Some questions about the paper

[Sakurakdx]

I have some questions about the paper that I would like to ask.
First, I am a little unclear about the way the prompt in the conversation history is constructed in the article. Are the strategies in the conversation history generated or annotated by the dataset?

[Sakurakdx]

For example, is the third strategy generated by another model or provided in the original dataset?

[maxlchen]

Hi @Sakurakdx -- thanks for reaching out. In the original paper we mapped dialogue acts to templated narrative instructions like here: https://github.com/maxlchen/Controllable-Mixed-Initiative-Dialogue-Generation/blob/main/agent/generator.py#L35

However, it is certainly a viable strategy to directly have a model generate a narrative instruction from a dialogue act in place of our templates ones.