Graphics Reference
In-Depth Information
by M-scores decreases. Krippendorff's alpha (Krippendorff, 2004) has
been suggested as an alternative to M, as it takes care of the biased
category distributions.
3. Representation Languages
Representation languages (or other forms such as script or mark-up
languages (Krenn et al., 2011)) have been designed over many years
to drive virtual agents' behavior. They provide information on the
type of signals to be displayed; when they should appear and for how
long; which other signals are also present, etc. Examples of existing
languages are VHML, MPML, APML, GESTYLE, etc. The various
existing languages may encode information as various as activities,
culture, voice quality, signal description, emotion, etc. The readers
are referred to Prendinger and Ishizuka (2004) and Krenn et al. (2011)
for a broader view of existing works. However, in this section we
will focus on three examples which are representatives of encoding:
communicative functions, affective states, and multimodal behavior.
The representation languages are respectively called FML (Heylen et
al., 2008), EmotionML (Schröder et al., 2011) and BML (Kopp et al.,
2006). Whilst, apart from EmotionML, the other two are not part of
any standard (such as W3C or ISO), but are widely used throughout
the agent community.
Before starting this description, we will introduce SAIBA, an
international initiative for defining a common framework when
modeling embodied conversational agents (see Figure 1). SAIBA
stands for Situation, Agent, Intention, Behavior, and Animation
(Kopp et al., 2006; Vilhjálmsson et al., 2007). SAIBA arises from the
observation that more current agent systems have followed a similar
architecture, namely to be composed of three main modules: the
first one, called Intent Planner, takes as input information from the
context (e.g. the environment, the conversation state). It outputs a list
of communicative intentions and emotional states which are encoded
through Function Mark-up Language—FML (Heylen et al., 2008) and
EmotionML (Schröder et al., 2011) (see next sub-section for further
information). The next module called Behavior Planner instantiates
Figure 1. SAIBA framework.
Search WWH ::




Custom Search