EEHF: Learning Event Argument Extraction from Hybrid Feedback

The event argument extraction (EAE) task is aim to extract a structured event record from an event text. Most existing approaches need learn from well-labeled training set. Due to the expensive and time-consuming manual annotation cost, insufficient training data seriously hinders the EAE model from achieving better performance. In this paper, we propose Event argument Extraction with Hybrid Feedback (EEHF), which can use reinforcement learning to learn EAE task on unsupervised corpus. The hybrid feedback containing correctness feedback, reasonableness feedback and validity feedback to reward the model in various dimensions. The experiments show EEHF achieves new state-of-the-art on three benchmarks with a significant margin in the EAE task. Hybrid feedback can effectively align language models to EAE tasks.