Improving Intent Detection and Slot Filling for Vietnamese

Intent detection and slot filling are critical tasks in spoken language understanding. In this paper, we propose a method to leverage word features for these tasks. We propose an extension of JointIDSF [1] with two additional contextual layers to incorporate bidirectional in-tent and slot information. We further investigate a variant of Co-Interactive Transformer (CoITrans) model [2] by employing BERT for sentence presentation as additional information for intent detection task. Using Vietnamese as a case study, the experimental results on PhoATIS dataset show that our proposals achieve the state-of-the-art performance.