Neuro-Symbolic WCAG-Compliant Polyphone Disambiguation:Vision-Impaired-Accessible RoBERTa with Dynamic Phonemic Grounding and Haptic-Auditory Synchronization

Abstract

Chinese polyphone disambiguation remains a critical challenge for NLP and assistive technologies, particularly for 285 million visually impaired users facing digital exclusion. This study pioneers a neuro-symbolic framework integrating cognitive computing, speech synthesis, and perceptual interfaces to overcome limitations in semantic-phonemic alignment, linguistic adaptability, and accessibility compliance. Our innovations include: (1) A dynamic phoneme knowledge graph enabling probabilistic coupling between semantic roles (agent/patient) and tonal patterns, adapting to linguistic evolution (e.g., post-pandemic semantic shifts of "冠[guān/guàn]"); (2) A 32-dimensional topological radical encoder decomposing 214 Chinese radicals into morphologically informed vectors, fused via conditional masking for hierarchical phoneme-character-morpheme interactions; (3) A tactile-auditory cross-channel interface reducing focus-switching latency from 650ms to 89ms through Ebbinghaus curve-optimized rhythm; (4) The first ISO 9241-171-aligned NLP paradigm achieving WCAG 2.1 AA compliance via WAI-ARIA annotations and keyboard-TTS synchronization.Evaluated on the Chinese Polyphones with Pinyin (CPP) benchmark, our RoBERTa-BiLSTM hybrid achieves SOTA performance (96.23% accuracy, 85.63% F1-score) with 83ms inference latency. The system demonstrates 200% font scalability and <200ms response under JAWS/NVDA tests, establishing a new standard for high-reliability applications (e.g., medical/legal domains) where semantic fidelity is paramount. This work bridges symbolic grounding in language models with cognitive accessibility, advancing assistive technology through multimodal fusion and dynamic linguistic adaptation.