Beyond In-Distribution Success: Scaling Curves of CoT Granularity for Language Model Generalization

TL;DR

This paper demonstrates that Chain-of-Thought reasoning improves language model generalization to out-of-distribution compound tasks, especially with finer granularity and less data, supported by theoretical and empirical analysis.

Contribution

It reveals the importance of CoT data granularity and sample efficiency for better OOD generalization, and provides theoretical insights into why CoT outperforms QA training.

Findings

Finer-grained CoT data enhances generalization performance.

CoT achieves similar accuracy with significantly less data compared to QA.

Transformer positional embeddings can further improve CoT generalization.

Abstract

Generalization to novel compound tasks under distribution shift is important for deploying transformer-based language models (LMs). This work investigates Chain-of-Thought (CoT) reasoning as a means to enhance OOD generalization. Through controlled experiments across several compound tasks, we reveal three key insights: (1) While QA-trained models achieve near-perfect in-distribution accuracy, their OOD performance degrades catastrophically, even with 10000k+ training examples; (2) the granularity of CoT data strongly correlates with generalization performance; finer-grained CoT data leads to better generalization; (3) CoT exhibits remarkable sample efficiency, matching QA performance with much less (even 80%) data. Theoretically, we demonstrate that compound tasks inherently permit shortcuts in Q-A data that misalign with true reasoning principles, while CoT forces internalization of valid dependency structures, and thus can achieve better generalization. Further, we show that transformer positional embeddings can amplify generalization by emphasizing subtask condition recurrence in long CoT sequences. Our combined theoretical and empirical analysis provides compelling evidence for CoT reasoning as a crucial training paradigm for enabling LM generalization under real-world distributional shifts for compound tasks.