A Property-Based Method for Acquiring Commonsense Knowledge

Commonsense knowledge is crucial in a variety of AI applications. However, one kind of commonsense knowledge that has not received attention is that of properties of actions denoted by verbs. To address this limitation, we propose an approach to acquiring commonsense knowledge about action properties. In this paper, we take self-motion actions as an example to present our method. We first identify commonsense properties of actions from their definitions. We then introduce a list of dimensions for acquiring commonsense knowledge based on adjectives. Finally, we extract commonsense knowledge from text by parsing sentences that involve actions. Experiments show that our method allows to obtain high-quality commonsense knowledge.