Language Model Priming for Cross-Lingual Event Extraction
This addresses the problem of event extraction for languages with sparse or noisy training data, offering a novel approach that enhances cross-lingual performance, though it is incremental in its method adaptation.
The paper tackled event extraction in low-resource and zero-shot cross-lingual settings by introducing a language-agnostic priming method that augments input based on specific triggers, resulting in significant improvements in trigger and argument detection and classification over state-of-the-art methods.
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.