Digital Signal Processing Reference
In-Depth Information
(Sim_MSDS); otherwise, it is called a sequential MSDS (Seq_MSDS). In a
Seq-MSDS, utterances of speakers are buffered first, and then they are
processed together. In this chapter we only consider Seq_MSDS.
In multi-speaker dialogues, speakers may cooperate to accomplish a
common goal or negotiate to solve conflicting opinions to achieve the same
goal. We defined two types of goals in MSDS (i.e., the individual goal and
the global goal). The individual goal is the answer that one speaker wants to
know from the inquiry. Since individual goals may conflict with each other,
the system should maintain a global goal in which it can integrate the
individual goals. The following examples will demonstrate different cases in
which individual goals do and do not conflict with each other. Depending on
the relationship between two individual goals, the interactions between
speakers and the system are classified as one of three types: independent,
cooperative, and conflicting. Examples are shown below, where
and
are
different speakers:
and
(i)
Independent interaction:
speakers
have independent goals
What's
the
weather in Taipei?
Where is the Tainan train station?
In the first example, the individual goal of each speaker is different and
independent.
(ii)
Cooperative interaction:
speakers have a common goal
Please find a place to eat.
I want to eat Japanese noodles.
In second example, the individual goal of S1 is to find a restaurant, and
the goal of is to eat Japanese noodles. The dialogue manager should detect
and integrate these individual goals to form the global goal, i.e., a place where
Japanese noodles are available.
(iii) Conflicting interaction:
speakers have conflicting goals
Tell me
a
Chinese restaurant.
I think we should go to an Italian restaurant.
In third example, wants to go to a restaurant which supplies Chinese
food while; in contrast, wants to go to an Italian restaurant. Their intentions
are similar, but the destinations conflict. The global goal should be adjusted
when speaker
has an individual goal different from that of
