SAS: Self-Augmentation Strategy for Language Model Pre-training

TL;DR

This paper introduces SAS, a self-augmentation strategy that uses a single network for both pre-training and contextualized data augmentation, improving performance and reducing computational costs.

Contribution

SAS eliminates the need for a separate generator in data augmentation, simplifying the pre-training process and enhancing model performance across various NLP tasks.

Findings

SAS outperforms ELECTRA and other models on GLUE tasks.

SAS achieves these results with similar or reduced computation.

The strategy is compatible with recent techniques like disentangled attention.

Abstract

The core of self-supervised learning for pre-training language models includes pre-training task design as well as appropriate data augmentation. Most data augmentations in language model pre-training are context-independent. A seminal contextualized augmentation was recently proposed in ELECTRA and achieved state-of-the-art performance by introducing an auxiliary generation network (generator) to produce contextualized data augmentation for the training of a main discrimination network (discriminator). This design, however, introduces extra computation cost of the generator and a need to adjust the relative capability between the generator and the discriminator. In this paper, we propose a self-augmentation strategy (SAS) where a single network is utilized for both regular pre-training and contextualized data augmentation for the training in later epochs. Essentially, this strategy eliminates a separate generator and uses the single network to jointly conduct two pre-training tasks with MLM (Masked Language Modeling) and RTD (Replaced Token Detection) heads. It avoids the challenge to search for an appropriate size of the generator, which is critical to the performance as evidenced in ELECTRA and its subsequent variant models. In addition, SAS is a general strategy that can be seamlessly combined with many new techniques emerging recently or in the future, such as the disentangled attention mechanism from DeBERTa. Our experiments show that SAS is able to outperform ELECTRA and other state-of-the-art models in the GLUE tasks with similar or less computation cost.