Customised GPTs for Lexicography

  • Gilles-Maurice de Schryver BantUGent & LT3, Ghent University, Ghent, Belgium; and Department of African Languages, University of Pretoria, Pretoria, South Africa (https://orcid.org/0000-0001-7272-9878)
DOI: https://doi.org/10.5788/35-2-2104

Abstract

About a year and a half ago, lexicographers realised that the enthusiasm they had when using ChatGPT-3.5 for English lexicography, could not be carried over to other languages, not even big ones like Portuguese. A year ago, the lexicographic community was made aware that ChatGPT-3.5 could not be used at all for small languages, and especially not for oral ones like Lusoga. In mid-2025, we notice that the models have improved drastically. Both Gemini-2.5 and especially GPT-4o perform miracles on languages like Portuguese, and for undocumented languages like Lusoga, customising GPT-4o by feeding it corpora as well as entire dictionaries is a major step forward, down to tackling neologisms. We are not yet there in full but moving forward. Keywords: generative artificial intelligence (GenAI), large language models (LLMs), generative pre-trained transformer (GPT), customised GPTs, lexicography, dictionaries, lexicographers, Portuguese, Lusoga
Published
2025-12-09
How to Cite
de Schryver, G.-M. (2025). Customised GPTs for Lexicography. Lexikos, 35(2), 397-422. https://doi.org/10.5788/35-2-2104
Issue
Vol. 35 No. 2 (2025)
Section
Artikels/Articles

Copyright of all material published in Lexikos will be vested in the Board of Directors of the Woordeboek van die Afrikaanse Taal. Authors are free, however, to use their material elsewhere provided that Lexikos (AFRILEX Series) is acknowledged as the original publication source.

Creative Commons License CC BY 4.0