Symbolic Mediation of Language-Based Decision Support in Tactical Contexts

Abstract

Language-based AI systems are increasingly explored as decision-support tools in tactical and operational contexts, where timely interpretation and action are critical. While recent advances enable language models to generate contextually grounded responses, their outputs are typically delivered as free-form text, placing interpretive burden on human decision-makers operating under time pressure. This paper addresses this representational gap by introducing a neuro-symbolic approach that mediates language model outputs through symbolic representations. We present a system architecture that transforms outputs from an already contextualized language model into standardized operational symbols using a symbolic inference layer grounded in rule-based mappings and confidence constraints. Rather than replacing language-based explanations, the system supports dual-mode presentation, preserving textual reasoning while rendering salient decision-relevant elements as symbols familiar to operational practice. This work contributes a representational framework for human-centered AI decision support in time-sensitive environments.