UniLMv2: Pseudo-Masked Language Models for Unified Language Model Pre-Training

TL;DR

UniLMv2 introduces a novel pseudo-masked language modeling pre-training method that unifies autoencoding and autoregressive tasks, leading to state-of-the-art performance on various NLP benchmarks.

Contribution

The paper presents a new pre-training approach called PMLM that combines autoencoding and autoregressive training within a single unified model architecture.

Findings

Achieves new state-of-the-art results on multiple NLP benchmarks.

Effectively unifies bidirectional encoding and sequence-to-sequence decoding.

Reduces redundant computation through shared context encodings.

Abstract

We propose to pre-train a unified language model for both autoencoding and partially autoregressive language modeling tasks using a novel training procedure, referred to as a pseudo-masked language model (PMLM). Given an input text with masked tokens, we rely on conventional masks to learn inter-relations between corrupted tokens and context via autoencoding, and pseudo masks to learn intra-relations between masked spans via partially autoregressive modeling. With well-designed position embeddings and self-attention masks, the context encodings are reused to avoid redundant computation. Moreover, conventional masks used for autoencoding provide global masking information, so that all the position embeddings are accessible in partially autoregressive language modeling. In addition, the two tasks pre-train a unified language model as a bidirectional encoder and a sequence-to-sequence decoder, respectively. Our experiments show that the unified language models pre-trained using PMLM achieve new state-of-the-art results on a wide range of natural language understanding and generation tasks across several widely used benchmarks.