ChatGPT as a COBUILD lexicographer
Robert Lew
Abstract
Abstract Creating high-quality dictionary entries has been a meticulous and time-consuming process, but current developments in Artificial Intelligence (AI) could help to speed up this process. This study explores the use of ChatGPT to create COBUILD-style English monolingual entries and evaluates its ability to generate accurate and high-quality sense definitions and example sentences on a sample of fifteen verbs of communication. AI-generated entries are produced using ChatGPT Plus, and human experts evaluate them alongside COBUILD entries in a blinded randomized process. The quality of sense definitions, example sentences, and overall entries are assessed using a five-point scale. The results indicate that AI-generated definitions are comparable to those created by human lexicographers, while AI-generated examples and overall entries receive lower ratings. The study also explores the possibility of fine-tuning AI models based on expert feedback to enhance their performance.