Abstract

We present ATOMIC, an atlas of everyday commonsense reasoning, organizedthrough 300k textual descriptions. Compared to existing resources that centeraround taxonomic knowledge, ATOMIC focuses on inferential knowledge organizedas typed if-then relations with variables (e.g., "if X pays Y a compliment,then Y will likely return the compliment"). We propose nine if-then relationtypes to distinguish causes v.s. effects, agents v.s. themes, voluntary v.s.involuntary events, and actions v.s. mental states. By generatively training onthe rich inferential knowledge described in ATOMIC, we show that neural modelscan acquire simple commonsense capabilities and reason about previously unseenevents. Experimental results demonstrate that multitask models that incorporatethe hierarchical structure of if-then relation types lead to more accurateinference compared to models trained in isolation, as measured by bothautomatic and human evaluation.

 

Introduction (beta)

None

 

Conclusion (beta)

None