Abstract:Commonsense reasoning is one of the abilities necessary for artificial intelligence to be as intelligent as humans. However, how to make AI understand commonsense has been a problem that has plagued artificial intelligence for more than 60 years. Existing efforts focus more on the means of knowledge acquisition and strive to enrich the capacity of commonsense knowledge (CSK) bases and dimensions of CSK through advanced methods. Unfortunately, this exuberance has obscured a general consideration of CSK, such as how to follow human habits to obtain the most representative knowledge we need to understand the world. In this paper, this representative knowledge is referred to as core CSK. The influence of core CSK is extensive, and it constitutes almost the fundamental element of human life and the most fundamental cognition of the world. Harnessing human curiosity to find solutions to the above problems is an effective and straightforward route. Specifically, we focus on a special corpus to mine core CSK, namely, why-questions. For example, we can harvest “the sky is blue” from “why is the sky blue?”. To this end, we propose a novel method to extract CSK from why-questions, which mainly consist of two modules. The first is a question classification module used to determine whether a question contains CSK. In this module, we propose a classifier based on a one-sided bootstrapping method and design several informative features for the classifier. The second is a crowdsourcing module used to improve the quality of the extracted commonsense. We conduct extensive experiments, and the experimental results show that our method effectively mines CSK from question corpora. Furthermore, statistical analysis demonstrates the feasibility of this curiosity-driven approach, implying that we provide a basic idea for collecting core CSK. Remarkably, today’s outstanding large language models do not have such simple knowledge summarization capabilities, demonstrating the barrier between the excellence of language models and the universality of CSK.

Don’t Ignore the Drive of Curiosity: Rethinking Subtleties Between Universality of Commonsense Knowledge and Excellence of Large Language Models

Coarse-to-Careful: Seeking Semantic-related Knowledge for Open-domain Commonsense Question Answering

Commonsense Generative Model for Chinese Automatic Knowledge Graph Construction

Building a commonsense knowledge base for context-awareness inference

Incorporating Domain Knowledge and Semantic Information into Language Models for Commonsense Question Answering

UNcommonsense Reasoning: Abductive Reasoning about Uncommon Situations

Evaluating Commonsense in Pre-trained Language Models

Mining Verb-Oriented Commonsense Knowledge

VoCSK: Verb-oriented commonsense knowledge mining with taxonomy-guided induction

What makes us curious? analysis of a corpus of open-domain questions

Improving Commonsense Question Answering by Graph-based Iterative Retrieval over Multiple Knowledge Sources

Enhancing Question Generation with Commonsense Knowledge

A Logical Model for Supporting Social Commonsense Knowledge Acquisition

Commonsense Knowledge Construction with Concept and Pretrained Model.

Large Language Models Are Also Good Prototypical Commonsense Reasoners

Teaching Pretrained Models with Commonsense Reasoning: A Preliminary KB-Based Approach

Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering

Ask & Explore: Grounded Question Answering for Curiosity-Driven Exploration

Joint Reasoning for Multi-Faceted Commonsense Knowledge

Visually Grounded Commonsense Knowledge Acquisition

CommonsenseVIS: Visualizing and Understanding Commonsense Reasoning Capabilities of Natural Language Models