Integrating Dialog History into End-to-End Spoken Language Understanding Systems

End-to-end spoken language understanding (SLU) systems that process\nhuman-human or human-computer interactions are often context independent and\nprocess each turn of a conversation independently. Spoken conversations on the\nother hand, are very much context dependent, and dialog history contains useful\ninformation that can improve the processing of each conversational turn. In\nthis paper, we investigate the importance of dialog history and how it can be\neffectively integrated into end-to-end SLU systems. While processing a spoken\nutterance, our proposed RNN transducer (RNN-T) based SLU model has access to\nits dialog history in the form of decoded transcripts and SLU labels of\nprevious turns. We encode the dialog history as BERT embeddings, and use them\nas an additional input to the SLU model along with the speech features for the\ncurrent utterance. We evaluate our approach on a recently released spoken\ndialog data set, the HarperValleyBank corpus. We observe significant\nimprovements: 8% for dialog action and 30% for caller intent recognition tasks,\nin comparison to a competitive context independent end-to-end baseline system.\n