Simple human discourse model

Overlook of the model

Basically, only two types of utterance exist in this model.

Example

[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)

Semantic dependencies

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.

Topic

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".

Subtopic

The switching of subtopic is far more frequent than topic, a subtopic can be changed by a question or statement.

A chat bot based on this model

The following features are required.

  • Learn how to make response with the context of previous 2 ~ 3 utterances.
  • Learn how to infer and control the topic / subtopic.
  • Learn how to determine next topic / subtopic and the timing to change.

A mind map

conversation model mindmap