A Cognitive Neural Architecture Able to Learn and Communicate through Natural Language

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Human communication relies on procedural knowledge for processing and producing language.
• Existing models struggle to integrate language processing with neural architecture from a tabula rasa state.
• Understanding the neural basis of language acquisition remains a significant challenge.

Purpose of the Study:

• To develop a comprehensive cognitive system based on neural architecture for language elaboration.
• To model the procedural knowledge involved in processing verbal and nonverbal communication.
• To create a tabula rasa system capable of learning language through interaction.

Main Methods:

• Developed a large-scale neural architecture with a central executive controlling working memory components.
• Implemented a neural network for the central executive, processing short-term memory states to generate mental actions.
• Utilized an open-ended, incremental learning process via a text-based interface for human interaction.

Main Results:

• The system learned to communicate in natural language without prior linguistic knowledge.
• Successfully acquired and utilized various word classes, including nouns, verbs, and adjectives.
• Validated on a corpus, demonstrating capabilities comparable to a 4-year-old child's language assessment.

Conclusions:

• The proposed neural system effectively models procedural knowledge for language acquisition.
• Demonstrates the feasibility of learning complex language skills from a tabula rasa state through interaction.
• Offers insights into the neural mechanisms underlying human language processing and development.