Dynamic Chain-of-thought for Low-Resource Event Extraction

ACL ARR 2024 December Submission352 Authors

Abstract:

Event extraction is a critical task that encompasses several interdependent sub-tasks. The complex interplay among these sub-tasks renders the overall task exceedingly challenging, particularly in low-resource scenarios where data availability is limited.
However, the inherent logical coherence among these sub-tasks presents a promising avenue for addressing these challenges.
This logical structure is particularly advantageous in low-resource settings, as it facilitates a deeper understanding of the tasks by the model and reduces dependence on available data.
Building on this observation, we explore the logical structure of event extraction with a focus on low-resource scenarios. Specifically, we propose a three-step Chain-of-Thought pattern to guide the model through the logical reasoning process. Additionally, we design a step-wise navigator that dynamically provides the model with relevant knowledge.
Empirical results demonstrate the robustness of our approach in low-resource event extraction.

Paper Type: Long

Research Area: Information Extraction

Research Area Keywords: Efficient/Low-Resource Methods for NLP

Contribution Types: NLP engineering experiment, Approaches to low-resource settings

Languages Studied: English

Submission Number: 352