Bridging the Gap Between Training and Inference of Bayesian Controllable Language Models
This work addresses the problem of controlling attributes like sentiment and topic in language generation for users needing precise text generation, representing an incremental improvement over existing BCLMs.
The paper tackles the mismatch between training and inference in Bayesian Controllable Language Models (BCLMs) for controllable language generation, proposing a 'Gemini Discriminator' that achieves new state-of-the-art results on sentiment and topic control tasks in both automatic and human evaluations.
Large-scale pre-trained language models have achieved great success on natural language generation tasks. However, it is difficult to control the pre-trained language models to generate sentences with the desired attribute such as topic and sentiment, etc. Recently, Bayesian Controllable Language Models (BCLMs) have been shown to be efficient in controllable language generation. Rather than fine-tuning the parameters of pre-trained language models, BCLMs use external discriminators to guide the generation of pre-trained language models. However, the mismatch between training and inference of BCLMs limits the performance of the models. To address the problem, in this work we propose a "Gemini Discriminator" for controllable language generation which alleviates the mismatch problem with a small computational cost. We tested our method on two controllable language generation tasks: sentiment control and topic control. On both tasks, our method reached achieved new state-of-the-art results in automatic and human evaluations.