*=Equal taxpayers
Giving voice assistants the ability to navigate multi-turn conversations is a challenging problem. Handling multi-turn interactions requires the system to understand various conversational use cases, such as steering, intent transfer, unflux, entity transfer, and repair. The complexity of this problem is compounded by the fact that these use cases are mixed together and often appear simultaneously in natural language. This paper proposes a non-autoregressive query rewrite architecture that can handle not only the five aforementioned tasks, but also complex compositions of these use cases. We show that our proposed model performs competitively on a single task compared to the reference approach, and even outperforms a fitted T5 model in use case compositions, despite being 15 times smaller in parameters and 25 times larger. fast latency.
