Seminar Artificial Intelligence - Prof. Max Louwerse, Tilburg University

Title: How language creates meaning: Implications for NLP, AI, Cognitive Science (and Rocket Science)

Abstract:

It is often stated that LLMs (Large Language Models) such as ChatGPT excel in converting unstructured data into structured insights. For LMMs to accomplish these impressive results, massive amounts of data are trained on a transformer deep learning model with massive numbers of parameters using massive computing power. But are LMMs really using unstructured data? And how come that humans are able to understand language with far less amounts of training data? Might the answers to these questions have to make us rethink LLMs? This talk will argue that the answer to the question how language attains meaning may not come so much from the computational power of LLMs, or the language instinct in humans, but from within the language system itself. Using examples from phonology, syntax and semantic – specifically word embeddings – the argument is made that language creates meaning. Such an argument would call for revisiting some common principles advocated in theoretical, computational and psycholinguistics, and has far reaching consequences for NLP, AI and the cognitive sciences at large.