Do Language Models Perform Generalizable Commonsense Inference?
This work addresses the problem of limited generalization in commonsense knowledge mining for AI systems, but it is incremental as it primarily analyzes existing limitations without proposing a new solution.
The paper analyzes the ability of pretrained language models to perform generalizable commonsense inference across multiple knowledge graphs, finding they adapt to different schemas but fail to generalize to new relations and show limited generalization to novel objects.
Inspired by evidence that pretrained language models (LMs) encode commonsense knowledge, recent work has applied LMs to automatically populate commonsense knowledge graphs (CKGs). However, there is a lack of understanding on their generalization to multiple CKGs, unseen relations, and novel entities. This paper analyzes the ability of LMs to perform generalizable commonsense inference, in terms of knowledge capacity, transferability, and induction. Our experiments with these three aspects show that: (1) LMs can adapt to different schemas defined by multiple CKGs but fail to reuse the knowledge to generalize to new relations. (2) Adapted LMs generalize well to unseen subjects, but less so on novel objects. Future work should investigate how to improve the transferability and induction of commonsense mining from LMs.