Complex Character

Model for a Non Player AI Character for Interactive Narrative Discourse

Hee Holmen

Department of Communication, Business and Information Technologies, Roskilde University,

Universitetsvej 1, Roskilde, Denmark

Keywords:

AI and Creativity, Interactive Narrative, Interactive Comics, None Player Agency.

Abstract:

Non Player Characters (NPC) in Interactive Drama, Fac¸ade, are built based on the Believable Agent model.

This model is made for effectively managing character behaviour, as believability is expressed by visible

actions. Yet NPCs in Fac¸ade do not render their ’rich characters.’ The dialogues do not respond well enough

to express any complexities the characters may have. For dramatic narratives, authors in Interative Narrative

(IN) need ways to reveal complex characters. How can AI be used to build a complex character for interaction?

More importantly, how should these complexities be revealed to the reader? This paper proposes design

contexts for a complex Non Player Character (NPC) for the interactive comics framework, Cyber Comix.

1 INTRODUCTION

Interactive Narrative (IN) research has two main

problems: practical and theoretical. These problems

can be related the early ideal model of what an IN

system could be, represented in Murray’s Holodeck

(Murray, 1997). Holodeck is a ﬁctional, virtual simu-

lator from TV series Star Trek. In Holodeck, the user

is in a photo-realistic fantasy theatre where all his en-

tertainment whims are met by Holodeck, be it a plot

change or interaction with characters in the story. One

of the practical problems of Holodeck-like systems is

that the storytelling is taking place in a 3D virtual en-

vironment. This constrains IN under the monstrous

task of managing animation in real time play.

Photo-realistic, animated visualization is techni-

cally possible, yet practicality it is a bottleneck be-

cause such technology requires most of the technical

resources in a production (Aylett et al., 2011). This

practical handicap prevents individual artists from

taking part in developing IN in general, simply be-

cause the complex technology is limited to highly

educated experts in computer programming and an-

imation. If we consider two IN systems developed

after Holodeck ideal, Fac¸ade (2005, in 3D on PC)

and EmoEmma (2009, in CAVE VR), most of their

research resources are used for managing 3D anima-

tion, which is more of a representational technology.

It is not surprising that none of the prototypes are

replicated outside of the lab. The efforts to realise

the ideal model of Holodeck should not be underesti-

mated. However, we could acknowledge that IN is not

limited to a 3D animated virtual environment. This

paper investigates an alternative graphical narrative

medium that is not necessarily animated, yet suitable

for IN systems. This may invite more participation

from general authors in developing IN. Disregarding

animation may be considered drastic by some, how-

ever I argue that the boundary of storytelling is much

bigger than animated expression, and it needs to be

explored.

A theoretical problem in IN research is that the

discussion of interactivity for narrative discourse is

user-centric: how the user beneﬁts from interaction.

We know that users beneﬁt from increased presence,

sense of agency, and pleasure of navigation in sim-

ulated interaction (Murray, 1997), (Carlquist, 2013)

as well as understanding what kinds of players they

can be (Aarseth, 2007). However, we know very little

about what interaction means for the authors. How

can interaction serve the narrative discourse? Should

the author make up a story ﬁrst and translate or adapt

interactivity for the users without any artistic bene-

ﬁts? In this regard, the author has been marginalized

as someone who must give up narrative control and

is in conﬂict with the user who wants more freedom.

This might be considered a misplaced claim, as we

did not seriously consider what interaction means for

creating narrative discourses from the author’s point

of view. For example, if we look at interaction as an

expressive form for storytelling, we may have differ-

ent ideas regarding what beneﬁts the users as well.

563

Holmen H..

Complex Character - Model for a Non Player AI Character for Interactive Narrative Discourse.

DOI: 10.5220/0005283505630568

In Proceedings of the International Conference on Agents and Artiﬁcial Intelligence (ICAART-2015), pages 563-568

ISBN: 978-989-758-074-1

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

The top-down approach in present IN practice can

be summarised as: here are what users like to do in

interactive narrative, make up a story that produces

such interaction. Narratives found in computer games

seem to be just such practices. This may lead authors

to limit their creations, as well as discourage potential

authors from participating.

To address both aspects to the beneﬁt of authors,

the Cyber Comix framework is developed (Holmen,

2014a); proposing an accessible technology in graph-

ical IN, and testing a different interaction that may be

artistically interesting for a discourse. The research

question regarding this paper is: how to reveal com-

plex characters in interactive narrative discourse?

Cyber Comix uses 2D comics (non animated)

graphics for IN system, and a chatbot character to

interact with. A user is role-playing in interactive

comics, and can type in dialogues to have conversa-

tions with a Narrative Non Player Character (NNPC)

within a narrative discourse. The interaction is con-

versational, and the focus of the IN research is how

to build a knowledge base for a complex character

in AI. In this paper, I will investigate a previous IN

system, Fac¸ade, and its character model of Believ-

able Agency, and propose a Complex Agency model

to improve expressing characters’ interior world.

2 NEO ARISTOTELIAN

THEATRE V. MODERN

LITERATURE

When PC and internet technology advanced in the

1990s, New Media was noted for its interactivity

and user controlled content. 3D Computer Generated

Images (CGI) technology also presented new possi-

bilities to build photo realistic 3D virtual environ-

ments. New Media’s hot trend was a VR system, also

known as CAVE, where a user experiences and in-

teracts in 3D virtual world. With this development,

Brenda Laurel proposed an IN model based on Neo-

Aristotelian Theatre (Laurel, 1993). As New Me-

dia world can be placed in a 3D virtual environment,

interaction and storytelling are also constrained by

3D physicality. A traditional theatre, another narra-

tive medium constrained by a physical world, seemed

to be a natural place to start. Neo-Aristotelian the-

atre emphasises plots (events, actions) over charac-

ters (Mateas, 2001), (Szilas, 2004), (Murray, 1997),

(Abbott, 2008). As on the stage, characters are meant

to be shown, and acted upon, to serve the plot. Be-

haviours (actions) of characters should be believable,

not to break the natural action. Aristotelian theatre

is about what events take place, not what charac-

ters do. The kind of stories it can tell is different

from other narrative media. For example, Abbott ex-

plained that the narrative (literary) tradition has long

been shifted from plot-oriented Aristotelian to char-

acter oriented since the 19th century(Abbott, 2008,

p.130-133). Modern narratives are about revealing

complexities of characters. In the literary tradition,

the author employs focalization, free indirect style,

and inner monologue to express complexities of hu-

man nature (Ibid.p.70-79). These are also important

tools for authors in other narrative media. Film, an-

other New Media, started as ﬁlmed stage, but found

its own expressive tools to reveal complex characters:

montage, camera angle, narration, frame composition

among others. Thus, we may expect to have similar

tools of interaction to reveal complex characters in IN.

3 FAC¸ ADE: NEO-ARISTOTELIAN

THEATER AND BELIEVABLE

AGENCY

Mateas and Stern’s Fac¸ade is a landmark achievement

in IN research. It is the ﬁrst, and presently only, work-

ing prototype that is still available for general users.

Mateas based his model on Laurel and Murray’s

Neo-Aristotelian theatre (Mateas and Stern, 2002b),

(Mateas and Stern, 2003), (Mateas and Stern, 2006)

The characters’ behaviour is driven by the progress

of the plot, and they should generally be believable.

Fac¸ade is an important example of dramatic story-

telling that used conversation as interactive method.

It is necessary to exam how Fac¸ade express rich char-

acters, and why it may not succeeded.

Fac¸ade’s plot is that the user, an old friend of the

NPCs (Trip and Grace, a couple), visits for a drink.

Soon he is faced with marriage trouble. The user may

play marriage counselling, however, the plot options

are clear: break up or reconcile. The user does not

have an avatar, and uses type-in dialogue to interact.

NPCs will reply with pre-recorded audio dialogue.

Fac¸ade has succeeded in the following:

• Procedural storytelling: the story and NPC action

are made up dynamically along with the user in-

put.

• Murray’s idea of structuring the user agent as a

visitor.

• Managing natural reactions (in real-time anima-

tion) of NPCs.

• Natural Language based interaction.

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

564

• Simulating Neo-Aristotelian theatre Interactive

Drama model.

Fac¸ade is designed to be short and intensive, with

a play duration of 20 minutes. Mateas envisioned

Fac¸ade as a work of art, which the user will experi-

ence a ’strong agency’ when played the ﬁrst time and

would want to replay again to explore the rich char-

acters and drama intensive plot (Mateas, 1999). How-

ever, a user experience evaluation in 2012 revealed

that users’ story experience (curiosity, suspense, and

identiﬁcation) does not change between the ﬁrst time

and the second (Roth et al., 2012). This means that

users of Fac¸ade did not beneﬁt in the narrative ex-

perience from the repeated opportunity. Journalistic

writings about Fac¸ade often mention how the users

realize soon that he does not inﬂuence the plot or NPC

behaviour, and soon try to test out the system by act-

ing extreme.

The causes for the failure of Fac¸ade as dramatic

narrative can be various, however, this paper will fo-

cus the main problem: NPCs.

• NPCs in Fac¸ade do not have capability to carry

conversations outside of the plot’s topic.

The plot is rather predictable: a couple is break-

ing up. To make the narrative interesting, we need

complex characters. Trip and Grace might have rich

characters, but the user does not have access to them.

The user can not engage in any meaningful conver-

sation with NPCs, nor are events designed to express

the complexities of characters. Then, what happened

to the rich characters Mateas envisioned?

In Fac¸ade, AI was applied to two system tools:

one for story making (procedural story), and another

for managing NPC reactions. The keywords for NPC

can be believability, expressiveness, and emotional re-

actions(Reilly and Bates, 1992), (Mateas and Stern,

2002a). The term Believable Agent was deﬁned in

the OZ project at Carnegie Mellon University, which

developed programming languages, Em and Hap, for

automated emotions and actions (Bates et al., 1991).

The Believable Agent took its cues from narrative

psychologists, who argued that agents will be more

comprehensible if their visible behaviour is structured

into narrative (Mateas and Sengers, 1999). For Bates,

Reilly, and Mateas, believability of a character should

be expressed by clearly visible action. Thus emo-

tions are visualized through facial expressions, and

behaviour means corporeal movement. For exam-

ple, a character who is frustrated may walk to a door,

pause, breath out a sigh, then open the door. A sad-

ness would be showed as a sad face, in crooked angle

of eyebrows and mouth shapes. Believable Agent is

a character who shows what he feels. These actions

will exhibit rich personality, according to Mateas.

Exterior-wise, Trip and Grace do act with reason-

ably proper movement; walking, angry face, irritated

expression, throwing out arms, etc. However, inte-

rior wise, Trip and Grace do not have a database for

detailed personal information. What we can ﬁnd out

about them is mainly conditions in the plot, which we

are clearly being told about. They have been married

for 10 years, Trip was a bartender in his youth and

somehow he is shamed of it (we do not know why,

though know that he is afraid of being poor), they are

not happy in their marriage, Grace wanted to be an

artists, etc. Strangely enough, the pre-recorded dia-

logues are strictly used to deliver informations for the

plot (unhappy marriage). There is no subtlety in Trip

and Grace’s dialogue and action. It is loud and clear

that they are unhappy together, and are not meant to

listen to what the user has to say. We learn the cause

(Trip was bartender and shamed about it) and the ef-

fect (Trip is afraid of being poor), but missing a lot

of irrational, emotional, yet dramatically interesting

personal information, in the middle. Any attempt the

user does to start a more personal conversation (e.g.

what is wrong with being a bartender: what does Trip

do for living now), is met by a wall, as NPCs do

not have ability to handle talk outside of plot-oriented

topics. In this regard, Fac¸ade failed interacting with

the users.

Believable Agent in Fac¸ade is more about ab-

stracting a general acting method. ABL (A Behaviour

Language), the programming language developed for

Fac¸ade (Mateas and Stern, 2002a), is described by

Mateas as AI implementation of the actor’s mind to

chose proper movement (Rauch, 2006). Yet the char-

acters of NPCs can only by felt by their dialogues, not

by their actions. Characteristics of Trip and Grace ex-

ist in the 5 hours of pre-recorded dialogues and not in

the behaviour. The trust in the idea of ”action speaks

louder” seems to be misplaced. When the user was

offered a drink made by Trip, the user can not choose

any other drinks but his cocktail. Should this say that

Trip is a jerk? Instead of interpreting the behaviour

as indications of character, the user may believe that

the interaction has failed, making conversations with

NPC meaningless. The pretension of conversation

will be lost. The fact that dialogues are not oriented

towards describing the characters but the plot, may

render Fac¸ade unsatisfying as a drama for many.

Mateas did not consider chatbot as a possible

model for building a character (Mateas, 1999, p.16),

and seems to have believed that expressing personal-

ity is best as pre-written dialogues. The problem of

this approach is that no author can make up replies

for all kind of input from the user. Unfortunately, as

bad conservationists, Trip and Grace turned out to be

ComplexCharacter-ModelforaNonPlayerAICharacterforInteractiveNarrativeDiscourse

565

generic characters that lack interesting and convinc-

ing personalities to make the drama more satisfying.

The lesson from Fac¸ade’s NPC model can be that

a character in IN should be a complete entity, that can

hold a conversation in various situations in his own

way. In fact, I will argue that a character in IN should

be built more like a chatbot, anticipating any inquiry.

4 THE COMPLEX CHARACTER

IN CYBER COMIX

FRAMEWORK

Cyber Comix framework (Holmen, 2014a) is an in-

teractive comics system. Cyber Comix deﬁnes its

user as the Reader-Player. The reader-player is role-

playing a character (Reader-Player Character, RPC),

and may interact with a chatbot character (Narrative

Non Player Character, NNPC) by typing in dialogues

in talking balloons. There are also Non Player Char-

acters (NPC) in a story, however, they would not

have a chatbot function. Taking 2D graphics as vi-

sual medium directs the research effort to the devel-

opment of complex characters and conversational in-

teraction. Cyber Comix is designed for lengthy seri-

alized comics, to take advantage of the unlimited nar-

rative time.

To reveal complexities of an interior world

of characters, the Cyber Comix framework uses

conversation-based interaction. The following are the

main concepts in the framework:

• Dynamic Bubble is the name for the chatbot-

enabling authoring tool for talking bubbles.

• The Complex Character is the framework for

Narrative Non Player Character (NNPC) chatbot

database structure.

• The story structure takes the design cue from Mad

Lips (1953) by Stern and Price.

• Simulating Character Centric Interactive Drama

model.

Deep Data Mars (DDM)(Holmen, 2014b) is the

ﬁrst prototype in Cyber Comix. The story is sci-

ence ﬁction, and has one reader-player character, one

complex character (NNPC), and multiple NPCs. The

comics is a short serial of 10 episodes, each episode

has 20-30 frames. Currently two episodes are pub-

lished at the time of this paper submission. The plot

is that Mary-Ann (RPC), wakes up after long hiberna-

tion, and slowly ﬁnds herself with KahToo (NNPC),

the robot that she built before. She lost her memory

due to her long hibernation, and ﬁnds out what hap-

pened in her past. The reader-player will decide if

she still loves her ex-lover, among many other small

decisions.

The chatbot is largely used for expressing the

complex character of NNPC, but also for explaining

other characters. NNPC may be viewed as the narra-

tor, who may explain other character’s interior world.

The knowledge-based structure is as follows and is

hierarchical:

1. World proﬁle: year, season, date, time, social eti-

quette, name of location (city, country), etc.

2. Character proﬁle: character name, age, gender,

marriage statutes, education, occupation or social

standing.

3. Personal preference: personal hobby, friends

with, engaged with, married to, in love with, etc.

4. Critical incidents: speciﬁc decisions that the user

character must give an answer to, that affects sig-

niﬁcantly the story experience.

5. Secrets : The main story device.

Using comics as the narrative medium gives ad-

vantages in authoring control of both literary and

graphical semiotics. The ﬂexible nature of inter-

preting comics is an advantage for IN. For exam-

ple, the comics images will change the meaning what

the reader-player character types in. In Figures 1

and 2, the reader-player character’s (the woman) and

NNPC’s (the robot) different interactions in the sam-

ple image are shown. With the different dialogue in-

teractions, it can be seen that the narrative tone has

been changed. The author may decide and direct the

reader-player character by graphics and chatbot de-

sign: however, the executive decision is with the user,

who can make the story whatever he likes it to be:

play-along, play-against, or something else.

The Complex Character, a framework for NNPC,

is combining managing plot progressing and the char-

acter. This is different from previous IN systems, that

separate AI for plot and characters. As a character in

the story, NNPC in DDM is a super intelligent robot,

who is in love with RPC. To manage the plot progress,

NNPC reveals a few important plot progressing dia-

logues along the narrative discourse. This is called

Critical Incidents, where the reader-player may de-

cide important facts for the narrative. For example,

in episode 2, RPC, Mary-Ann, will be told by Kah-

Too that her ex-lover is suspected as the murderer of

her father. In later episodes Mary-Ann may decide if

her ex-lover indeed killed her father or not, and then

decide if she still loves him after all.

All categories are adjustable after RPC interac-

tions. Cyber Comix has a pre-production period to

test out its interaction design. This is after the Wiz-

ard of Oz method (Green and Wei-Haas, 1985), or

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

566

Figure 1: The user is playing along.

Figure 2: The user is playing something else.

heuristic approach to adjust after empirical data. In

these periods, interactions are reviewed by the author

to adjust balloons and responses accordingly.

NNPC is story speciﬁc. NNPC in DDM may not

be used for other stories. NNPC may have a person-

alized dialogue style and way to react to RPC inquiry,

uniquely in DDM.

5 RELATED WORKS

• ELIZA: The ﬁrst and one of the most successful

believable characters in AI. The chatbot creator

Joseph Weizenbaum was credited as the ﬁrst liter-

ary writer in computer medium by Murray (Mur-

ray, 1997, p.72), ELIZA is the ﬁrst successful

chatbot that is built on the simple principle of con-

versational pattern of a Rogerian psychotherapist.

The chatbot inﬂuenced Interactive Fiction as well

as other various chatbot character developments

(Ibid).

• Expressive AI: The concept by Mateas is for de-

vising interdisciplinary art and technology. ”This

effect of producing psychologically readable be-

haviour is not limited only to explicit anthropo-

morphic characters within the game world, but

also to intelligences operating behind the scenes.”

(Mateas, 2003)

• Abbott’s explanations for narrative discourse in

literature, theatre and ﬁlm promote the impor-

tance of characters (Abbott, 2008).

• Szilas investigated the problem regarding duration

of interaction in 3D real time Interactive Drama,

and proposed three different ways to tackle it: ca-

sual, semi-autonomy, and elliptic (Szilas, 2004).

6 CONCLUSION

Cyber Comix is a framework in progress. Com-

plex Character model for IN may solve customizing

NPCs for speciﬁc stories for conversational interac-

tion. Currently it is planned to have ’non ﬂexible’

graphics, that is not changed by the meaning of con-

versation. However, to improve the drama experi-

ence, it is possible to consider implementing a sep-

arate drama managing AI to have dynamic graphics

as well. To manage the chatbot, it might be more ef-

fective to have a unique programming language ded-

icated to Cyber Comix than using conventional chat-

bot AIML. The collaboration part of AI design is the

awareness of an incomplete story, that can be a prob-

lem implementing the framework. The reader-player

may or may not play along with the story, or the

RPC. System AI for drama, or story progress, is de-

signed by the authors graphical references in comics

and NNPCs character. It is also important to be aware

that NNPC in DDM will only exist when a reader in-

teracts with him. Cyber Comix has a capacity to ad-

just NNPCs database according to reader-player in-

teractions. When the database is built up by interac-

tions with readers, NNPC will be formed as a new,

bigger database, which is not deﬁned by the author.

Thus NNPC exists in the intersection of the author,

the story, and the reader. Thus NNPC in DDM is

largely unknown at the moment.

ComplexCharacter-ModelforaNonPlayerAICharacterforInteractiveNarrativeDiscourse

567

REFERENCES

Aarseth, E. (2007). I fought the law: Transgressive play and

the implied player. Situated Play. Proc. DiGRA, pages

24–28.

Abbott, H. P. (2008). The Cambridge introduction to nar-

rative. Cambridge University Press.

Aylett, R., Louchart, S., and Weallans, A. (2011). Re-

search in interactive drama environments, role-play

and story-telling. In Interactive Storytelling, pages 1–

12. Springer.

Bates, J., Loyall, B., and Reilly, W. S. (1991). Broad agents.

ACM SIGART Bulletin, 2(4):38–40.

Carlquist, J. (2013). Playing the story: computer games as

a narrative genre. Human IT: Journal for Information

Technology Studies as a Human Science, 6(3).

Green, P. and Wei-Haas, L. (1985). The rapid development

of user interfaces: Experience with the Wizard of Oz

method. In Proceedings of the Human Factors and Er-

gonomics Society Annual Meeting, volume 29, pages

470–474. SAGE Publications.

Holmen, H. (2014a). Cyber Comix framework. Online ac-

cess at http://cybercomix.studioh.dk/.

Holmen, H. (2014b). Deep Data Mars: Cyber Comix. On-

line access at http://cybercomix.studioh.dk/ddm/.

Laurel, B. (1993). Computers as theatre.

Mateas, M. (1999). An Oz-centric review of interactive

drama and believable agents. In Wooldridge, M.

and Veloso, M., editors, Artiﬁcial Intelligence Today,

volume 1600 of Lecture Notes in Computer Science.

Springer Berlin Heidelberg, Berlin.

Mateas, M. (2001). A preliminary poetics for interactive

drama and games. Digital Creativity, 12(3):140–152.

Mateas, M. (2003). Expressive AI: Games and artiﬁcial in-

telligence. In DIGRA Conf.

Mateas, M. and Sengers, P. (1999). Narrative intelligence.

In Proceedings AAAI Fall Symposium on Narrative In-

telligence, pages 1–10.

Mateas, M. and Stern, A. (2002a). A behavior language for

story-based believable agents. IEEE Intelligent Sys-

tems, 17(4):39–47.

Mateas, M. and Stern, A. (2002b). Towards integrating plot

and character for interactive drama. In Socially Intel-

ligent Agents, pages 221–228. Springer.

Mateas, M. and Stern, A. (2003). Fac¸ade: An experiment in

building a fully-realized interactive drama. In Game

Developers Conference, pages 4–8.

Mateas, M. and Stern, A. (2006). Interaction and narrative.

The game design reader: A rules of play anthology,

1:642–669.

Murray, J. H. (1997). Hamlet on the Holodeck: The future

of narrative in cyberspace. Simon and Schuster.

Rauch, J. (2006). Interviews: ”beyond space in-

vaders”. Atlantic Unbound. online access at

http://www.theatlantic.com/magazine/archive/2006/11/sex-

lies-and-videogames/305293/.

Reilly, W. S. and Bates, J. (1992). Building emotional

agents.

Roth, C., Vermeulen, I., Vorderer, P., and Klimmt, C.

(2012). Exploring replay value: shifts and continu-

ities in user experiences between ﬁrst and second ex-

posure to an interactive story. Cyberpsychology, Be-

havior, and Social Networking, 15(7):378–381.

Szilas, N. (2004). Stepping into the interactive drama. In

Technologies for Interactive Digital Storytelling and

Entertainment, pages 14–25. Springer.

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

568