Conversational Neuro-Symbolic Commonsense Reasoning

• DOI: 10.1609/aaai.v35i6.16623
• Publication Date: 2022-09-08T18:39:54Z
• Authors: Forough Arabshahi; Jennifer Lee; Mikayla Gawarecki; Kathryn Mazaitis; Amos Azaria; Tom Mitchell
• Keywords: Commonsense reasoning; Commonsense knowledge; Presumption; Natural language understanding

In order for conversational AI systems to hold more natural and broad-ranging conversations, they will require much commonsense, including the ability identify unstated presumptions of their partners. For example, in command "If it snows at night then wake me up early because I don't want be late work" speaker relies on commonsense reasoning listener infer implicit presumption that wish woken only if enough cause traffic slowdowns. We consider here problem understanding such imprecisely stated language commands given form if-(state), then-(action), because-(goal) statements. More precisely, we identifying allow requested action achieve desired goal from state (perhaps elaborated by making explicit). release a benchmark data set this task, collected humans annotated with presumptions. present neuro-symbolic theorem prover extracts multi-hop chains, apply problem. Furthermore, accommodate reality current lack full coverage, also an interactive framework built our system, conversationally evokes knowledge complete its chains.