A UNIFIED APPROACH FOR RECONCILING CHARACTERS
AND STORY IN THE REALM OF AGENCY
Rossana Damiano and Vincenzo Lombardo
Dipartimento di Informatica e CIRMA, Universit
`
a degli Studi di Torino
C.so Svizzera 185, Torino, Italy
Keywords:
Interactive Drama, Artificial Characters, Intelligent Agents.
Abstract:
In the last decade, a number of computational systems for entertainment and communication have appeared,
that share a set of common features, including the use of artificial characters and the reference to drama
and storytelling. Systems for interactive storytelling and drama rely on agent theories to model characters,
and adopt planning techniques to cope with non-determinism at the story level, combining them according to
sophisticate architectural designs. However, a consolidated approach has not emerged yet, that fully reconciles
these two dimensions. In this paper, we propose a unifying framework to accommodate the tension between
story control and character behavior and we claim that the accurate modeling of agency is a prerequisite to the
success of attempts to solve this tension. By using this framework to analyze practical systems we point out
that the importance of agency is acknowledged by successful systems, although only implicitly in most cases.
1 INTRODUCTION
In the last decade, a number of systems for enter-
tainment and communication have appeared that
notwithstanding different design goals and concep-
tions share a number of common features, includ-
ing the use of artificial characters and the adoption of
storytelling to structure the interaction with the user.
These systems rely on multiple modalities for com-
municating with the user, such as natural language,
graphics or virtual reality, and support different styles
of interaction, such as dialogue, direct manipulation
or embodiment.
For example, the Fac¸ade entertainment system in-
volves the user as an active character in a dramatic
situation in which a couple whose marriage is falling
apart invites her/him for a drink, with wife and hus-
band trying to bring the user on their respective sides.
In the interactive storytelling system by (Pizzi et al.,
2007), the user is immersed in a virtual reality envi-
ronment, where he/she plays the part of one of the
characters of the French novel “Madame Bovary”, in-
teracting with the other characters and affecting their
feelings and behavior. In the Dramatour guide appli-
cation for cultural heritage (Damiano et al., 2006), a
virtual character, the spider Carletto, accompanies the
visitor in a historical location from a portable device,
exhibiting an inner conflict between ‘guide’ and ‘sto-
ryteller’. In the FearNot! edutainment system (Aylett
et al., 2007), the dynamics underlying bullying inci-
dents in schools is dramatized in a cartoon-like en-
vironment, in which the child user intervenes as an
empathic observer to give advice to the victim.
Given this heterogeneity of goals and instruments,
a first, broad distinction has been established in the lit-
erature between story-based and character-based sys-
tems (Mateas and Stern, 2003; Pizzi et al., 2007; The-
une et al., 2003; Riedl and Young, 2006). Story-
based systems are characterized by centralized archi-
tectures, in which the behavior displayed by the sys-
tem, possibly through characters’ mediation, is driven
by the unifying principle of a story. The story to be
conveyed is usually underspecified in some way, so as
to provide some limited support to non-determinism
and interactivity. Character-based systems rely on the
autonomous behaviour of characters to create situa-
tions, which are then interpreted as emergent narrative
structures (Spierling, 2007).
Whatever the chosen approach, it is widely ac-
knowledged that the author’s control over the story
430
Damiano R. and Lombardo V. (2009).
A UNIFIED APPROACH FOR RECONCILING CHARACTERS AND STORY IN THE REALM OF AGENCY.
In Proceedings of the International Conference on Agents and Artificial Intelligence, pages 430-437
DOI: 10.5220/0001662004300437
Copyright
c
SciTePress
is related with communicative effectiveness; how-
ever, it must be traded off against the autonomy and
the believability of the characters. For some specific
forms of communication and entertainment, clear de-
sign strategies have emerged: for example, in video
games, the quality of playability, anchored in care-
fully shaped and strongly constrained stories, is pre-
ferred over the definition of psychologically believ-
able, autonomous characters. On the contrary, AI sys-
tems generally envisage interactivity as a main ob-
jective, sustained by a rich literature on interactive
storytelling and drama (Murray, 1998; Ryan, 2006;
Wardrip-Fruin and Harrigan, 2004). However, a con-
solidated approach has not emerged yet that fully rec-
onciles the two conflicting dimensions of story and
characters.
In this paper, we sketch a formal system that sys-
tematizes the functions of story and characters in a
unified theoretical framework.
By using this framework, we accommodate the
components of a variety of practical systems, pointing
out the relevance of the notion of agency as a prereq-
uisite to reconciling story and characters.
2 A FORMAL FRAMEWORK
FOR DRAMA
The theoretical framework presented in this section
lays out the ‘language of drama’, independently of the
form and the media through which specific dramas are
realized. Based on previous work by (Damiano et al.,
2005), revised here with a particular concern towards
the contribution of AI and agents, the framework en-
compasses the main feature of the practical systems
mentioned above.
Given the drama literature (Egri, 1946; McKee,
1997), for ‘drama’ we intend a form of narrative that
describes the story via characters’ action in present
time and has a carefully crafted premise, i.e., an au-
thorial direction that shapes the dramatic climax until
its solution.
Given this definition, the drama framework con-
sists of two levels, the directional level, that encodes
the specific traits of drama, and the actional level, that
connects such traits with the notion of agency. The di-
rectional level of the system (Fig. 1, top) is centered
on the notion of drama unit. A drama unit (du) is any
segment of the drama that contributes to the story ad-
vancement. Each advancement is due to a change of
polarity of some value at stake for one or more char-
acters (McKee, 1997); values refer to emotional states
or belief states.
Originally expressed by Aristotle as “unity of ac-
tion”, a drama unit provides an effective direction for
the story advancement operated by the segment of the
story to which it is associated.
Emotional states refer to a cognitive model of
emotions like the one stated by Ortony, Clore and
Collins (Ortony et al., 1988); this model, largely
employed in interactive drama (starting from (Bates
et al., 1994)), defines emotion types based on an
agent’s appraisal of self and other agents’ actions,
thus naturally lending itself to accommodate the di-
alectics of characters in drama. Belief states can be
any predicate logic formula.
A drama unit can be recursively expanded into a
number of children drama units, forming the plot tree
(Fig. 1, top left). The plot tree is licensed by the
formal rules of drama, which pose constraints on the
expansion relationship of dus (dominance relation in
formal grammar terms). In particular, the direction
of the parent du must include all the value at stake of
children dus, with consistent initial and final polarities
in case of subsequent changes.
The expansion of drama units into sub drama units
stops when, at the basic recursion, drama units are
expanded into a sequence of beats (bs in Fig. 1, bot-
tom). This sequence constitutes the actional level of
drama. Beats are the minimal units for the story ad-
vancement, that will be exposed to the audience.
Beats consist of incident pairs, where incidents are
characters’ actions – executed as part of their plans –
or unintentional events (represented by pairs of adja-
cent boxes in the central part of Fig. 1). While only
beliefs and emotions are represented at the directional
level,
at the actional level a model of agency, such as the
belief–desire–intention (BDI) model (Bratman et al.,
1988), provides the glue that links the behavior of
characters into a coherent causal chain and sustains
the believability of the characters.
The prescription that the direction is achieved
through conflict is captured by constraint that, for
each beat,
the intended outcome of the second incident
should be incompatible with the intended outcome of
the first incident, directly or indirectly.
In order to illustrate how this framework ap-
plies to specific drama, we resort to a well-known
example, the ‘nunnery scene’ from Shakespeare’s
Hamlet, describing how the characters’ rational and
emotional states are plotted by the author to achieve
the direction of the scene.
1
For all the segments
that compose the scene, we sketch the structure of
1
The scene has been analysed also by (Damiano and
Pizzo, 2008) in an actional perspective.
A UNIFIED APPROACH FOR RECONCILING CHARACTERS AND STORY IN THE REALM OF AGENCY
431
directional level
actional level
Character1’s
values
Character2’s
values
Figure 1: A graphical representation of the formal system of drama. On top, the directional level of drama, i.e. the way it
accomplishes its direction through a sequence of elements situated at different levels of abstraction (dus), which manipulate
characters’ values; below, the actional level, represents the anchoring of the direction into the characters’ actions contained in
beats and rooted in a rational and emotional model of agency.
the scene in terms of characters’ values (beliefs and
emotions), with the goal of showing how the plot
incidents affect these values (a formal derivation is
omitted for space reasons). Emotional values refer to
the ‘ontology of emotions’ described in (Ortony et al.,
1988); the modifications of values are represented by
their changes of polarity.
The scene has a tripartite structure; the value
at stake of the overall scene is Hamlet’s attitude
toward Ophelia (changing from positive to neg-
ative), h f eel(Hamlet, love f or(Ophelia)), +i.
In the first part, Ophelia is sent to Hamlet by
Claudius to induce him to reveal his inner feelings,
has goal(Ophelia, tell(Hamlet, inner f eelings)))),
as a way to confirm the hypothesis that his mad-
ness is caused by his rejected love for Ophelia.
In order to do so, she has formed a simple plan
that consists of starting an interaction with Hamlet,
and asking him to talk about his inner feelings
(has plan(Ophelia, (meet(Ophelia, Hamlet), start
interaction(Ophelia, Hamlet), ask(Ophelia, tell(
Hamlet, inner f eelings)))). Hamlet, who does
not want to foolish Ophelia (for whom he still
has a strong affection), tries to leave her, but
is forced to answer by Ophelia’s insistence, as
she makes subsequent attempts to start an in-
teraction with him. This segment of the scene
establishes Polonius’ value at stake, a belief state
in which he knows that Hamlet is mad because of
rejected love hbel(Polonius, mad(Hamlet)), −i;
Ophelia’s value is an emotional state that
includes fear about Hamlet’s madness,
h f eel(Ophelia, f ear f or(Hamlet madness)), −i
and hope (related with the fulfillment of her task
to convince Hamlet to reveal his inner feelings,
tell(Hamlet, inner f eelings).
At the beginning of the second part,
Hamlet reactively forms the goal of sav-
ing Ophelia from the corruption of Elsi-
nore court, has goal(Hamlet, save(Hamlet,
Ophelia)), by inducing her to go to a nunnery
(has goal(Hamlet, has goal(Ophelia, go(Ophelia,
nunnery)))). In order to do so, he resorts to a rhetor-
ical plan to convince her that moral and affective
values do not hold anymore, and that everybody,
including himself, is corrupted, with the intention
that she would spontaneously decide to go to the nun-
nery (bel(Hamlet, (bel(Ophelia, corrupted(court))
ICAART 2009 - International Conference on Agents and Artificial Intelligence
432
has goal(Ophelia, go(Ophelia, nunnery))))).
Hamlet’s value at stake is now set to hope about
Ophelia’s decision (and her consequent salva-
tion), h f eel(Hamlet, hope(go(Ophelia, nunnery)
)), +i, while Ophelia’s value at stake is a be-
lief state about the corruption of the court,
hbel(Ophelia, corrupted(court)), −i.
Finally, in the third part of the scene, Ham-
let verifies whether he has convinced Ophelia
that the court is corrupted, i.e., if h(bel(Ophelia,
corrupted(court)), +i holds. Prompted by the
fact a contingent event that he has noticed
Claudius and Polonius hidden behind the curtains,
hbel(Hamlet, (overhearing(Claudius)), +i, he real-
izes that she maybe be lying and asks her where her
father is. For Hamlet, this is a way to verify if Ophe-
lia is still subdued to his father’s authority. As she
answers that her father is at home, her lie, for Hamlet,
counts as a confirmation that his plan has failed (since
the belief that the entire court is corrupted should re-
tain her from obeying her father): so, Hamlet is upset.
He starts feigning madness again, thus
confirming Claudius’ and Polonius’ beliefs
about him, hbel(Polonius, mad(Hamlet)), +i;
the scene ends with Ophelia complaining
about her tragic destiny. Hamlet’s value at
stake is his emotional attitude towards Ophe-
lia h f eel(Hamlet, love f or(Ophelia)), −i,
while his hope to convince Ophelia to go
to the nunnery turns into disappointment,
h f eel(Hamlet, hope (go(Ophelia, nunnery))), −i,
and anger, h f eel(Hamlet, anger(not(bel(Ophelia,
corrupted(court))), +i, At the same time,
Ophelia experiences the emotional state of
fear confirmed about Hamlet’s madness,
h f eel(Ophelia, f ear con f irmed(Hamlet madness)),
+i, not at all mitigated by the fact of having ac-
complished her initial task h f eel(Ophelia, sel f
satis f action(done(tell(Hamlet, inner f eelings))),
+i.
The directional level corresponds to the authorial con-
trol over drama; at this level, the author puts the char-
acters’ values at stake, abstracting from the actions
represented in the plot, with the intent of establish-
ing the story meaning by manipulating the charac-
ters’ values. The outcome of an incident, be it action
or event, affects the characters’ values that are put at
stake at the directional level, so that the the story ad-
vances at the directional level as prescribed by the au-
thor. However, the outcomes of characters’ actions
not only affect their values, but must be coherent with
their goals; also, unintentional events must be plau-
sible as naturally occurring (like the revelation that
Polonius and Claudius are hidden behind the curtains
in this scene).
For instance, in the third part of the example scene,
Hamlet formulates the goal of knowing whether
Ophelia is lying or not, as a way to monitor the ac-
tual outcome of his rhetorical plan to convince Ophe-
lia that the court is corrupted and subtract her from
the submission to her father. By choosing to do it by
asking Ophelia about her father, he opportunistically
exploits the circumstance of which he has just be-
come aware that Claudius and Polonius have been
hiding behind the curtains to eavesdrop. Hamlet’plan
to know if Ophelia is sincere consists in asking her
directly, and when he comes to know that Ophelia is
lying, his value at stake of hope is consequently set
to a negative value. At the same time, since drama
units are part of a hierarchical structure, Ophelia’s an-
swer affects Hamlet’s value
at stake for the overall
scene, i.e., it turns Hamlet’s love to a negative value,
a change of major importance for the entire play.
As this short excerpt illustrates, at the ac-
tional level, even sophisticated characters like Shake-
speare’s show the typical behavior of rational agents:
they recognizably have goals, they form plans to
achieve them and monitor the execution of those
plans, a behavior that is apparent in the acts performed
by Ophelia and Hamlet. Moreover, they reactively
form new goals (e.g., Hamlet decides to act so as to
save Ophelia from the corruption of the court when
he realizes the necessity and the opportunity to do it)
and drop the goals that have proven unachievable (see
Hamlet’s utter disappointment when confronted with
his failure, and the similar feeling displayed by Ophe-
lia by the end of the scene). Rationality, then, is in-
tegrated with the emotional states that are determined
by the appraisal of the characters’ own goals and of
the other characters’ goals; emotional states in turn
affect the deliberative processes of the characters, by
providing the motivation for action, thus yielding the
complex interplay of emotions and rationality that can
be observed in drama.
3 APPLYING THE FORMAL
FRAMEWORK TO PRACTICAL
ARCHITECTURES
In the following, we use the framework sketched in
the previous section as a conceptual instrument to an-
alyze a sample of systems, that span a multiplicity of
relevant approaches and are characterized by the use
of a variety AI theories and techniques. With respect
to the dichotomy between story-based and character-
based approaches, the adoption of a unifying formal
A UNIFIED APPROACH FOR RECONCILING CHARACTERS AND STORY IN THE REALM OF AGENCY
433
framework brings the advantage of situating differ-
ent systems against a background in which the two
dimensions of story and characters are explicitly put
in relation, thus providing a common evaluation grid.
Moreover, the way each system copes with the tension
between story design and characters’ autonomy has
relevant consequences for the role of the procedural
author (Murray, 1998) who will input the knowledge
in the system.
3.1 Directional Level
In practical systems, setting the focus on the direc-
tional level corresponds to giving the author direct
control on the direction of the story and frustrating the
autonomy of characters. Since the directional level
manipulates facts that relate, more or less directly,
with the properties of characters (represented by the
‘values at stake’ in the drama framework), the inter-
nal structure of characters a sum of emotions and
beliefs must be represented in a more or less explicit
way.
Without a representation of this kind, it would not
be possible, for example, to set up intriguing situa-
tions in which the characters’ values are affected by
their intentional behavior, but in the opposite way
than planned by the characters themselves. For ex-
ample, Hamlet’s attempt to save Ophelia, motivated
by his love for her, is not only frustrated but ends up
turning into disappointment the feeling that initially
prompted him to act.
The intentional aspects of the characters’ behav-
ior, then, are dealt with in different way. A strategy
consists of infusing a representation of the intentions
of the character at the level of story control, so as to
constrain the evolution of the story to coherent char-
acter behavior; or, reversing the perspective, the au-
tonomy of characters in the system can be constrained
to sequences of actions that are known to be consis-
tent in advance, a typical property of pre-compiled
plans.
In general, story-based systems tend to operate
at the directional level and incorporate sophisticated
story models (object-level knowledge about the se-
mantics of drama according to the drama framework)
to account for the structural aspects of narration and
drama, ranging from semiotic structuralism (Szilas,
2003; Peinado and Gerv
´
as, 2004; Hartmann et al.,
2005) to cognitive models of story understanding
(Swartjes and Theune, 2006). The knowledge about
story generation has often been encoded in the form of
logical rules, like in DEFACTO (Sgouros, 1999) and
the IDtension (Szilas, 2003). These systems closely
resemble expert systems, mixing the empirical and
theoretical knowledge of the author in a set of rules
that the system uses to generate the story; the effec-
tiveness of these systems seems to be directly con-
nected to their ability to integrate actional operators
and emotional aspects.
Alternatively, a family of successful systems re-
sort to AI planning for the generation of the story. In
these systems, the planner may replace the author in
devising the plot: planning operators are represented
by a set of possible plot incidents (Riedl and Young,
2006), which can be combined in a consistent se-
quence to achieve a transition from a author-defined
initial and final state, with explicit constraints on the
ordering of incidents and their causal relations (where
intentionality can represent a weak form of causa-
tion). Or, the planner may be in charge of solving
a planning problem from the perspective of the story
characters, delegating the control over the direction to
the author’s capability of encoding full-fledged drama
units into joint planning operators (Mateas and Stern,
2005).
The control over the story direction is largely
preferred to the manipulation of characters when it
comes to designing authoring tools, as exemplified
by a recent generation of systems (Weiss et al., 2005;
Iurgel, 2006; Medler and Magerko, 2006). In partic-
ular, (Medler and Magerko, 2006) propose a hybrid,
layered language for drama representation in author-
ing tools, aimed at allowing the author to design the
story at the directional level, as a partially ordered
set of plot points, while leaving the story generation
system the task of mapping this representation onto a
planning formalism. Similarly, the Automated Story
Director (Riedl and Stern, 2006) confines the auton-
omy of characters to the generation of low-level be-
haviors based on action libraries, giving to a ‘drama
manager’ (Mateas and Stern, 2003) the responsibil-
ity for monitoring the story advancement to preserve,
during the interaction with the user, the story goals
encoded by the author.
Moreover, within each family of systems (rule-
based, planning-based and hybrid), several alterna-
tives are available concerning the techniques em-
ployed and the system architecture.
The Mimesis (Riedl and Young, 2006) storytelling
system, explicitly aims at constraining the interaction
with the user to the realization of a specific direc-
tion. The story is generated offline by a partial or-
der planner given the direction posed by the author;
the planner assembles actions executed by a set of
characters in order to construct an overall plan that
fulfills the goal stated by the author. The plan is an-
alyzed to detect the potential impact of user’s input,
and converted into a conditional plan in which this in-
ICAART 2009 - International Conference on Agents and Artificial Intelligence
434
put is accounted for in advance. This powerful tech-
nique, called ‘narrative mediation’, allows the author
to know, in full detail, the possible forms of the plot.
At the same time, the system sees to it that the in-
tentions of the characters remain clearly recognizable
to the user along the plan, and that no actions are in-
serted into the plan only to meet the needs of drama,
without being part of the motivational structure of the
characters. The core of this system clearly operates
at the directional level; however,the representation of
the characters as agents is acknowledged, though indi-
rectly, by the use of “intention frames”, motivational
accounts with which the actions included in the plot
are annotated. A drawback of this strategy is that the
mental state of the characters are accounted for only
indirectly, blurring the conceptual distinction between
the goals that are individually pursued by the charac-
ters and the drama goal.
The Fac¸ade system (Mateas and Stern, 2005) is
designed to conduct an interactive drama to a clearly
stated set of outcomes, in which the couple of protag-
onists either split or remain together (see Section 1),
with the user being neutral or sympathizing for one
of the two sides. In Fac¸ade, the richness of the user
experience resides in the user becoming a protagonist
of the story, triggering (but not controlling) the evo-
lution of the plot towards one of the available direc-
tions. The generation of the plot is obtained through a
hierarchical plan language (ABL), that encodes multi-
agent, joint plans; the plan language also accounts for
the role of the user in the joint action. Using ABL, the
author defines a set of beats, which represent interper-
sonal conflicts among the characters. The inclusion of
the joint plans in the interactive story is then affected
by the interaction with the user and guided by a mea-
sure of the plot emotional value.
This system lends itself to represent stereotypical
situations of western drama, in which the perspec-
tive on characters is somewhat reversed with respect
to the rational and emotional approach to agency un-
derpinning the formalization we propose: in Fac¸ade,
the characters tend to be shaped by the actions they
perform and the things they say in the space of the
interpersonal relations, instead of being stated by an
a priori definition of their mental state from which
intentions can subsequently be derived, according to
the standard practice that characterizes agent architec-
tures (Rao and Georgeff, 1991; Wooldridge and Par-
sons, 1999). For this reason, even if this system op-
erates at the directional level, it is situated mid-way
between the directional level and the actional level.
3.2 Actional Level
In general, character-based systems are situated at
the actional level of the drama framework proposed
here. They tend to take an improvisational approach
to drama, close to the “comedy of the art” tradition,
first translated in a computational architecture by the
work of Hayes-Roth (Hayes-Roth et al., 1995). Ac-
cording to this paradigm, drama emerges from the in-
teraction of a set of characters, constrained to specific
roles. This approach – whose realization has been en-
couraged by the availability of conceptual and practi-
cal tools to implement the characters’ deliberation
conflicts with the realization of a specific direction.
This situation corresponds to having a strongly un-
derdetermined direction, subsuming a large variety of
plots, or not having a specified direction at all.
It is worth noting that practical systems tend to
equate the autonomy of characters to the deliberative
processes and to use planning techniques for this pro-
cesses, relegating the representation of the characters’
mind to the use of truth maintaining systems and dele-
gating the plan monitoring activity to the built-in fea-
tures of planning–and–execution environments. This
approach, although effective in the practice, does not
allow the author to explicitly plan effects like frus-
tration or self-reflections that, although sophisticated,
are pervasive in drama. For example, turning back to
the excerpt analyzed in Section ??, think of the iter-
ated attempts performed by Hamlet to take revenge on
Claudius all along the play, and his frustrated attempt
to subtract Ophelia from the corruption of the court,
conducted through a rhetorical plan whose execution
encompasses monitoring actions of primary dramatic
importance.
As an example of this approach, consider the
‘Friends’ system (Cavazza et al., 2002), in which the
behavior of each character is generated by a hierar-
chical planner. Characters are committed to specific
goals (like seducing another character), and reactively
form intentions (i.e., partially instantiated plans) to
achieve them. The user interacts with as a ‘deus ex
machina’ with them by cooperating (for example, giv-
ing suggestions) or conflicting (for example, prevent-
ing actions); the actions of the user affect the behavior
of the characters, either influencing their future delib-
eration (characters’ high level plans are detailed out
only when execution approximates, leaving room for
the user’s advice) or forcing them to replan. The char-
acters are executed concurrently, so that further con-
flicts that are not envisaged by the autor may emerge
from their interaction.
From the author’s point of view, the use of the hi-
erarchical task network (HTN) planning paradigm al-
lows a more direct control of the development of the
A UNIFIED APPROACH FOR RECONCILING CHARACTERS AND STORY IN THE REALM OF AGENCY
435
plot, since it directly connects the characters’ behav-
ior with the specification of their goal; as a drawback,
it reduces the responsiveness of the system, that must
recur to replanning techniques to display a flexible be-
havior. In this sense, the system is a compromise be-
tween the autonomy of the characters at the actional
level and the control over the story, since their behav-
ior is constrained to a well defined set of hierarchical
plans. Although a proper representation of the char-
acters as agents is not present in the system, the use
of HTN planning indirectly gives a certain stability to
the behavior of characters, who keep at the same time
the capability to replan, a requirement posed by the
analysis of real plots. In practice, stating the drama
direction in terms of the characters’ goals releases the
control over story; from a theoretical perspective, it
collapses the distinction between the directional and
the actional level, since the former should abstract
from how actions are represented in the system.
In system described by (Pizzi et al., 2007), the fo-
cus is on the responsiveness of characters’ emotional
attitudes. The behavior of each character is generated
by a heuristic-search planner, and planning is limited
to the selection of the next action, to cope with asyn-
chronous user intervention without resorting to re-
planning. The planning operators represent ‘feelings’,
that manipulate the mental state of the characters. An
informal ontology, extracted from the novel inspir-
ing the story, Madame Bovary, defines how charac-
ters’ feelings vary and evolve as a consequence of the
changes in their beliefs, and how they affect the char-
acters’ behavior.
The author’s control over the system is confined to
the actional level, and consists of defining the initial
mental state of the characters and a set of planning
operators for each character. The resulting initial sit-
uation is open to opposite endings, depending on the
user’s input and the moment this input is provided.
The notion of conflict is mostly internal to the charac-
ters’ mental states, which can evolve towards different
final outcomes as a consequence of feelings.
This system is strongly committed to the actional
level of drama: within this dimension, it clearly priv-
ileges the accuracy of the emotional modeling of the
characters; it does not contain an explicit notion of
direction to provide a stronger control on the story
development; characters, in order to appear respon-
sive and thus to gain the user engagement, exhibit an
emotionally-based behavior rather than being driven
by explicitly coded intentions, encouraging the user to
explore, rather than to ‘direct’, the advancement of the
story. From the author’s point of view, the possibility
for the user to affect the beliefs and feelings of the
characters at any time, and the use of heuristic–search
planning (HSP), are likely to determine a large plan
space, leading to hypothesize a methodology consist-
ing of iterative testing to tune the behavior of the sys-
tem to the (unstated) author’s direction.
4 DISCUSSION AND
CONCLUSIONS
Although there is a general agreement about the role
played by characters and plot in practical systems,
the complex relations between the two have received
much less attention, a situation that can be partly at-
tributed to the fact that the notions of character and
story, taken in isolation, rely on generally agreed upon
models (autonomous agents, planning and graph the-
ories).
As a partial acknowledgment of the importance of
the direction to build an operational model of drama,
however, it has been amply recognized that the main
difficulty encountered by the interactive systems is to
reconcile the authorial control over the plot (i.e., the
direction) with the interaction with the user (see In-
troduction).
In this paper we have sketched a formal frame-
work to represent the domain of drama, aimed at ac-
counting for the interdependencies of the two levels of
drama, i.e., direction and actionality, at which the no-
tions of plot and characters are respectively situated.
The informal analysis of a classical example through
this framework has helped out to point out that a de-
tailed model of agency is necessary to grasp the be-
havior of characters in actual plot. Successful sys-
tems seem to confirm this assumption, since they rely
on some key notions of agent and multi-agent theories
(such as the properties of intentions, reactivity, meta-
deliberation aspects and so on) to reconcile story and
characters, although these notions are often only im-
plicitly modeled.
As a future work, we envisage the tasks of giving
to the framework a more complete formalization, and
to test its validity on a larger sample of systems and
plots, giving an accurate account of how the various
aspects of agency are tied to the properties of plot at
the directional level of drama, and how their explicit
modelling affects the expression of the author’s goals.
In particular, future research should address aspects
like the passive and active monitoring of action effects
(clearly visible in Ophelia’s attempts to gain Ham-
let’s attention in the first part of the scene analyzed in
Section ??), the reasoning about one’s own intentions
(visible, for example, in Hamlet’s sudden decision to
affect Ophelia’s behavior by convincing her to go to
a nunnery) and meta-deliberation capabilities (Ham-
ICAART 2009 - International Conference on Agents and Artificial Intelligence
436
let’s decision to return to his pretended madness as a
result of his personal failure with Ophelia, dropping
the infeasible goal to save Ophelia). Finally, since the
definition of drama lays out a multi-agent context, an
account is also needed of the relations between indi-
vidual agents, including cooperation, conflict and the
social and institutional relations that affect the moti-
vational structure of the characters.
REFERENCES
Aylett, R., Vala, M., Sequeira, P., and Paiva, A. (2007).
Fearnot!–an emergent narrative approach to virtual
dramas for anti-bullying education. LNCS, 4871:202.
Bates, J., Loyall, A., and Reilly, W. (1994). An architecture
for action, emotion, and social behaviour. Artificial
Social Systems, LNAI 830.
Bratman, M. E., Israel, D. J., and Pollack, M. E.
(1988). Plans and resource-bounded practical reason-
ing. Computational Intelligence, 4:349–355.
Cavazza, M., Charles, F., and Mead, S. (2002). Interact-
ing with virtual characters in interactive storytelling.
In Proc. of the First Int. Joint Conf. on Autonomous
Agents and Multiagent Systems.
Damiano, R., Lombardo, V., Nunnari, F., and Pizzo, A.
(2006). Dramatization meets information presenta-
tion. In Proceedings of ECAI 2006, Riva del Garda,
Italy.
Damiano, R., Lombardo, V., and Pizzo, A. (2005). Formal
encoding of drama ontology. In LNCS 3805, Proc. of
Virtual Storytelling 05, pages 95–104.
Damiano, R. and Pizzo, A. (2008). Emotions in drama char-
acters and virtual agents. In AAAI Spring Symposium
on Emotion, Personality, and Social Behavior.
Egri, L. (1946). The Art of Dramatic Writing. Simon and
Schuster, New York.
Hartmann, K., Hartmann, S., and Feustel, M. (2005). Mo-
tif definition and classification to structure non-linear
plots and to control the narrative flow in interactive
dramas. Third International Conference, ICVS, pages
158–167.
Hayes-Roth, B., Brownston, L., and vanGent, R. (1995).
Multi-agent collaboration in directed improvisation.
In First International Conference on Multi-Agent Sys-
tems, San Francisco.
Iurgel, I. (2006). Cyranus–an authoring tool for interactive
edutainment applications. In Pan, Z., Aylett, R., Di-
ener, H., Jin, X., Goebel, S., and Li, L., editors, Tech-
nologies for E-Learning and Digital Entertainment,
volume LNCS 3942. Springer.
Mateas, M. and Stern, A. (2003). Integrating plot, charac-
ter and natural language processing in the interactive
drama fa
´
ade. In TIDSE 03.
Mateas, M. and Stern, A. (2005). Structuring content in the
fa¸ade interactive drama architecture. In Proceedings
of Artificial Intelligence and Interactive Digital Enter-
tainment.
McKee, R. (1997). Story. Harper Collins, New York.
Medler, B. and Magerko, B. (2006). Scribe: A tool for
authoring event driven interactive drama. In Proc. of
TIDSE 2006, volume LNCS 4326, Darmstadt (Ger-
many). Springer.
Murray, J. (1998). Hamlet on the Holodeck. The Future of
Narrative in Cyberspace. The MIT Press, Cambridge,
MA.
Ortony, A., Clore, G., and Collins, A. (1988). The Cognitive
Stucture of Emotions. Cambrigde University Press.
Peinado, F. and Gerv
´
as, P. (2004). Transferring Game Mas-
tering Laws to Interactive Digital Storytelling. Proc.
of TIDSE 2004.
Pizzi, D., Charles, F., Lugrin, J., and Cavazza, M. (2007).
Interactive storytelling with literary feelings. In
ACII2007, Lisbon, Portugal, September. Springer.
Rao, A. and Georgeff, M. P. (1991). Modeling rational
agents within a BDI-architecture. In Proc. 2th Int.
Conf. Principles of Knowledge Representation and
Reasoning (KR:91), pages 473–484, Cambridge, MA.
Riedl, M. and Stern, A. (2006). Believable Agents and In-
telligent Story Adaptation for Interactive Storytelling.
Proceedings of the 3rd International Conference on
Technologies for Interactive Digital Storytelling and
Entertainment (TIDSE06).
Riedl, M. and Young, M. (2006). From linear story gen-
eration to branching story graphs. IEEE Journal of
Computer Graphics and Applications, pages 23–31.
Ryan, M. (2006). Avatars of Story. University of Minnesota
Press.
Sgouros, N. (1999). Dynamic generation, management and
resolution of interactive plots. Artificial Intelligence,
107(1):29–62.
Spierling, U. (2007). Adding Aspects of “Implicit Cre-
ation” to the Authoring Process in Interactive Sto-
rytelling. Proceedings of Virtual Storytelling 2007,
LNCS 4871:13.
Swartjes, I. and Theune, M. (2006). A fabula model for
emergent narrative. Technologies for Interactive Dig-
ital Storytelling and Entertainment (TIDSE), LNCS,
4326:95–100.
Szilas, N. (2003). Idtension: a narrative engine for interac-
tive drama. In TIDSE 2003, Darmstadt (Germany).
Theune, M., Faas, S., Heylen, D., and Nijholt, A. (2003).
The virtual storyteller: Story creation by intelligent
agents. In Proceedings TIDSE 03, pages 204–215.
Fraunhofer IRB Verlag.
Wardrip-Fruin, N. and Harrigan, P. (2004). First Person:
New Media As Story, Performance, and Game. MIT
Press.
Weiss, S., M
¨
uller, W., Spierling, U., and Steimle, F. (2005).
Scenejo–an interactive storytelling platform. In Sub-
sol, G., editor, LNCS 3805, Proc. of Virtual Story-
telling 05, pages 77–80. Springer.
Wooldridge, M. and Parsons, S. (1999). Intention reconsid-
eration reconsidered. In M
¨
uller, J., Singh, M., and
Rao, A., editors, Proc . of ATAL-98, volume 1555,
pages 63–80. Springer-Verlag: Heidelberg, Germany.
A UNIFIED APPROACH FOR RECONCILING CHARACTERS AND STORY IN THE REALM OF AGENCY
437