Beyond I.I.D.: Three Levels of Generalization for Question Answering on Knowledge Bases

TL;DR

This paper introduces GrailQA, a large-scale dataset for question answering on knowledge bases with three levels of generalization, and proposes a BERT-based model to improve KBQA performance across these levels.

Contribution

It presents a new dataset and evaluation framework for three levels of KBQA generalization, and demonstrates the effectiveness of BERT in enhancing model generalization.

Findings

GrailQA enables evaluation of i.i.d, compositional, and zero-shot generalization.

BERT-based KBQA model outperforms previous approaches across all generalization levels.

Pre-trained embeddings like BERT significantly improve zero-shot and compositional generalization.

Abstract

Existing studies on question answering on knowledge bases (KBQA) mainly operate with the standard i.i.d assumption, i.e., training distribution over questions is the same as the test distribution. However, i.i.d may be neither reasonably achievable nor desirable on large-scale KBs because 1) true user distribution is hard to capture and 2) randomly sample training examples from the enormous space would be highly data-inefficient. Instead, we suggest that KBQA models should have three levels of built-in generalization: i.i.d, compositional, and zero-shot. To facilitate the development of KBQA models with stronger generalization, we construct and release a new large-scale, high-quality dataset with 64,331 questions, GrailQA, and provide evaluation settings for all three levels of generalization. In addition, we propose a novel BERT-based KBQA model. The combination of our dataset and model enables us to thoroughly examine and demonstrate, for the first time, the key role of pre-trained contextual embeddings like BERT in the generalization of KBQA.