Generating Dialogue
<P>
Naoyuki Okada
<P>
Artificial Intelligence, Kyushu Institute of Technology<P>
Iizuka 820, Japan<P>
e-mail: okada@pluto.ai.kyutech.ac.jp
<P>
Dialogue understanding or generation involves a mental process
reflecting the real world fact to which the dialogue refers as
material. This research aims the linkage of non-linguistic, mental
states or processes with the deep structures of sentences from AI point
of view. A prototype for sentence generation was constructed this year.
The result is roughly sketched as follows: 

     First, a method for representing mental states or processes
according to the model of mind that we proposed previously is shown. A
mental object such as unit data or program is called "module" and
represented by a frame structure. A mental state or process is
captured as a series of activations of modules called "chain
activation," and represented by a dynamic network.  

     Second, situation, intention, and belief which are important in
dialogue processing are clarified on the network. For instance, a
situation is interpreted as "coherence" of a chain activation. 

     Next, generating deep structures of sentences from mental states
or processes is given. When a module is activated, deep case frames
which describes the behavior and result of the module are extracted.
Thus, a series of deep case frames are obtained along a chain
activation, forming the seep structures of generated sentences. The
generation process was implemented to be confirmed. 

    Finally, several problems towards dialogue generation are discussed.
<P>