Variational model for low-resource natural language generation in spoken dialogue systems

Natural Language Generation (NLG) plays a critical role in Spoken Dialogue Systems (SDSs), aims at converting meaning representation into natural language utterances. Recent deep learning-based generators have shown improving results irrespective of providing sufficient annotated data. Nevertheless, how to build generator that can effectively utilize as much knowledge from low-resource setting data is crucial issue for NLG SDSs. This paper presents variational-based framework tackle the problem having limited two scenarios, domain adaptation and in-domain training Based on this framework, we propose novel adversarial taclking former issue, while latter also handled by second proposed dual variational model. We extensively conducted experiments four different domains variety which experimental show methods not only outperform previous when dataset but its ability work acceptably well there small amount or adapt quickly new with target