Reinforcement Learning for Few-Shot Text Generation Adaptation

Controlling the generative model to adapt a new domain with limited samples is a difficult challenge and it is receiving increasing attention. Recently, few-shot learning has shown promising process in domain adaptation. However, the texts generated by few-shot learning are typically devoid of linguistic diversity. To address this shortcoming, we frame the adaptation of text generation systems as a reinforcement learning problem and provide a new approach to make text generation models easily adaptable to target domain with the minimal amount of in-domain data. Experimental results on five target domains in two few-shot configurations demonstrate that our method significantly outperforms domain adaptation when very few in-domain samples are available.