeling for virtual characters could be used
characters to subscribe to information sources based on the to study the spread of debatable ethical or moral influence
source biases and opinions, share new information with one and media bias. Characters could choose to accede to peer
another, and form and exchange their opinions on the vari- pressure (from the media or society) and change their be-
ous issues at hand. haviors in order to feel a mix of both private acceptance
For virtual characters to be socially adept and add to the (that they are acting based on their views) and public con-
experience of the player, they must have a sizable expres- formity (to gain acceptance by the group). We believe the
sive range of conversational repertoire. We advance an ab- behaviors resulting from virtual characters modeled by this
system would be more believable and improve a player’s in-
teractive experience.
Related Work works previously mentioned are that our agents can reevalu-
Prior work discusses how designing for richer social be- ate their biases or changes over time by subscribing to new
haviors and interactions amongst virtual characters improve opinion pools from their peers or other sources of informa-
the believability of the character and the player’s interac- tion. We hope our model allows for a more natural conversa-
tive experience with the system (Afonso and Prada 2008; tion flow, with agents advancing and modifying their opin-
Swartout et al. 2006). Vinciarelli et al. describe the so- ions over time. We hope our system will add to the believ-
cial signals as “accounting for our attitudes towards other ability and behaviors generated by these works by provid-
participants in the current social context.” (Vinciarelli et ing further motivation for character relationships and inter-
al. 2008). Researchers have approached NPC social net- actions.
working through simulation of interactions between a col-
lection of NPCs that are reactive, appear intelligent, and Background
motivated (Riedl and Stern 2006; Mateas and Stern 2003; Group Formation
Ryan, Mateas, and Wardrip-Fruin 2016; Samuel et al. 2016).
These systems model worlds with a society of NPCs that Group formation has been studied in depth by social sci-
have individual goals, beliefs, and desires upon which they entists, historians, and psychologists to understand how hu-
act. mans respond to group (or societal) archetypes and opinions.
Research has been conducted on social rules and prac- When modeling group conversations, the physical or virtual
tices that virtual agents in the system must conform to as space where conversationalists congregate can be used to
designed by authors and designers (Mateas and Stern 2003). contextualize the interaction, allowing us to incorporate the
In Versu (Evans and Short 2014), we see the virtual char- history, physical affordances, or cultural significance of the
acters that interact with one another using the notion of geographic location or the topic in question. Merely read-
common “social practices” and templates. Characters un- ing the news enables one to gain a perspective of humans
dertake interactions based on their desires and goals, with forming groups to support various issues. These could be
social practices authored to be agnostic about which char- geographic groups, with articles describing how the Scot-
acters are assigned to the roles undertaken. Similarly with tish voted to “overwhelmingly remain” in the Brexit vote;
social systems such as CiF, in Prom Week, McCoy et al. or political ideology groups, with reports on Democrats dis-
constructs a social physics architecture model to decide cussing immigration resolutions; groups based on shared in-
how NPCs behave in a variety of different social scenar- terests, with news on Whovians that approve or condone rep-
ios rich with the potential for drama (McCoy et al. 2011a; resentation of women in Doctor Who (Jowett 2014); or by
2011b). Our system aims to add to the richness and diver- grouping an occupation, with articles describing how Tech
sity of the virtual character’s roles, the interactions they un- executives are contrite about election meddling. Latour dis-
dertake with the player and one another, and the charac- cusses how individuals relating to one group or another is an
ter archetypes generated. Our approach varies from these ongoing process made up of uncertain, fragile, controversial
social-physics scenarios. Virtual characters consider them- and ever-shifting ties (Latour 2005).
selves belonging to a new group based on their recognition
of their internal attitudes at a given moment corresponding Self Perception Agents for Opinion Dynamics
with the opinions of the society around them. These groups We review the problem of simulating agents capable of con-
could now form their own social rules over time as inter- versing and sharing opinions with one another. We model
actions that go against the group’s values would be looked the change in the opinions of the agents based on the model
upon unfavorably by its members. We believe this would re- proposed by Wang, Huang, and Sun in their 2014 paper
duce the authoring burden of the social rules or templates (Wang, Huang, and Sun 2014). Individual agents can in-
(Evans and Short 2014; McCoy et al. 2011b), allowing for fluence each other’s views and construct their self-opinions
interesting emergent gameplay. over the course of multiple interactions with one another.
Other research explores the simulation of conversations Agents are modeled as individual nodes in a social net-
and influence amongst NPCs. They view how societal norms work graph. Agents may exchange opinions with other
and popular opinions could affect the behavior of gener- agents if an edge links the nodes in the graph. Wang, Huang,
ated NPCs in a multi-agent system. In Social Role Aware- and Sun defines how agents every agent’s feelings on a topic
ness (Prendinger and Ishizuka 2001), agents choose conver- is informed by an inner “attitude” towards the topic that
sational responses based on their perception of their roles cannot be perceived by other agents, an outward expressed
within the social context. A secretary addressing her man- “opinion” and the level of “uncertainty” they feel about
ager could be more polite and responsive than one address- their opinion. Agents may adjust their internal attitudes or
ing an aspirant visiting the office. PyschSim (Pynadath and express modified opinions from their attitudes, on hearing
Marsella 2005) models influence amongst group members the opinion of other agents (Wang, Huang, and Sun 2014).
by examining how participants in a conversation view their Due to space restrictions, we refer readers to the Wang,
relationships with one another and their beliefs and moti- Huang, and Sun (2014) paper, and the Asch (1955) paper
vations about the world. Other work has virtual characters for the details on their experiment. We recognize that the
sharing their knowledge or gossiping about the world with threshold values and model evaluated in the Wang, Huang,
one another with their bias (Evans and Short 2014). The and Sun paper may not exactly conform to an exhaustive
most significant differences between our approach and the list of objects of discussion or topics of discourse. However,
Topics Objects of Discussion Source Rating
Political Issues e.g. Immigration, Gun Control Individual news articles Online or Print Media Political Bias or Affiliation
Political Issues e.g. Immigration, Gun Control Political candidates Articles, Interviews, Candidate Rally Approval Rating
Research Topics e.g. AI, Games Conference Papers Journals, Conference Proceedings Journal or Conference Rankings
Film Genres e.g. Horror, Sci-Fi Movies Movie Studios Rotten Tomatoes ratings
Table 1: Examples showing how discussions can be simulated on various datasets using the proposed knowledge model
their proposed agent model combines normative social in- obtained from a Source. The Source and the Object of Dis-
fluences with a continuous dynamics model in a novel ap- cussion are associated with a Rating. Multiple objects of dis-
proach. Our objective is to extend these current theories of cussion can be clustered to form a Topic.
dynamic opinion modeling research to the narrative intelli- A major contribution of our paper is that this model of the
gence community with the goal to simulate virtual societies knowledge base can be used for a large variety of datasets
capable of exploring complex issues of politics, religion, or while affording the same discussion and opinion modeling.
even simply movie ratings. For instance, simulating debates among NPCs about cur-
Towards this goal, our contribution builds on that of rent news articles clustered by political issues and ranked
Wang, Huang, and Sun’s in the following ways: by their bias. Similarly, we could use our model to discuss
• Prior work fails to model the complex and ever-changing the merits of various journal articles clustered together by
social relationships between conversationalists. The au- research topics and ranked by journal rankings or have au-
thors assume a grid-based society where the same neigh- dience members discuss their movie preferences clustered by
boring agents surround an individual throughout their movie genres and ranked by their Rotten Tomatoes rankings.
simulation. Our method proposes a more utilitarian defini- Some datasets considered during the design phase have been
tion of social relationships where NPCs with differing or highlighted in Table 1.
similar opinions could change relationships over time, al-
lowing their old social connections to dissolve over time. Rating of the Information
• Instead of a single object of discussion, we allow charac- We define the rating as the value of the information learned
ters to discuss a variety of information clustered by topics. by the NPC in the system. This rating could represent ei-
This allows for relationships where characters that agree ther (1) the personal judgment or favor associated with the
over a few views but disagree over others to change their presentation of the information, or (2) a measure of the im-
affinity for one another over time. partiality of the unit of information. The rating is the merit
or value of the information that is debated by the NPCs in
• We allow for the simulation to add new concepts and top- our system. For instance, this could represent the ratings for
ics over time. We believe this could lead to virtual charac- a movie, reviews for a paper, or a bias rating for a media
ters to extend their knowledge base while retaining their source.
individual views on existing knowledge.
Topics
Goals We describe topics as a clustering of information regarding
We list our goals for the project as follows: a specific subject, or field of information. A specific infor-
• Account for bias in characters where agents may have a mation unit can be a part of multiple topics at the same time.
predisposition to adopt a specific view from prior experi- For instance, a discussion of procedural content generation
ence. could belong to the topics of both artificial intelligence or
game design. A virtual character may periodically reevalu-
• Account for bias in the information. Information and ate his rating of a topic by considering the rating of all the
sources producing information may have an inherently bi- objects of discussion within a topic.
ased perspective.
• Ability for characters with similar opinions to form rela- Object of Discussion
tionships, and allow ad-hoc groups developing during so- This single unit of information forms the basis of our dis-
cial interactions to discuss their opinions on various top- cussion model. While interacting with one another, virtual
ics. characters search through their knowledge base and conver-
• Be able to use the same discussion model for a variety sational repertoire, choosing a single object of discussion to
of different data sources to simulate opinion modeling on debate. An NPC that adds a new object of discussion to his
discussions. knowledge base will note the original authorial rating in-
tended to be affiliated with the information, and associate
with it their own opinions on the topic. These views could be
Our Model of Knowledge based on prior discussions of the information with conversa-
The model we use to define in-game knowledge is described tionalists that introduce the character to the information, as
as follows. For a single discussion, the participants in the well as on the character’s current view of the topic to which
discussion choose an Object of Discussion to converse on, the information belongs.
Figure 1: Case Study: Example of a topic of discourse, Russia, and some news articles associated with it, each labeled with
their own media bias (AllSides 2018)
Sources have lower confidence in their attitude if (1) information
may create information covering a wide variety of objects in their existing knowledge base inadequately back them,
of discussions and topics. Sources may also have associated (2) if contradictory opinions are presented to the agent
with them a rating, representing the expected rating of the in- with high certainty, or (3) if the agent is surrounded by a
formation they produce. NPCs may use this rating to choose society a majority of whom disagrees with him. unc is a
to subscribe or unsubscribe to these over time based on their real number in the range [0, 1].
current inclinations. For instance, an NPC studying in the • Public Compliance Threshold (pub thr): When the
computer science domain may subscribe to AAAI for peri- strength of the public opinion exceeds this value, the agent
odic information on the research in their field. will choose to comply with the public opinion to feel ac-
cepted within the community. pub thr has a default value
Modeling a Character’s Views of 0.6.
Every participant in the discussion has their own Bias and • Private Acceptance Threshold (pri thr): When the
View on the information and can express their opinions on strength of the public opinion is below this value, the
the object of discussion at hand. These elements and our agent will choose to stand by their views. The pri thr
dataset have been described in further detail below. The at- is a real number in the range [0, 1]. Professors or experts
tributes of an agent’s view are modeled based on those by on a particular topic in our simulation would have higher
Wang, Huang, and Sun. values to indicate their expertise.
• Bias: We define an agent’s bias to be the agent’s predispo-
sition to adopt a particular view on a topic in a discussion. Social Interactions and Discussions
This bias is informed by either (1) the agent’s views inher- During initial generation, the NPC population is assigned
ited from their parents or (2) a mean of their views on all random cultural biases on topics in the knowledge base.
objects of discussion under the said topic or (3) the initial From this stage on, children inherit as bias the mean of their
bias they learn from the conversationalists when the topic parent’s biases (i.e., representing “nature”) while tending to
was added to their knowledge base during a discussion. agree more with one parent or the other about individual is-
• Attitude (att): the agent’s private views on a specific is- sues. However, the children may change these opinions over
sue. Attitude is a real number in the range [−1, 1], and time (i.e., representing “nurture”) over the course of several
represents an evaluation of the object of discussion. social interactions between the agents.
• Opinion (op): an agent’s outwardly expressed or shared
Discussion Method
views on a specific issue. Like attitude, opinion is a real
number in the range [−1, 1] and reveals the agent’s opin- We begin by clustering the expressed opinions of all par-
ion on the object of discussion to the other dialogists. ticipants of the conversation using the Jenks optimization
There may be a discrepancy in the attitudes and opinions method (Jenks 1967) and choose the grouping with the low-
of the character since a character may not represent their est square error. The number of opinion groups formed in-
attitudes accurately to participants. A human example of dicates whether a public consensus on the matter has devel-
the situation where this is apparent can be seen in exam- oped and the presence of normative social influence (or peer
ples of an employee in conversation with his managers pressure). The fewer the groups that form, the more likely it
who choose not to express his disagreement to avoid be- is that an agent who maintains their views contrary to public
ing punished. opinion will feel rejected.
• Uncertainty (unc): a measure of an agent’s confidence in Public Opinion Formed If the agent has high uncertainty
their view. The higher the uncertainty, the more likely the (i.e., agent.unc > 0.8), they are more likely to accept the
agent is to change his mind or accept other perspectives. views of their fellow dialogists. We assign these agents the
As an example, an NPC may express opinions about the attitude and opinion equal to the mean of the largest group
legality of abortion in their town. However, the agent may in the consensus.
If the agent has low uncertainty (i.e., agent.unc ≤ 0.8), Case Study: Political Ideologies
we find the largest clustered opinion group with views clos- In this divisive age, it is difficult (yet unavoidable) to dis-
est to that of the agent. We then calculate the public opinion cuss current political events with family or friends. APIs for
strength for the selected group and decide if an agent’s atti- major media sources are available with access to news arti-
tudes or opinions are affected. The public opinion strength cles on various topics. As a case study, our simulation uses
(op str) is calculated as described by Wang, Huang, and Sun a corpus of news articles (AllSides 2018), grouped by their
by normalizing and finding the mean of the sum of the fol- political issues. Characters are initially assigned political af-
lowing factors: filiations and biases. The rating system, in this case study,
• The size (fa ) of the group. The larger the group, the is based on that of the U.S political-ideological system. For
stronger the public opinion. the simulation, in the beginning, characters are subscribed to
sources that confirm their political bias. For instance, a Cen-
0,if xa ≤ 1
trist NPC may subscribe to the Associated Press as a news
fa = xa /10,if 1 < xa ≤ 10 source.
1,if xa > 10 News Source AllSides Media Bias Ranking
New York Daily News Left
• The homogeneity (fb ) in the opinion of the group defining New York Times Lean Left
if the group come to a consensus
Associated Press Center
fb = 1/(1 + e24xb −6 ) Boston Herald Lean Right
Fox News Editorial Right
• The discrepancies (fc ) in the agent’s opinion and attitude.
Table 2: Examples of the AllSides Media Bias Rankings ob-
fc = 1/(1 + e−12xc +6 ) tained for NPC subscriptions to media sources
Next, the agent measures their own uncertainty with the
strength of the public opinion by calculating two threshold • Rating: We use media bias as our rating and associate
values, th1 = 1 − agent.unc and th2 = max(0.6, th1 ). with each bias a value as follows: Left(−1.0), Lean
Left(−0.5), Center(0.0), Lean Right(0.5), Right(1.0).
• Low Opinion Strength (op str < th1 ): If the opinion The bias ratings in our dataset are obtained from All-
strength is too weak, the conversationalist does not change Sides using a combination of blind bias surveys, editorial
their mind, recognizing the discrepancy between their in- reviews, third-party research, independent research, and
ternal attitudes and ideas and those of the group. community votes to calculate media bias of the informa-
• Moderate Opinion Strength (th1 ≤ op str < th2 ): tion (AllSides 2018) as can be seen in Table. 2.
– Members with a low uncertainty find the opinion • Topics: We use U.S. Political Issues such as Civil Rights,
strength of their group strong enough to modify their Immigration, Healthcare, Free Speech, Gun Control, and
opinions to the mean of the group. Agents then find Abortion (AllSides 2018) each with an equal number of
their internal attitudes, and their expressed behaviors articles representing every bias.
are inconsistent, and so change their attitudes to match. • Objects of Discussions: Individual news articles are our
In this case, agents believe that the change in their objects of information. A character will note the original
views are a natural and expected evolution, and do not authorial bias of the information and associate with it their
realize they are bending to public opinion. views based on their current attitude towards to the topic,
– Agents with large uncertainty realize that they are con- their overall political affiliations, and their discussions on
ceding the discussion, and bending to public opinion. the article with other conversationalists.
They change their external opinions and internal atti- • Sources: Sources are media sources that publish articles
tudes to match. on a wide variety of issues. NPCs may subscribe or unsub-
• High Opinion Strength (op str ≥ th2 ): The agent realizes scribe to these over time based on their current political
the strength of the opinion. In this case, the agent may inclinations. Overall Political Affiliation: is a weighted
choose to conform to the public opinion with their out- average of the agent’s attitudes of all topics in the agent’s
wardly expressed views and change their opinion to the knowledge base (ranked by an agent’s priorities). For in-
mean of the group. However, they do not change their in- stance, a simple measure how Liberal or Conservative a
ner attitudes, and in the absence of external pressure will person is could be expressed as a weighted average of
revert to their attitudes. their attitudes on the topics of gun control, abortion, ho-
mosexuality, tax reform, and so on.
No Public Opinion Formed If public opinion has not
formed yet, then after clustering the agent finds the clus- Social Interactions and Discussions
ter of opinions with the opinions most similar to that of the We simulate a town where characters can interact with one
NPC. The NPC modifies their opinion to the mean of the another. Our preliminary experiment allows for two types of
cluster and their internal attitudes on the information being organizations, Schools, and Businesses, to facilitate group
discussed. discussion.
Figure 2: Sample discussion outcome involving four virtual characters on a news article from NYTimes.
Schools Schools choose a subset of topics from the world his views after reading the article. As such his uncertainty
to teach their students. Professors are modeled to have a low on the subject reduces, but his views stay the same. Vickie,
uncertainty value regarding their views. This in combination whose political views were aligned Right (att = 1.0) before
with the fact that they are regarded as authority figures in the the discussion changes her views slightly over the course of
simulation implies that a student is more likely to adopt their discussion (att = 0.948) and finds herself a little more un-
views. In Fig. one can see the knowledge base of a recent certain about her view on the article.
graduate after he reevaluates his views on Immigration. However, since the internal attitudes of all four partici-
Businesses NPCs may apply to work at open positions in pants on the article and the topic of Immigration (not shown
various local businesses. The application to these positions in Fig. 2) remain the same, their overall Political Affiliations
is based on the knowledge as well as the opinions an NPC do not change. . . yet.
acquires over time. For instance, an NPC may be required
to have specific views on the topic of abortion as a qualifi- Future Work
cation to work at a local hospital that matches those of their
colleagues. In the future, we hope to be able to simulate cultural or re-
gional opinion preferences by associating opinions with lo-
cations at the beginning of the population generation. For in-
stance, NPCs originating in Japan may have a bias for highly
restrictive gun control laws. Additionally, some articles or
knowledge may be regional, prompting stronger opinions
among members directly associated with an issue or allow-
ing for the modeling of an exchange of cultural knowledge.
We believe that NPCs with an ability to share and exchange
opinions could lead to the generation of a virtual society
that has more diversity in beliefs and preferences. A greater
awareness regarding the variety of opinions that exist could
be extrapolated in the future causing them to revisit the cer-
Figure 3: The political news and opinions knowledge base tainty of their opinions on other topics of discourse; thus
for a character that graduated from school allowing for more realistic machine enculturation.
We aim to enable the creation of virtual communities shar-
Sample Discussion Outcomes : We decode in prose a typ- ing opinions that form to discuss their views. These groups
ical outcome for a discussion from our simulation as shown could then inform classes of actions available to their mem-
in Fig. 2. NPCs discuss an article titled “Room for Debate: bers. For instance, a group of students could petition to
Should ‘Birthright Citizenship’ Be Abolished” at work with reduce the school’s carbon footprint. We hypothesize the
colleagues. The article falls under the topic of Immigration spread of opinions and influences will enable us to study
and is published by the source NY Times with an original how more believable information dissemination could occur
authorial bias calculated by AllSides as Leaning Left. The in simulated populations and narratives.
duration of the discussion is 11 minutes, representing the
number of times the algorithm is run, and the views of the Acknowledgement
participants are updated.
Ruth and Suzanne learn about the article for the first time. We thank Chung-Che Hsiao from NC State University for
They choose to accept the outcome of the discussion as their his insight towards the research of this paper. We would also
opinion after applying any pre-existing bias on the topic of like to express our gratitude to Scott J. McDonald and the
Immigration. Richard, whose political views Leaned Left AllSides team for providing us with API endpoints and bias
(att = −0.5) before the discussion, is more convinced about rated media articles to use in our Political News Case Study.
References a survey. In Proceedings of the 10th international confer-
Afonso, N., and Prada, R. 2008. Agents that relate: Im- ence on Multimodal interfaces, 61–68. ACM.
proving the social believability of non-player characters in Wang, S.-W.; Huang, C.-Y.; and Sun, C.-T. 2014. Modeling
role-playing games. In International Conference on Enter- self-perception agents in an opinion dynamics propagation
tainment Computing, 34–45. Springer. society. Simulation 90(3):238–248.
AllSides. 2018. Balanced news via media bias ratings for
an unbiased perspective.
Asch, S. E. 1955. Opinions and social pressure. Readings
about the social animal 193:17–26.
Evans, R., and Short, E. 2014. Versua simulationist story-
telling system. IEEE Transactions on Computational Intel-
ligence and AI in Games 6(2):113–130.
Jenks, G. F. 1967. The data model concept in statistical map-
ping. International yearbook of cartography 7:186–190.
Jowett, L. 2014. The girls who waited? female compan-
ions and gender in doctor who. Critical Studies in Television
9(1):77–94.
Latour, B. 2005. Reassembling the social: An introduction
to actor-network-theory. Oxford university press.
Mateas, M., and Stern, A. 2003. Façade: An experiment in
building a fully-realized interactive drama. In Game devel-
opers conference, volume 2.
McCoy, J.; Treanor, M.; Samuel, B.; Mateas, M.; and
Wardrip-Fruin, N. 2011a. Prom week: social physics as
gameplay. In Proceedings of the 6th International Confer-
ence on Foundations of Digital Games, 319–321. ACM.
McCoy, J.; Treanor, M.; Samuel, B.; Wardrip-Fruin, N.; and
Mateas, M. 2011b. Comme il faut: A system for authoring
playable social models. In AIIDE.
Prendinger, H., and Ishizuka, M. 2001. Social role aware-
ness in animated agents. In Proceedings of the fifth interna-
tional conference on Autonomous agents, 270–277. ACM.
Pynadath, D. V., and Marsella, S. C. 2005. Psychsim: Mod-
eling theory of mind with decision-theoretic agents. In IJ-
CAI, volume 5, 1181–1186.
Riedl, M. O., and Stern, A. 2006. Believable agents and in-
telligent scenario direction for social and cultural leadership
training. In Proceedings of the 15th Conference on Behavior
Representation in Modeling and Simulation.
Ryan, J.; Mateas, M.; and Wardrip-Fruin, N. 2016. A simple
method for evolving large character social networks. Proc.
Social Believability in Games.
Samuel, B.; Ryan, J.; Summerville, A. J.; Mateas, M.; and
Wardrip-Fruin, N. 2016. Bad news: An experiment in com-
putationally assisted performance. In International Confer-
ence on Interactive Digital Storytelling, 108–120. Springer.
Swartout, W.; Hill, R.; Gratch, J.; Johnson, W. L.; Kyri-
akakis, C.; LaBore, C.; Lindheim, R.; Marsella, S.; Miraglia,
D.; and Moore, B. 2006. Toward the holodeck: Integrating
graphics, sound, character and story. Technical report, Uni-
versity Of Southern California Marina Del Rey CA Inst For
Creative Technologies.
Vinciarelli, A.; Pantic, M.; Bourlard, H.; and Pentland, A.
2008. Social signals, their function, and automatic analysis: