ABSTRACT
This chapter explores how conversational AI systems capture and process meanings through word embeddings, representing words as coordinates in a vector space. This approach allows AI to process synonyms and simulate speech that appears to us as meaningful. The chapter discusses distributional semantics, highlighting how words in similar contexts share meanings, and touches on the numerical basis of neural networks that generate semantic embeddings. These concepts are illustrated through a simple example involving common desktop items. This discussion emphasises how AI captures linguistic subtleties, setting the stage for an exploration of context in LLMs and parallels in urban environments.
