Generative Zero-Shot Prompt Learning for Cross-Domain Slot Filling with Inverse Prompting

TL;DR

This paper introduces a generative zero-shot prompt learning framework with inverse prompting for cross-domain slot filling, significantly improving generalization, robustness, and performance on unseen slots.

Contribution

It proposes a novel inverse prompting strategy and an efficient prompt-tuning method to enhance zero-shot cross-domain slot filling performance.

Findings

Achieves +13.44% F1 improvement on unseen slots

Demonstrates better generalization and robustness than previous models

Effective in distinguishing slot types with inverse prompting

Abstract

Zero-shot cross-domain slot filling aims to transfer knowledge from the labeled source domain to the unlabeled target domain. Existing models either encode slot descriptions and examples or design handcrafted question templates using heuristic rules, suffering from poor generalization capability or robustness. In this paper, we propose a generative zero-shot prompt learning framework for cross-domain slot filling, both improving generalization and robustness than previous work. Besides, we introduce a novel inverse prompting strategy to distinguish different slot types to avoid the multiple prediction problem, and an efficient prompt-tuning strategy to boost higher performance by only training fewer prompt parameters. Experiments and analysis demonstrate the effectiveness of our proposed framework, especially huge improvements (+13.44% F1) on the unseen slots.