2019/09/12 by Abhinav Rastogi, Rastogi, Abhinav, Xiaoxue Zang +7 · 38 citations
Computer Science · #Topic Modeling #AI in Service Interactions #Speech and dialogue systems
paper · pdf · doi:10.48550/arxiv.1909.05855
Virtual assistants such as Google Assistant, Alexa and Siri provide a\nconversational interface to a large number of services and APIs spanning\nmultiple domains. Such systems need to support an ever-increasing number of\nservices with possibly overlapping functionality. Furthermore, some of these\nservices have little to no training data available. Existing public datasets\nfor task-oriented dialogue do not sufficiently capture these challenges since\nthey cover few domains and assume a single static ontology per domain. In this\nwork, we introduce the the Schema-Guided Dialogue (SGD) dataset, containing\nover 16k multi-domain conversations spanning 16 domains. Our dataset exceeds\nthe existing task-oriented dialogue corpora in scale, while also highlighting\nthe challenges associated with building large-scale virtual assistants. It\nprovides a challenging testbed for a number of tasks including language\nunderstanding, slot filling, dialogue state tracking and response generation.\nAlong the same lines, we present a schema-guided paradigm for task-oriented\ndialogue, in which predictions are made over a dynamic set of intents and\nslots, provided as input, using their natural language descriptions. This\nallows a single dialogue system to easily support a large number of services\nand facilitates simple integration of new services without requiring additional\ntraining data. Building upon the proposed paradigm, we release a model for\ndialogue state tracking capable of zero-shot generalization to new APIs, while\nremaining competitive in the regular setting.\n