![](bg2.png)
S: What class would you like to fly? Please, reply “Business” or “Economy”.
U: Economy.
S: What is your credit card number? Please, read the number figure by figure.
U:
One, two, three, four, five, six.
S: A seat in economy class is booked for you on 12
th
April, flight to Porto, Portugal.
Your credit card number is 123456. Is it correct? Please, reply “Yes” or “No”.
U: Yes.
S: Okay. Thank you. Good bye.
As you can see, the dialogue is acceptable for, say, novice user
1
. But an experienced
one would for sure like to “speed up” the dialogue causing the system to get confused
and terminate the dialogue immediately by means of some kind of escape utterance.
Usually the commercial systems put the user through to a human operator after the
escape utterance. Clearly, the above depicted situation is an implementative solution
of a human-computer dialogue but it is only a crutch after all. A full-featured, mature
human-computer interaction on the given topic should look e.g. like this (“S” is the
system, “U” is the user):
S: Hello, this is ARS—Airplane-ticket Reservation System.
How can I help you?
U:
I want to book a seat on flight to Porto on April, 12
th
.
S:
Do you mean Porto in Portugal?
U: Well, of course, do you fly to any other Porto?
S: There is Porto in Corsica where you may get via Calenzana.
U: Oh, I see. But I want to Porto in Portugal.
S: What class would you like to fly?
U: I prefer economy class.
S: What is your credit card number?
U: Twelve, thirty four, fifty six.
S:
A seat in economy class is booked for you on 12
th
April, flight to Porto, Portugal.
Your credit card number is 123456. Is it correct?
U:
Yeah.
S: Okay. Thank you. Good bye.
Such a dialogue with computer is generally possible—we tested e.g. CSLU Toolkit
package (see [1]) to design it (for English). The DIS
2
worked well but the recognition
accuracy was below the acceptable level for public release, mainly because of very
wide grammar covering all the possible dialogue situations. The restrictive power of the
grammar was so weak that it was not able to suppress the invalid recognized utterance
hypotheses. Unfortunately, Czech language has properties that make this task even more
complicated as described in section 2.
Our LICS research group examined the publicly available dialogue information sys-
tems operated in Czech language with the goal to explore their dialogue strategies. The
1
Unfortunately some kind of understanding the whole matter and a good deal of obedience is
necessary. And these are in practical operation quite rare.
2
Dialogue Information System
160