What Would Happen Next? Predicting Consequences from An Event Causality Graph

TL;DR

This work tackles the challenge of predicting subsequent events from historical data by arguing that an Event Causality Graph (ECG) provides richer priors than conventional event-script chains. It introduces SeDGPL, a distance-sensitive graph prompt model with three components—Distance-sensitive Graph Linearization, Event-Enriched Causality Encoding, and Semantic Contrast Event Prediction—that linearize ECGs into prompts, fuse event-level semantic and schema information, and apply semantic-contrast learning to select the ground-truth consequent from a large candidate set. Experiments on CGEP-MAVEN and CGEP-ESC show that ECG-based predictions with SeDGPL outperform script-based methods and several baselines, including large language models, underscoring the value of graph-structured priors for complex historical reasoning. The approach advances event forecasting by integrating causal graph structure with prompt-based PLMs, enabling more accurate near-future predictions in domains like dialogue, discourse understanding, and story generation.

Abstract

Existing script event prediction task forcasts the subsequent event based on an event script chain. However, the evolution of historical events are more complicated in real world scenarios and the limited information provided by the event script chain also make it difficult to accurately predict subsequent events. This paper introduces a Causality Graph Event Prediction(CGEP) task that forecasting consequential event based on an Event Causality Graph (ECG). We propose a Semantic Enhanced Distance-sensitive Graph Prompt Learning (SeDGPL) Model for the CGEP task. In SeDGPL, (1) we design a Distance-sensitive Graph Linearization (DsGL) module to reformulate the ECG into a graph prompt template as the input of a PLM; (2) propose an Event-Enriched Causality Encoding (EeCE) module to integrate both event contextual semantic and graph schema information; (3) propose a Semantic Contrast Event Prediction (ScEP) module to enhance the event representation among numerous candidate events and predict consequential event following prompt learning paradigm. %We construct two CGEP datasets based on existing MAVEN-ERE and ESC corpus for experiments. Experiment results validate our argument our proposed SeDGPL model outperforms the advanced competitors for the CGEP task.

Figures (6)

• Figure 1: Illustration of the Causality Graph Event Prediction task, forecasting consequential event based on an Event Causality Graph.
• Figure 2: Data Processing Flowchart: The data processing involves transforming an original ECG into multiple data instances, with each instance specifically predicting a single consequential event.
• Figure 3: The SeDGPL model consists of three modules: (1) Distance-sensitive Graph Linearization (DsGL); (2) Event-Enriched Causality Encoding (EeCE); (3) Semantic Contrast Event Prediction (ScEP).
• Figure 4: Results on CGEP-MAVEN with different loss ratio $\beta$.
• Figure 5: A case of SeDGPL on causality graph event prediction and causality script event prediction tasks.