Basically, only two types of utterance exist in this model.
[Topic: job] [Subtopic: B's job] A: How is your new job? (question) B: That is great (statement) A: You seems really loves the job (statement) [Subtopic: A's job] B: What is your job? (question) A: I'm a egg seller. (statement) B: Sounds cool (statement) [Topic: egg] [Subtopic: whether B likes eggs] A: By the way, do you like eating eggs? (question) B: Huh, not really. (statement)
In this example, we can see the content of responses are heavily relying on previous utterances. It turns out that previous 2 or 3 utterances.
At each time in the conversation, we all have a topic. And this topic changes over the time, but only changes at specified conditions. When a person changes the topic, he will use some transition phrases like "By the way".
The switching of subtopic is far more frequent than topic, a subtopic can be changed by a question or statement.
The following features are required.