Empirical Investigation of Neural Symbolic Reasoning Strategies

TL;DR

This paper empirically examines how different neural symbolic reasoning strategies, especially regarding step granularity and chaining, impact reasoning accuracy and extrapolation performance.

Contribution

It systematically analyzes the effects of reasoning strategies on neural symbolic reasoning, revealing configurations that achieve near-perfect performance.

Findings

Reasoning strategy choice significantly affects performance.

Longer extrapolation lengths increase performance gaps.

Certain strategies enable near-perfect extrapolation performance.

Abstract

Neural reasoning accuracy improves when generating intermediate reasoning steps. However, the source of this improvement is yet unclear. Here, we investigate and factorize the benefit of generating intermediate steps for symbolic reasoning. Specifically, we decompose the reasoning strategy w.r.t. step granularity and chaining strategy. With a purely symbolic numerical reasoning dataset (e.g., A=1, B=3, C=A+3, C?), we found that the choice of reasoning strategies significantly affects the performance, with the gap becoming even larger as the extrapolation length becomes longer. Surprisingly, we also found that certain configurations lead to nearly perfect performance, even in the case of length extrapolation. Our results indicate the importance of further exploring effective strategies for neural reasoning models.