Although data-driven techniques are commonly used for Natural Language Understanding in dialogue systems, their efficacy is often hampered by the lack of appropriate annotated training data in sufficient amounts. We present an approach for rapid and cost-effective annotation of training data for classification-based language understanding in conversational dialogue systems. Experiments using a web-accessible conversational character that interacts with a varied user population show that a dramatic improvement in natural language understanding and a substantial reduction in expert annotation effort can be achieved by leveraging non-expert annotation.
展开▼