Information Technology Reference
In-Depth Information
Discussion
Subjects in our study did not gracefully merge Agneta and Frida and the web
content into one narrative whole. Sometimes they enjoyed the contents of the
web pages, sometimes they were amused by the comments by Agneta and Frida,
and at some points web browsing and interaction were integrated into the story
of the two characters. Although the study taught us important aspects of inter-
action with embodied believable agents, it did not generate the result we had
hoped for. One reason for this may be the fact that we had too high expec-
tations of the creativity of the user in weaving the two narratives together. As
mentioned, many of the comments and behaviors of Agneta and Frida were
not really related to the web information on the pages, but of a more general
nature. One lesson to be learned form this study is that the agents' comments
must be tightly connected to the information displayed if disturbance effects
are to be avoided. On the face of it, it seems like the systems presented by Isbis-
ter & Doyle (Chapter 14) appropriated these features to a greater degree than
our system did. This, however, is also related to the objective of the system (e.g.
supporting serious wayfinding, guiding or explorative play), and in this respect
it seems like the Agneta & Frida system has a somewhat different function than
the systems of Isbister and Doyle.
The study also made it clear that evaluation of systems with believable
agents needs to take into consideration a larger context of cultural user expec-
tations. Humor, for instance, which was a rather central aspect in the system,
is based on personal preferences and socio-cultural dimensions. The question
whether Agneta and Frida are likable or not for a given subject is to large de-
gree a function of those contextual parameters, which an evaluation study must
take into account. In retrospect, we should have asked more about the prefer-
ences of the users, for instance their general attitude towards embodied agents
in interfaces.
While the results did not come out as we had hoped, we still believe that the
methods developed and deployed could be valuable in evaluating narrativity in
interactive systems.
Acknowledgements
The work reported here was done as part of the EC- and SITI-funded PER-
SONA project. Jarmo Laaksolahti assisted in the implementation. Kristina Tull-
gren and Jenny Holm conducted the analysis of narrative and spatial verbs and
