Dual-Channel Reasoning Model for Complex Question Answering

Multihop question answering has attracted extensive studies in recent years because of the emergence human annotated datasets and associated leaderboards. Recent have revealed that systems learn to exploit annotation artifacts other biases current datasets. Therefore, a model with strong interpretability should not only predict final answer, but more importantly find supporting facts’ sentences necessary answer complex questions, also known as evidence sentences. Most existing methods sequence or simultaneously, which inhibits ability models path reasoning. In this paper, we propose dual-channel reasoning architecture, where two channels sentences, respectively, while sharing contextual embedding layer. The can simply use same structure without additional network designs. Through experimental analysis based on public datasets, demonstrate effectiveness our proposed method