Digital Signal Processing Reference
In-Depth Information
5.2
Statistics of Tasks
In Phase I, we specified a sequence of three tasks: 1) traffic; 2) restaurant;
and 3) tourist attraction. In Phase II, we did not specify and ordered set of
tasks. Speakers could switch dialog tasks freely during their interactions with
the operator. This is a less restrictive approach for corpus collection, and a
more varied set of dialog tasks appear in Phase II. Therefore, it is of much
interest to analyze the number and types of dialog tasks as well as the task
transitions attested in the corpus collected from Phase II.
Figure 3-11.
Task frequency ration in the set A and B.
Analysis of the collected dialogs revealed that the total number of dialog
tasks in Phase II was 1,593, indicating an average number of tasks per speaker
of 8.1. As described in Section 3.3, speakers in Phase II generated two sets of
dialog task sets (
A
and
B)
according to their interests. These two sets
corresponded to the task before starting the trip on the first day, and the task
before leaving the hotel on the second day, respectively. The total number of
tasks in sets
A
and
B
were 1,002 and 591, respectively, indicating that the
average number of tasks per set per speaker was 5.1 and 3.0, respectively.
Figure 3-11 contains a comparison of the task frequency ratios in the set
A
and
B.
The three most desired types of information in the corpus were traffic
information, tourist attraction information, and restaurant information.
Naturally, the desire for hotel information disappeared in task set
B.
The most
frequently requested information (task) on the first day was traffic
information. On the other hand, the most frequently requested information on
