Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Texte intégral / Full text (pdf, 20 MiB) - Infoscience - EPFL
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Chapter 6. Experimental Validations - Social Phobia<br />
2.80 when considering only the subjects who looked at the character throughout the whole<br />
exercise. This corresponds to a neither normal nor abnormal evaluation when considering all<br />
subjects. However, it corresponds to a fairly normal evaluation when considering only the<br />
subjects who looked at the character throughout the talking exercise.<br />
From our results, we can see that the subjects have clearly identified the difference between<br />
a bored and an attentive character. Not only are the scores to the bored and attentive<br />
characters clearly categorized, but the SDs are also small, which indicates small variations<br />
in scoring between the 12 subjects. These values are shown in parentheses in Table 6.11.<br />
Moreover, from the responses we obtained for the 5th question, many subjects reacted very<br />
negatively to the bored character. We received feedbacks such as “he could not care less<br />
about what I was saying”, “I almost made her cry”, or “he was more than distracted, almost<br />
condescending”. We also received feedbacks such as “she was making me think of a lover<br />
who was drinking my words”, “he was very interested and receptive”, or “he was faking he<br />
was interested” for the always attentive character. The results to the random version of the<br />
scene were slightly above the mean value of 3 with SDs of 0.80 and 1.08 for the first and<br />
second questions respectively. This seems quite logical since the characters for this scene<br />
would alternatively be attentive or bored, and that we usually are more receptive to negative<br />
attitudes than to positive ones. For some of the subjects, there were more positive attitudes<br />
than negative ones and vice versa, which explains the higher SD values. For the tracked<br />
version, the results we obtained were very similar to those of the random version. However,<br />
when considering only the people who looked at the character throughout the whole talking<br />
exercise, these values came closer to those of the attentive version, which equally seems logical.<br />
Indeed, we hypothesized that such a closed loop induced by the subject’s gaze behavior<br />
characterized an interaction through which the subject is empowered as he/she can perceive<br />
he/she has an influence on the outcome of the interaction (note that the subjects were not<br />
told about the potential influence of their gaze behavior). However for such a feedback to<br />
be built, it is first necessary for the subject to make eye contact with the virtual character.<br />
Hence, a lack of motivation from the subject prevents the triggering of such a positive feedback<br />
loop. The SDs equally sustain this hypothesis since their values are quite high when<br />
considering all subjects and decrease when considering only the subjects who looked at the<br />
character during the exercises.<br />
The lack of involvement from some subjects may explain the answers to the fourth question<br />
concerning the normality of the characters. The bored version scored slightly under 4,<br />
corresponding to fairly abnormal, which is what we expected to obtain. However, for the<br />
attentive version, the score corresponds to a slightly normal evaluation. It is not surprising<br />
to see a better scoring than for the bored version, but we did not expect it to be the best out<br />
of the four. It seems as though the subjects were expecting the virtual character to be always<br />
attentive to them. The characters from the random version appeared to be more normal to the<br />
subjects than the characters from the bored version, which is what we expected. However,<br />
they seemed less normal than characters from the attentive version, possibly for the reason<br />
highlighted above (i.e. a bias from the subject towards an expectation of an always attentive<br />
virtual character). For the tracked version, the same comments apply. However, it is important<br />
to note that the characters seemed more normal to the subjects in the tracked version of<br />
the scene than in the random one which is in line with our second hypothesis.<br />
100