Self-supervised Knowledge Triplet Learning for Zero-shot Question Answering

The aim of all Question Answering (QA) systems is to be able to generalize to\nunseen questions. Current supervised methods are reliant on expensive data\nannotation. Moreover, such annotations can introduce unintended annotator bias\nwhich makes systems focus more on the bias than the actual task. In this work,\nwe propose Knowledge Triplet Learning (KTL), a self-supervised task over\nknowledge graphs. We propose heuristics to create synthetic graphs for\ncommonsense and scientific knowledge. We propose methods of how to use KTL to\nperform zero-shot QA and our experiments show considerable improvements over\nlarge pre-trained transformer models.\n