Mapping the Course for Prompt-based Structured Prediction

2025-08-20

AI Summary

To address hallucination and logical inconsistency in large language models (LLMs) for structured prediction, this paper proposes a fine-tuning-free LLM–symbolic reasoning synergy framework. It dynamically integrates LLMs’ generative capabilities with symbolic reasoning modules—such as combinatorial optimization and constraint solving—guided by prompt engineering to elicit structured intermediate representations from the LLM, while the symbolic module enforces syntactic and logical consistency. Confidence calibration and lightweight, structure-targeted fine-tuning further enhance reliability. Experiments demonstrate substantial improvements in both predictive accuracy and structural consistency across diverse structured prediction tasks—including semantic parsing, constrained text generation, and logical form mapping—outperforming purely generative baselines. The results validate the effectiveness of the “generation + reasoning” paradigm in preserving LLM generalizability while improving trustworthiness. Moreover, the work unveils a novel pathway for co-optimizing prompting strategies and symbolic constraints to achieve robust, verifiable structured outputs.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Symbolic Representations
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

LLMs have been shown to be useful for a variety of language tasks, without requiring task-specific fine-tuning. However, these models often struggle with hallucinations and complex reasoning problems due to their autoregressive nature. We propose to address some of these issues, specifically in the area of structured prediction, by combining LLMs with combinatorial inference in an attempt to marry the predictive power of LLMs with the structural consistency provided by inference methods. We perform exhaustive experiments in an effort to understand which prompting strategies can effectively estimate LLM confidence values for use with symbolic inference, and show that, regardless of the prompting strategy, the addition of symbolic inference on top of prompting alone leads to more consistent and accurate predictions. Additionally, we show that calibration and fine-tuning using structured prediction objectives leads to increased performance for challenging tasks, showing that structured learning is still valuable in the era of LLMs.

Problem

• Addressing LLM hallucinations in structured prediction tasks
• Combining LLMs with combinatorial inference for consistency
• Evaluating prompting strategies for confidence estimation

Innovation

• Combining LLMs with combinatorial inference
• Using prompting strategies to estimate confidence
• Adding symbolic inference for consistent predictions