Complex Reasoning over Logical Queries on Commonsense Knowledge Graphs.

Event commonsense reasoning requires the ability to reason about therelationship between events, as well as infer implicit context underlying thatrelationship. However, data scarcity makes it challenging for language modelsto learn to generate commonsense inferences for contexts and questionsinvolving interactions between complex events. To address this demand, wepresent COM2 (COMplex COMmonsense), a new dataset created by sampling multi-hoplogical queries (e.g., the joint effect or cause of both event A and B, or theeffect of the effect of event C) from an existing commonsense knowledge graph(CSKG), and verbalizing them using handcrafted rules and large language modelsinto multiple-choice and text generation questions. Our experiments show thatlanguage models trained on COM2 exhibit significant improvements in complexreasoning ability, resulting in enhanced zero-shot performance in bothin-domain and out-of-domain tasks for question answering and generativecommonsense reasoning, without expensive human annotations.