Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Abductive Inference in Retrieval-Augmented Language Models: Generating and Validating Missing Premises

Shiyin Lin

TL;DR

The paper tackles the problem of incomplete evidence in retrieval-augmented language models by introducing abductive inference to generate and validate missing premises that bridge reasoning gaps. It defines a formal task where a missing premise $P$ is sought such that $E \wedge P \vdash A$, and presents a four-stage abductive-RAG pipeline: insufficiency detection, abductive premise generation, premise validation, and answer generation. Empirical results on abductive reasoning and multi-hop QA benchmarks show consistent improvements over strong baselines, including a $+7.2$ EM gain on EntailmentBank and a HotpotQA F1 of $75.3$, indicating better robustness and faithfulness. The work advances explainability in RAG by making the reasoning path explicit through abductively generated premises that are tested for consistency and plausibility before yielding answers, suggesting practical benefits for knowledge-intensive NLP tasks.

Abstract

Large Language Models (LLMs) enhanced with retrieval -- commonly referred to as Retrieval-Augmented Generation (RAG) -- have demonstrated strong performance in knowledge-intensive tasks. However, RAG pipelines often fail when retrieved evidence is incomplete, leaving gaps in the reasoning process. In such cases, \emph{abductive inference} -- the process of generating plausible missing premises to explain observations -- offers a principled approach to bridge these gaps. In this paper, we propose a framework that integrates abductive inference into retrieval-augmented LLMs. Our method detects insufficient evidence, generates candidate missing premises, and validates them through consistency and plausibility checks. Experimental results on abductive reasoning and multi-hop QA benchmarks show that our approach improves both answer accuracy and reasoning faithfulness. This work highlights abductive inference as a promising direction for enhancing the robustness and explainability of RAG systems.

Abductive Inference in Retrieval-Augmented Language Models: Generating and Validating Missing Premises

TL;DR

The paper tackles the problem of incomplete evidence in retrieval-augmented language models by introducing abductive inference to generate and validate missing premises that bridge reasoning gaps. It defines a formal task where a missing premise is sought such that , and presents a four-stage abductive-RAG pipeline: insufficiency detection, abductive premise generation, premise validation, and answer generation. Empirical results on abductive reasoning and multi-hop QA benchmarks show consistent improvements over strong baselines, including a EM gain on EntailmentBank and a HotpotQA F1 of , indicating better robustness and faithfulness. The work advances explainability in RAG by making the reasoning path explicit through abductively generated premises that are tested for consistency and plausibility before yielding answers, suggesting practical benefits for knowledge-intensive NLP tasks.

Abstract

Large Language Models (LLMs) enhanced with retrieval -- commonly referred to as Retrieval-Augmented Generation (RAG) -- have demonstrated strong performance in knowledge-intensive tasks. However, RAG pipelines often fail when retrieved evidence is incomplete, leaving gaps in the reasoning process. In such cases, \emph{abductive inference} -- the process of generating plausible missing premises to explain observations -- offers a principled approach to bridge these gaps. In this paper, we propose a framework that integrates abductive inference into retrieval-augmented LLMs. Our method detects insufficient evidence, generates candidate missing premises, and validates them through consistency and plausibility checks. Experimental results on abductive reasoning and multi-hop QA benchmarks show that our approach improves both answer accuracy and reasoning faithfulness. This work highlights abductive inference as a promising direction for enhancing the robustness and explainability of RAG systems.

Paper Structure

This paper contains 23 sections, 5 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Retrieval-Augmented Generation
  4. Abductive and Multi-hop Reasoning
  5. Premise Validation and Context Modeling
  6. Theoretical Perspectives
  7. Premise Validation and Faithfulness
  8. Problem Definition
  9. Insufficiency Detection
  10. Abductive Premise Generation
  11. Premise Validation
  12. Answer Generation
  13. Experiments
  14. Datasets
  15. Baselines
  16. ...and 8 more sections

Figures (2)

  • Figure 1: Abductive-RAG pipeline. The system detects insufficiency, abductively generates candidate premises, validates them via entailment and retrieval plausibility, selects $p^{*}$, and answers with $(Q,E,p^{*})$. Dashed arrows denote optional or shortcut paths.
  • Figure 2: Case study comparing Baseline RAG and Abductive-RAG. Our method generates and validates a missing premise $p^{*}$ to bridge incomplete evidence, avoiding hallucination and yielding a supported answer.