Graphics Reference
In-Depth Information
turn are much more evenly distributed between categories in the
person's speech than in the artificial interviewee's speech. The artificial
interviewee is as stated before very consistent, he decides what to say
beforehand and sticks to that. After listening to the recordings of the
person speaking there is a lot more variation occurring; decisions are
being made and changed at the spur of the moment leading to more
inconsistencies in prosody. An example of that is a person giving a
short answer with prosody that can be perceived as giving turn and
then adding to the answer and again ending with a give-turn prosody
(e.g. “My favorite actor is Will Smith. and Ben Affleck.”).
When the agent interviews 10 consecutive people, we analyzed
which patterns were most widely used at the end of a turn. We found
that four patterns out of nine are seen in up to 80% of turn-transitions
(see Figure 6). None of these patterns have an end pitch above session
average. This might be due to the fact that people are not asking the
agent any questions—and questions tend to end on a higher-than-
average pitch.
We further analyzed the use of the final fall pattern Down_Below ,
both as turn-transition and within turn. The use of final fall, both at
end of turn and within turn, varies considerably between participants.
The person that uses final fall the most at end of turn uses it in 41%
of the ends of turns, while the person that uses it the least only uses
it in 2.7% of cases (see Table 5). This is surprising as the pool of
participants are all from the same cultural background and we would
thus speculate more similarities in behavior.
Table 5. Usage of Down Below in the 10-person study.
Participant
At end
Within
1
7.7%
14.9%
2
14.8%
7.3%
3
34.8%
6.7%
4
6.3%
9.1%
5
2.7%
7.1%
6
27.3%
15.4%
7
41.0%
8.7%
8
18.8%
5.0%
9
11.1%
2.5%
10
25.0%
19.2%
Search WWH ::




Custom Search