Simulating Action Dynamics with Neural Process Networks

TL;DR

This paper introduces Neural Process Networks, a model that simulates action dynamics in procedural text to better understand unstated causal effects, improving reasoning and interpretability.

Contribution

The paper presents Neural Process Networks, a novel approach that explicitly models actions as state transformers for procedural text understanding.

Findings

Improved reasoning about unstated causal effects.

More accurate contextual understanding of procedural text.

Enhanced interpretability of internal representations.

Abstract

Understanding procedural language requires anticipating the causal effects of actions, even when they are not explicitly stated. In this work, we introduce Neural Process Networks to understand procedural text through (neural) simulation of action dynamics. Our model complements existing memory architectures with dynamic entity tracking by explicitly modeling actions as state transformers. The model updates the states of the entities by executing learned action operators. Empirical results demonstrate that our proposed model can reason about the unstated causal effects of actions, allowing it to provide more accurate contextual information for understanding and generating procedural text, all while offering more interpretable internal representations than existing alternatives.