Contextualized Attention-based Knowledge Transfer for Spoken Conversational Question Answering

Spoken conversational question answering (SCQA) requires machines to model\ncomplex dialogue flow given the speech utterances and text corpora. Different\nfrom traditional text question answering (QA) tasks, SCQA involves audio signal\nprocessing, passage comprehension, and contextual understanding. However, ASR\nsystems introduce unexpected noisy signals to the transcriptions, which result\nin performance degradation on SCQA. To overcome the problem, we propose CADNet,\na novel contextualized attention-based distillation approach, which applies\nboth cross-attention and self-attention to obtain ASR-robust contextualized\nembedding representations of the passage and dialogue history for performance\nimprovements. We also introduce the spoken conventional knowledge distillation\nframework to distill the ASR-robust knowledge from the estimated probabilities\nof the teacher model to the student. We conduct extensive experiments on the\nSpoken-CoQA dataset and demonstrate that our approach achieves remarkable\nperformance in this task.\n