Reasoning-Grounded Natural Language Explanations for Language Models

2025-03-14

AI Summary

To address the low faithfulness and weak consistency between natural language explanations and answers generated by large language models (LLMs), this paper proposes the “Reasoning Anchoring” framework. It explicitly incorporates reasoning sequences as contextual input to jointly generate answers and explanations, thereby decoupling their mutual dependency. The core contributions are: (1) the first end-to-end grounding mechanism that maps latent reasoning processes to textual explanations; and (2) a novel autoregressive architecture featuring a joint prediction-explanation head and in-context reasoning encoding. Evaluated across multiple domains, our method achieves significant improvements—+12.3% in explanation faithfulness and +5.8% in answer accuracy—while enabling direct reuse of reasoning segments and ensuring tight alignment between answers and explanations.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Large Multimodal Models (LMMs)
• 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
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

We propose a large language model explainability technique for obtaining faithful natural language explanations by grounding the explanations in a reasoning process. When converted to a sequence of tokens, the outputs of the reasoning process can become part of the model context and later be decoded to natural language as the model produces either the final answer or the explanation. To improve the faithfulness of the explanations, we propose to use a joint predict-explain approach, in which the answers and explanations are inferred directly from the reasoning sequence, without the explanations being dependent on the answers and vice versa. We demonstrate the plausibility of the proposed technique by achieving a high alignment between answers and explanations in several problem domains, observing that language models often simply copy the partial decisions from the reasoning sequence into the final answers or explanations. Furthermore, we show that the proposed use of reasoning can also improve the quality of the answers.

Problem

• Develops a method for generating faithful natural language explanations.
• Uses reasoning sequences to improve answer and explanation alignment.
• Enhances answer quality by integrating reasoning into model outputs.

Innovation

• Grounding explanations in reasoning process
• Joint predict-explain approach for faithfulness
• Reasoning sequence improves answer quality