XLNet: Generalized Autoregressive Pretraining for Language Understanding

TL;DR

XLNet introduces a generalized autoregressive pretraining method that models bidirectional contexts without masking, outperforming BERT on multiple NLP tasks by integrating permutation-based training and Transformer-XL ideas.

Contribution

It proposes a novel permutation-based autoregressive pretraining approach that captures bidirectional context and overcomes BERT's limitations, integrating Transformer-XL techniques.

Findings

Outperforms BERT on 20 NLP tasks

Achieves significant improvements in question answering and natural language inference

Demonstrates the effectiveness of permutation-based pretraining

Abstract

With the capability of modeling bidirectional contexts, denoising autoencoding based pretraining like BERT achieves better performance than pretraining approaches based on autoregressive language modeling. However, relying on corrupting the input with masks, BERT neglects dependency between the masked positions and suffers from a pretrain-finetune discrepancy. In light of these pros and cons, we propose XLNet, a generalized autoregressive pretraining method that (1) enables learning bidirectional contexts by maximizing the expected likelihood over all permutations of the factorization order and (2) overcomes the limitations of BERT thanks to its autoregressive formulation. Furthermore, XLNet integrates ideas from Transformer-XL, the state-of-the-art autoregressive model, into pretraining. Empirically, under comparable experiment settings, XLNet outperforms BERT on 20 tasks, often by a large margin, including question answering, natural language inference, sentiment analysis, and document ranking.