Selective Token Generation for Few-shot Language Modeling

Natural language modeling with limited training data is challenging problem, and many algorithms make use of large-scale pretrained language models (PLMs) for this due to its great generalization ability. Among these transfer learning algorithms from PLMs, additive learning that incorporates a task-specific adapter on top of the fixed PLM has been popularly used to alleviate the severe overfitting problem in the few-shot setting. However, this added task-specific adapter is generally trained by maximum likelihood estimation that can easily suffer from the so-called exposure bias problem, especially in sequential text generation. Therefore, in this work, we develop a novel additive learning algorithm based on reinforcement learning (RL) for few-shot natural language generation (NLG) tasks. In particular, we propose to use a selective token generation between the transformer-based PLM and the task-specific adapter during both training and inference. This output token selection between the two generators allows the adapter to take into account only on the task-relevant parts in sequence generation, and therefore makes it more robust to overfitting as well as more stable in RL training. In addition, in order to obtain the complementary adapter from the PLM for each few-shot task, we exploit a separate selecting module that is also simultaneously trained using RL. Experimental results on various few-shot NLG tasks including data-to-text generation and text summarization demonstrate that the proposed selective token generation significantly outperforms the previous additive learning algorithms based on the PLMs.