Lifelong Pretraining: Continually Adapting Language Models to Emerging Corpora

TL;DR

This paper explores continual pretraining of language models to adapt to new data streams while retaining prior knowledge, demonstrating effective distillation methods for better temporal generalization and knowledge transfer.

Contribution

It introduces a lifelong pretraining framework with continual learning algorithms, evaluating their effectiveness on domain-incremental and chronological data streams.

Findings

Distillation-based methods best retain earlier domain performance.

Algorithms improve adaptation to new data and knowledge transfer.

Enhanced temporal generalization in distribution-shift scenarios.

Abstract

Pretrained language models (PTLMs) are typically learned over a large, static corpus and further fine-tuned for various downstream tasks. However, when deployed in the real world, a PTLM-based model must deal with data distributions that deviate from what the PTLM was initially trained on. In this paper, we study a lifelong language model pretraining challenge where a PTLM is continually updated so as to adapt to emerging data. Over a domain-incremental research paper stream and a chronologically-ordered tweet stream, we incrementally pretrain a PTLM with different continual learning algorithms, and keep track of the downstream task performance (after fine-tuning). We evaluate PTLM's ability to adapt to new corpora while retaining learned knowledge in earlier corpora. Our experiments show distillation-based approaches to be most effective in retaining downstream performance in earlier domains. The algorithms also improve knowledge transfer, allowing models to achieve better downstream performance over the latest data, and improve temporal generalization when distribution gaps exist between training and evaluation because of time. We believe our problem formulation, methods, and analysis will inspire future studies towards continual pretraining of language models.