Grounding vs. Compositionality: On the Non-Complementarity of Reasoning in Neuro-Symbolic Systems

TL;DR

This paper empirically investigates the relationship between grounding and reasoning in neuro-symbolic AI, showing that grounding alone does not lead to generalization and that reasoning requires explicit training.

Contribution

It introduces the $i$LTN architecture and demonstrates that joint training on grounding and reasoning tasks is necessary for generalization.

Findings

Models trained only on grounding fail to generalize to new entities and relations.

The full $i$LTN achieves high zero-shot accuracy on reasoning tasks.

Grounding is necessary but not sufficient for reasoning generalization.

Abstract

Compositional generalization remains a foundational weakness of modern neural networks, limiting their robustness and applicability in domains requiring out-of-distribution reasoning. A central, yet unverified, assumption in neuro-symbolic AI is that compositional reasoning will emerge as a byproduct of successful symbol grounding. This work presents the first systematic empirical analysis to challenge this assumption by disentangling the contributions of grounding and reasoning. To operationalize this investigation, we introduce the Iterative Logic Tensor Network ($i$LTN), a fully differentiable architecture designed for multi-step deduction. Using a formal taxonomy of generalization -- probing for novel entities, unseen relations, and complex rule compositions -- we demonstrate that a model trained solely on a grounding objective fails to generalize. In contrast, our full $i$LTN, trained jointly on perceptual grounding and multi-step reasoning, achieves high zero-shot accuracy across all tasks. Our findings provide conclusive evidence that symbol grounding, while necessary, is insufficient for generalization, establishing that reasoning is not an emergent property but a distinct capability that requires an explicit learning objective.