Learning Symbolic Rules over Abstract Meaning Representations for Textual Reinforcement Learning

TL;DR

This paper introduces NESTA, a neuro-symbolic approach that combines semantic parsing and rule induction to create interpretable policies in text-based reinforcement learning, outperforming neural methods in generalization and data efficiency.

Contribution

The paper presents a novel neuro-symbolic framework, NESTA, that learns interpretable rules for text-based RL, improving generalization and reducing training data needs.

Findings

NESTA outperforms deep RL methods on benchmark games.

NESTA generalizes better to unseen games.

NESTA learns effectively from fewer interactions.

Abstract

Text-based reinforcement learning agents have predominantly been neural network-based models with embeddings-based representation, learning uninterpretable policies that often do not generalize well to unseen games. On the other hand, neuro-symbolic methods, specifically those that leverage an intermediate formal representation, are gaining significant attention in language understanding tasks. This is because of their advantages ranging from inherent interpretability, the lesser requirement of training data, and being generalizable in scenarios with unseen data. Therefore, in this paper, we propose a modular, NEuro-Symbolic Textual Agent (NESTA) that combines a generic semantic parser with a rule induction system to learn abstract interpretable rules as policies. Our experiments on established text-based game benchmarks show that the proposed NESTA method outperforms deep reinforcement learning-based techniques by achieving better generalization to unseen test games and learning from fewer training interactions.