Linguistic competence, our capacity to produce and understand sentences and words, was given a mythical and grandiose status of being a uniquely human capacity that requires innate or hardwired rules and shapes the very structure of our thoughts. This account, which I will call The Inherited View (IV) is shared among the intertwined fields such as analytic philosophy, especially philosophy of language as its main subfield, theoretical linguistics, and cognitive science. The Cartesian internalist position (CI) constitutes the core of IV. According to CI, we are the sole authorities in cognizing our inner modes of presentation through which we grow to recognize and even name objects. A more literal, radical, and seemingly testable pairing of thought with linguistic expression came with Fodor’s Language-of-Thought Hypothesis (LOTH). Any methodology that rejects LOTH or a theoretical framework that doesn’t hinge on the CI qua ontological assumption about the nature of linguistic competence is inherently flawed, since it fails to mirror IV.

The resurgence of deep learning (DL) models for natural language processing (NLP) in the last decade, as well as large language models (LLMs) currently implemented in commercial chatbots, seems to provide a perspective on the nature of linguistic competence that runs counter to LOTH and discredits CI. These models arguably vindicate empiricist ambitions of grounding cognitive processes in raw data and different learning algorithms. In my talk, I will argue that the most disruptive element that the empiricist view of the nature of linguistic competence presents is the decoupling of language from thought. Methodologically, this opposes LOTH and may position philosophy of language as somewhat transformed by LLMs, but without jeopardizing its conceptual foundations. However, if one maintains that such an element must bring about full refutation of the CI, then philosophers of language should start considering the process of transmutation of their subfield, which means the IV has to go. Finally, I will outline two implications for the philosophy of language when it comes to the endorsement of the novel empiricist view of the nature of linguistic competence based on the performance of LLMs.

Vanja Subotić earned her Ph.D. from the University of Belgrade and presently works as a Research Associate at the Institute of Philosophy within the Faculty of Philosophy of the same university. She holds MA and BA in Philosophy (cum laude) and MSc in Computational Methods for Social Sciences and Humanities. Subotić was an EPSA Research Fellow at the University of Turin and a visiting lecturer at the University of Rijeka. Her areas of expertise encompass the philosophy of cognitive neuroscience, philosophy of AI, general methodology of science, and experimental philosophy of language. So far, Subotić published one edited volume and her papers appeared in venues like Synthese, The Journal of Value Inquiry, Review of Philosophy and Psychology, etc. She served as a consultant for the HORIZON 2020 project TechEthos, offering insights into ethical challenges associated with emerging technologies, most notably chatbots and conversational AI. Subotić is also a certified scientific communicator and maintains a close collaboration with the Center for the Promotion of Science in Belgrade.