Empirical Investigation of Neural Symbolic Reasoning Strategies
This work addresses the problem of enhancing neural reasoning models for symbolic tasks, but it is incremental as it builds on existing knowledge about intermediate steps without introducing a new paradigm.
The paper investigates how generating intermediate reasoning steps improves neural symbolic reasoning accuracy, finding that specific strategies like step granularity and chaining significantly boost performance, with some configurations achieving nearly perfect results even in length extrapolation scenarios.
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.