BERTs are Generative In-Context Learners

TL;DR

This paper shows that masked language models like DeBERTa can perform in-context learning and generative tasks without extra training, revealing complementary strengths with causal models and suggesting hybrid approaches.

Contribution

Demonstrates that masked language models can exhibit in-context learning and generative abilities using a simple inference method, challenging the focus on causal models.

Findings

Masked models outperform causal models on certain tasks

Causal models excel in different categories of tasks

Hybrid approaches could leverage strengths of both architectures

Abstract

While in-context learning is commonly associated with causal language models, such as GPT, we demonstrate that this capability also 'emerges' in masked language models. Through an embarrassingly simple inference technique, we enable an existing masked model, DeBERTa, to perform generative tasks without additional training or architectural changes. Our evaluation reveals that the masked and causal language models behave very differently, as they clearly outperform each other on different categories of tasks. These complementary strengths suggest that the field's focus on causal models for in-context learning may be limiting - both architectures can develop these capabilities, but with distinct advantages; pointing toward promising hybrid approaches that combine the strengths of both objectives.