Scholarlink

<p>Given the extremely diverse landscape of languages in India, the need arises for democratising STEM education in the country sometime in the future through inclusive AI models that can handle multimodal, multilingual educational content. Most approaches deal only with English or simply monolingual frameworks. They do not align visual, symbolic and linguistic domains well, nor do they personalize the learning of native languages. Hence, comprehension, engagement and retention are limited, especially among non-English speaking regional learners. The current work presents an integrated AI pipeline which leverages Indic languages for STEM education through five novel components to tackle this issue. The Multi Modal Indic Cognitive Alignment Network (MICAN) aligns textual, visual, and symbolic content through a tri-attentional transformer architecture where deep multimodal grounding is enabled. The Contextual Phoneme-Semantic Error Calibration (C-PSEC) capability improves TTS/ASR by calibrating phoneme errors according to semantic significance, thereby improving auditory instruction. Indic STEM Concept Transfer Graph (ISCTG): This ensures the cross-lingual semantic preservation of STEM concepts through a graph-attention-based multilingual alignment framework. To personalize learning, the Adaptive Learning Indexing using Lexico-Cognitive Indicators (ALILCI) model generates personalized content paths using lexical complexity, cognitive load, and temporal patterns. Finally, it uses pedagogical clarity and syntactic integrity to evaluate instructional content and hence giving out actionable feedback under the terms STEM-PED Evaluation using Pragmatic-Syntactic Indic Metrics (SPESIM). Together, these models form a tightly coupled ecosystem that ensures semantic grounding, language-agnostic learning, pedagogical validity, and learner-specific adaptation. The results show an overall increase of 22.7% in STEM comprehension, 17.5% in retention, and 23% in instructional efficacy. In new conditions, this work represents a model for aid-seeking, high-tech and culturally sensitive STEM education in multilingual low-resource settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An AI-driven multimodal framework for language inclusive STEM education in indic languages

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Pradnya Patil,
  • Priyanka Deshmukh,
  • Minal Sonkar

摘要

Given the extremely diverse landscape of languages in India, the need arises for democratising STEM education in the country sometime in the future through inclusive AI models that can handle multimodal, multilingual educational content. Most approaches deal only with English or simply monolingual frameworks. They do not align visual, symbolic and linguistic domains well, nor do they personalize the learning of native languages. Hence, comprehension, engagement and retention are limited, especially among non-English speaking regional learners. The current work presents an integrated AI pipeline which leverages Indic languages for STEM education through five novel components to tackle this issue. The Multi Modal Indic Cognitive Alignment Network (MICAN) aligns textual, visual, and symbolic content through a tri-attentional transformer architecture where deep multimodal grounding is enabled. The Contextual Phoneme-Semantic Error Calibration (C-PSEC) capability improves TTS/ASR by calibrating phoneme errors according to semantic significance, thereby improving auditory instruction. Indic STEM Concept Transfer Graph (ISCTG): This ensures the cross-lingual semantic preservation of STEM concepts through a graph-attention-based multilingual alignment framework. To personalize learning, the Adaptive Learning Indexing using Lexico-Cognitive Indicators (ALILCI) model generates personalized content paths using lexical complexity, cognitive load, and temporal patterns. Finally, it uses pedagogical clarity and syntactic integrity to evaluate instructional content and hence giving out actionable feedback under the terms STEM-PED Evaluation using Pragmatic-Syntactic Indic Metrics (SPESIM). Together, these models form a tightly coupled ecosystem that ensures semantic grounding, language-agnostic learning, pedagogical validity, and learner-specific adaptation. The results show an overall increase of 22.7% in STEM comprehension, 17.5% in retention, and 23% in instructional efficacy. In new conditions, this work represents a model for aid-seeking, high-tech and culturally sensitive STEM education in multilingual low-resource settings.