Abstract
Existing research on response generation for chatbot focuses on \textbf{FirstResponse Generation} which aims to teach the chatbot to say the first response(e.g. a sentence) appropriate to the conversation context (e.g. the user'squery). In this paper, we introduce a new task \textbf{Second ResponseGeneration}, termed as Improv chat, which aims to teach the chatbot to say thesecond response after saying the first response with respect the conversationcontext, so as to lighten the burden on the user to keep the conversationgoing. Specifically, we propose a general learning based framework and developa retrieval based system which can generate the second responses with theusers' query and the chatbot's first response as input. We present the approachto building the conversation corpus for Improv chat from public forums andsocial networks, as well as the neural networks based models for responsematching and ranking. We include the preliminary experiments and results inthis paper. This work could be further advanced with better deep matchingmodels for retrieval base systems or generative models for generation basedsystems as well as extensive evaluations in real-life applications.