Year: 2026
Type: article
Abstract: In this paper, I deal with the growing role of Large Language Models (LLMs) in linguistic research, asking if their use represents just a methodological innovation or a broader paradigmatic change. Drawing on the distinction between corpus-based and corpus-driven linguistics, I propose three approaches to AI-assisted language research: LLM-based, LLM-driven, and LLM-native linguistics. In LLM-based research, models serve as tools for analyzing linguistic data within previously defined questions and hypotheses. In LLM-driven research, patterns identified by the model provide the basis for formulating new research questions and hypotheses. Both approaches constitute methodological innovations, since the object of inquiry remains human language. LLM-native linguistics, by contrast, treats LLM-generated texts and the models themselves as autonomous objects of linguistic investigation. This approach raises important questions about the status of AI-generated texts and the boundaries of linguistic inquiry. I believe that LLM-native linguistics may change what linguistics recognizes as its object of study.
Author: Daniel Słapek
Institution: Jagiellonian University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7226035904
