Unlocking Efficiency in Task-Oriented Dialogue Systems with Self-Training and Constrained Decoding

Introduction Task-oriented dialogue systems have become increasingly popular, thanks to advancements in natural language generation (NLG). These systems, however, often require substantial amounts of annotated data to generate coherent and contextual...