Towards Faithful Chain-of-Thought: Large Language Models are Bridging Reasoners

2024-05-29

arXiv.org

AI Summary

This work addresses the instability of chain-of-thought (CoT) reasoning across diverse tasks and its frequent lack of faithfulness—i.e., generated answers not being fully supported by the intermediate reasoning steps. To this end, we introduce the first step-level causal modeling framework that explicitly distinguishes centralized versus distributed reasoning paradigms, uncovering causal dependencies among context, reasoning chains, and final answers. We propose “Inference Bridging”, a novel method integrating attribution analysis and semantic consistency as dual criteria to jointly optimize CoT filtering and prompt enhancement. Our pipeline comprises context retrieval, CoT generation, and step-aware re-ranking. Extensive experiments demonstrate significant improvements in both reasoning faithfulness and answer accuracy across multiple benchmarks, confirming effectiveness and strong cross-task generalization. The core contributions are: (1) step-granular causal modeling of CoT reasoning, and (2) a dual-criterion co-optimization framework for faithful, robust inference.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Reasoning with Beliefs

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs) suffer from serious unfaithful chain-of-thought (CoT) issues. Previous work attempts to measure and explain it but lacks in-depth analysis within CoTs and does not consider the interactions among all reasoning components jointly. In this paper, we first study the CoT faithfulness issue at the granularity of CoT steps, identify two reasoning paradigms: centralized reasoning and distributed reasoning, and find their relationship with faithfulness. Subsequently, we conduct a joint analysis of the causal relevance among the context, CoT, and answer during reasoning. The result proves that, when the LLM predicts answers, it can recall correct information missing in the CoT from the context, leading to unfaithfulness issues. Finally, we propose the inferential bridging method to mitigate this issue, in which we use the attribution method to recall information as hints for CoT generation and filter out noisy CoTs based on their semantic consistency and attribution scores. Extensive experiments demonstrate that our approach effectively alleviates the unfaithful CoT problem.

Problem

• Analyzing factors affecting Chain-of-Thought (CoT) effectiveness and faithfulness.
• Identifying issues with unfaithful CoT due to missing information in reasoning.
• Proposing an algorithm to improve CoT by enhancing information recall and evaluation.

Innovation

• Analyzes CoT effectiveness via problem difficulty, information gain, flow
• Identifies unfaithful CoT through question-CoT-answer interaction analysis
• Proposes algorithm enhancing CoT with extra question information recall