The idea in plain language.
Everyday reasoning needs likely causes and effects, not only category labels.
How it works
The dataset records if-then knowledge about events: what might motivate an action, what may happen afterward, and how the people involved might react. Its relation types separate these different kinds of inference. The paper trains models to generate plausible inferences for events, testing whether shared structure across relation types helps them generalize.
What to keep in mind
Commonsense statements are contextual and can reflect annotation bias. A likely inference should not be treated as a fact about a particular person or situation.
Source: ATOMIC: An Atlas of Machine Commonsense for If-Then Reasoning. The original manuscript contains the methods, experiments, figures, and references. An arXiv posting date may follow an earlier conference publication. Read the linked record for version history.