Воплощенный ИИ и нейровычислительная семантика

2026/2027

ID 1121860

Статус: Маго-лего

Кто читает: Департамент анализа данных и искусственного интеллекта

Где читается: Факультет компьютерных наук

Охват аудитории: для своего кампуса

Преподаватели: Скрынникова Инна Валериевна

Язык: русский

Кредиты: 6

Контактные часы: 40

Дополнительные материалы в LMS Задать вопрос

Аннотация

This graduate course explores the convergence of embodied cognition, artificial intelligence, computational linguistics and neurocomputational approaches to semantic representation. Grounded in the premise that meaning is rooted in sensorimotor experience, the course introduces Embodied Construction Grammar (ECG) as a formal model for representing language as structured simulations of perceptual and action-based schemas. Through a series of lectures and interactive seminars, students examine how intelligent agents—both biological and artificial—ground linguistic meaning in sensorimotor experience and environmental interaction. The curriculum critically evaluates contemporary paradigms such as language-conditioned reinforcement learning, multimodal foundation models with embodied interfaces that acquire semantics through physical interaction. Seminar sessions emphasize hands-on analysis of experimental methodologies, replication studies from embodied AI research. The course equips students to design and evaluate AI systems whose semantic capabilities emerge from situated interaction rather than purely statistical pattern recognition. Learning outcomes include the ability to critique disembodied language models, implement simple embodied semantic architectures, and articulate the theoretical foundations linking bodily experience to representational content in both natural and artificial intelligence.

Цель освоения дисциплины

The course seeks to provide students with theoretical knowledge as well as hands-on skills in formalizing cognitive-linguistic theories into executable computational models, implementing simulation-based semantic engines that map sensorimotor schemas to abstract concepts, and integrating these neurocomputational architectures with contemporary embodied AI frameworks for language-conditioned robotic control and multimodal reasoning.

Планируемые результаты обучения

formalizing cognitive-linguistic theories into executable computational models, including writing construction schemas that map sensorimotor schemas to linguistic constructions;
implementing simulation-based semantic engines that ground abstract concepts in perceptual and action-based schemas, moving beyond symbolic or purely distributional representations;
designing and evaluating neuro-symbolic architectures that integrate formal semantic representations with multimodal foundation models and language-conditioned robotic agents;
critiquing and comparing disembodied vs. embodied language models, identifying the theoretical and empirical limitations of purely statistical approaches to semantic representation;
conducting replication studies and experimental methodologies from embodied AI and cognitive linguistics research, including hands-on analysis of simulation-based language understanding systems;
articulating the theoretical foundations linking neural computation, bodily experience, and representational content in both biological and artificial cognitive systems.

Содержание учебной дисциплины

Module 1: Foundations of Embodied Cognition and Neural Semantics Introduction to Embodied Cognition and the Neural Theory of Language (NTL)
Image Schemas and Spatial Primitives as Semantic Building Blocks
Neural Computation and the Biological Substrates of Meaning
Conceptual Metaphor Theory and Cross-Domain Mappings
Foundations of Embodied Construction Grammar (ECG): Syntax, Semantics, and Unification
Simulation-Based Semantic Engines: Grounding Language in Perceptual-Motor Routines
Module 2: Computational Modeling, Implementation, and Embodied AI Integration. Implementing ECG Grammars: Tools and Formalisms
Neuro-Symbolic Architectures for Language and Action Integration
Embodied AI in Practice: Language-Conditioned Robotics and Multimodal Agents
Comparative Evaluation, Open Challenges, and Future Directions

Элементы контроля

HW_1_ECG
Test_M1_Theory
Sem_Active
HW_2_Simulation
Proj_Final

Промежуточная аттестация

2026/2027 3rd module
Final = 0.15 × HW_1_ECG + 0.15 × Test_M1_Theory + 0.10 × Sem_Active + 0.25 × HW_2_Simulation + 0.35 × Proj_Final

Список литературы

Авторы

Скрынникова Инна Валериевна
Антропова Лариса Ивановна

Программа дисциплины