MultiTalk: A Highly-Branching Dialog Testbed for Diverse Conversations

We study conversational dialog in which there are many possible responses to a given history. present the MultiTalk Dataset, corpus of over 320,000 sentences written that balances high branching factor (10) with several conversation turns (6) through selective branch continuation. make multiple contributions generation highly setting. In order evaluate diverse set generations, we propose simple scoring algorithm, based on bipartite graph matching, optimally incorporate references. language tasks at different levels predictive depth, using textual attributes induced automatically from pretrained classifiers. Our culminating task is challenging theory mind problem, controllable requires reasoning about expected reaction listener.