Digital Signal Processing Reference
In-Depth Information
adopted, which is based on the selection of the output having the maximum
likelihood.
2.3
Natural Language Understanding
The Natural Language Understanding (NLU) module (being developed at
Daimler-Chrysler AG) has to take care of the spontaneous speech input and
possible incomplete or ungrammatical sentences. The parser must be able to
deal with all these factors and provide a semantic representation of the user's
sentence. The whole structure of an utterance is not currently checked for
consistency, rather single elements containing the phrase have to be
identified. The resulting semantic representation is composed by a set of slots
instantiated by the meaningful words of the utterance, taking into account the
context of the on-going dialogue.
2.4
Dialogue Manager
At the core of the spoken dialogue system, the Dialog Manager aims at
reacting appropriately to the user's spontaneous requests in all circumstances.
The input to this module is the semantic information delivered by the NLU;
the output consists of queries to the databases through the CWW module,
instructions to the Response Generator, and the selection of grammars for the
language understanding module according to the dialogue status. Basically,
the DM must be able to identify the current task requested by the user and to
obtain the information needed to fulfill it. This has to be done also in case of
incomplete, ambiguous or erroneous input, to integrate or clarify the
information provided by the driver, and to accept a request of modification in
a possible negotiation step.
The implemented DM is based on a description of the application, which
is basically characterized by a set of contexts, each one containing some
semantically relevant concepts, and by a set of procedures that describe the
actions to be performed during the dialogue. The goal of the DM is to fill the
concepts of the active context in a consistent way, then to require some
external actions and finally to move to another context.
The Dialog Manager should be aware of the profile of the user, retrieved
via the Car Wide Web. The user profile should contain information about
user's preferences and habits, like preferred hotel chains, restaurants,
familiarity with the system. For example DM should be able to customize the
interaction in order to avoid to be petulant with expert users or to be too
concise with novice users. A sample dialogue is reported in Figure 6-2 as an
