Information Technology Reference
In-Depth Information
-
context-sensitivity and negotiability:
In behavior-based systems, the 'mean-
ing' of a behavior is thought of as always the same: the name the designer
gives the internally-defined behavior. But in narrative comprehension,
meaning is not a matter of identifying already-given symbols, but comes
out of a complex process of negotiation between the interpreter and the
events being interpreted. The meaning of the same event can change radi-
callybasedonthecontextinwhichitoccurs,aswellasonthebackground,
assumptions, knowledge, and perspective of the interpreter. In order to
design narratively expressive agents, designers must respect (rather than
attempt to override) the context- and audience-dependency of narrative
comprehension.
-
intentional state entailment
: In most behavior-based systems, the reason a
behavior is run is implicit in its action-selection mechanism. The behavior
is then necessarily communicated to the user on a “just the facts, ma'am”
basis: it is usually easy to see
what
an agent is doing, but hard to tell
why
.
But in narrative, the reasons or motivations behind actions are just as im-
portant as - if not more so than - what is done. People do not want to
know just the events that occur in the narrative, but also the motivations,
thoughts, and feelings behind them. Supporting narrative comprehension
means communicating clearly not just what the agent does, but its reason
for doing it.
-
diachronicity
: Behavior-based agents jump from behavior to behavior ac-
cording to what is currently optimal. Each of these behaviors is designed
independently, with minimal interaction. But a fundamental property of
narrative is its diachronicity; it relates events over time. In a narrative,
events do not happen randomly and independently; they are connected
to and affect one another. Narrative support in a behavior-based agent
requires normally independent behaviors to be able to influence each
other, to present a coherent picture of narrative development to the user
over time.
These properties are the motivation for the Expressivator, an agent architecture
that focuses on the narrative expression of agent behavior. The Expressivator
is an extension of Bryan Loyall's Hap (Loyall & Bates 1991; Loyall 1997), a
behavior-based language designed for believable agents. The Expressivator has
been tested in The Industrial Graveyard, a virtual environment in which the Pa-
tient, a discarded lamp character implemented with the Expressivator, attempts
to eke out a miserable existence while being bullied about by the Overseer, an
agent implemented in Hap.
