# Reasoning in Transformers -- Mitigating Spurious Correlations and Reasoning Shortcuts

arXiv: https://arxiv.org/abs/2403.11314 · HTML: https://mlsanity.com/paper/2403.11314/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 6 of its 24 references that are in the map.
- First posted on arXiv 17 Mar 2024.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 6 of 6)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Generative Language Modeling for Automated Theorem Proving](https://mlsanity.com/paper/2009.03393/) (2020)
- [PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Change](https://mlsanity.com/paper/2206.10498/) (2022)
- [Massive Exploration of Neural Machine Translation Architectures](https://mlsanity.com/paper/1703.03906/) (2017)
- [Neuro-Symbolic AI: An Emerging Class of AI Workloads and their Characterization](https://mlsanity.com/paper/2109.06133/) (2021)
- [Mathematical Reasoning via Self-supervised Skip-tree Training](https://mlsanity.com/paper/2006.04757/) (2020)

## Built on by (most cited first, 5 of 5)

- [Large Language Model Reasoning Failures](https://mlsanity.com/paper/2602.06176/) (2026)
- [FINEREASON: Evaluating and Improving LLMs' Deliberate Reasoning through Reflective Puzzle Solving](https://mlsanity.com/paper/2502.20238/) (2025)
- [IslamicLegalBench: Evaluating LLMs Knowledge and Reasoning of Islamic Law Across 1,200 Years of Islamic Pluralist Legal Traditions](https://mlsanity.com/paper/2602.21226/) (2026)
- [Integrating Expert Knowledge into Logical Programs via LLMs](https://mlsanity.com/paper/2502.12275/) (2025)
- [Long-Term Prediction Accuracy Improvement of Data-Driven Medium-Range Global Weather Forecast](https://mlsanity.com/paper/2407.01598/) (2024)

## Abstract (arXiv)

Transformer language models are neural networks used for a wide variety of tasks concerning natural language, including some that also require logical reasoning. However, a transformer model may easily learn spurious patterns in the data, short-circuiting actual reasoning. In this paper we investigate to what extent transformers can be trained to a) approximate reasoning in propositional logic while b) avoiding known reasoning shortcuts via spurious correlations in the training data. To do so, we use a dataset with known spurious correlation between truth and e.g. the number of rules in the problem. We augment the data with proofs, and train two models: a generative transformer, WP-BART, trained on problems and their whole proofs, and a neuro-symbolic model, SIP-BART, trained on individual proof steps and combining the generative transformer model BART with a symbolic proof checker. We find that SIP-BART succeeds in avoiding reasoning shortcuts, while WP-BART does not. For SIP-BART, we then identify a few remaining reasoning errors, not previously described in the literature, arising from using a pre-trained language model. These are qualitatively analysed to create a taxonomy of four different types of additional pitfalls.