RestKB: A Library of Commonsense Knowledge about Dining at a Restaurant

TL;DR

This paper introduces RestKB, a modular commonsense knowledge library about restaurant scenarios, developed in ALM, enhancing reasoning about restaurant narratives and exceptions.

Contribution

It expands action language ALM to effectively encode restaurant commonsense knowledge, improving construction, testing, and reasoning capabilities.

Findings

RestKB improves reasoning accuracy in restaurant scenarios.

Encoding in ALM enhances knowledge base quality and flexibility.

System successfully reasoned about stereotypical restaurant activities.

Abstract

This paper presents a library of commonsense knowledge, RestKB, developed in modular action language ALM and containing background knowledge relevant to the understanding of restaurant narratives, including stories that describe exceptions to the normal unfolding of such scenarios. We highlight features that KR languages must possess in order to be able to express pertinent knowledge, and expand action language ALM as needed. We show that encoding the knowledge base in ALM facilitates its piecewise construction and testing, and improves the generality and quality of the captured information, in comparison to an initial ASP encoding. The knowledge base was used in a system for reasoning about stereotypical activities, evaluated on the restaurant domain.