The end of lexicography? Can ChatGPT outperform current tools for post-editing lexicography?

Abstract

In this paper we present a small English dictionary consisting of 99 sample entries generated fully automatically using the ChatGPT engine. We briefly introduce ChatGPT and the underlying machinery (an autoregressive transformer-based neural network) but primarily focus on discussing the performance of the system, factors that influence the quality of the output and limitations that we have established. We show that while the system clearly represents part of the state-of-the-art of automatic generation for some entry components, it also has significant limitations which the lexicographic community should be aware of.