Segatron: Segment-Aware Transformer for Language Modeling and Understanding

TL;DR

Segatron introduces a segment-aware positional encoding mechanism to Transformer models, enhancing their contextual understanding and performance across language modeling and NLP tasks.

Contribution

The paper proposes a novel segment-aware encoding method for Transformers, improving language model perplexity and NLP task performance over standard models.

Findings

Achieves 17.1 perplexity on WikiText-103 with Transformer-XL.

SegaBERT outperforms vanilla BERT on multiple NLP tasks.

Outperforms RoBERTa in zero-shot sentence representation learning.

Abstract

Transformers are powerful for sequence modeling. Nearly all state-of-the-art language models and pre-trained language models are based on the Transformer architecture. However, it distinguishes sequential tokens only with the token position index. We hypothesize that better contextual representations can be generated from the Transformer with richer positional information. To verify this, we propose a segment-aware Transformer (Segatron), by replacing the original token position encoding with a combined position encoding of paragraph, sentence, and token. We first introduce the segment-aware mechanism to Transformer-XL, which is a popular Transformer-based language model with memory extension and relative position encoding. We find that our method can further improve the Transformer-XL base model and large model, achieving 17.1 perplexity on the WikiText-103 dataset. We further investigate the pre-training masked language modeling task with Segatron. Experimental results show that BERT pre-trained with Segatron (SegaBERT) can outperform BERT with vanilla Transformer on various NLP tasks, and outperforms RoBERTa on zero-shot sentence representation learning.