Language Model Priming for Cross-Lingual Event Extraction

TL;DR

This paper introduces a language-agnostic priming method for language models to improve cross-lingual event extraction, especially in low-resource and zero-shot scenarios, by providing task-specific prompts at runtime.

Contribution

It proposes a novel priming technique that enhances language models' ability to perform event extraction across languages without additional training.

Findings

Significant improvement in trigger detection accuracy

Enhanced argument classification in low-resource languages

Effective zero-shot cross-lingual performance

Abstract

We present a novel, language-agnostic approach to "priming" language models for the task of event extraction, providing particularly effective performance in low-resource and zero-shot cross-lingual settings. With priming, we augment the input to the transformer stack's language model differently depending on the question(s) being asked of the model at runtime. For instance, if the model is being asked to identify arguments for the trigger "protested", we will provide that trigger as part of the input to the language model, allowing it to produce different representations for candidate arguments than when it is asked about arguments for the trigger "arrest" elsewhere in the same sentence. We show that by enabling the language model to better compensate for the deficits of sparse and noisy training data, our approach improves both trigger and argument detection and classification significantly over the state of the art in a zero-shot cross-lingual setting.