Collaborate SLM and LLM with latent answers for event detection

Youcheng Yan, Jinshuo Liu, Donghong Ji, Jinguang Gu, Ahmed Abubakar Aliyu, Xinyan Wang, Jeff Z. Pan

Abstract: Highlights•We propose a novel event detection framework, namely LSLAED.•We compare the advantages and disadvantages of SLMs and LLMs and then combine them organically.•We construct three types of latent answers: answer-aware examples, structure-aware examples, and answer candidates.•Our approach avoids the dependence on large high-quality datasets and generates latent answers-enhanced prompts.•Our approach achieves improvements in supervised and low-resource event detection.