Testing ChatGPT on Terminology Generation, Definitions: Translation, and Ontology Creation in German, English and Polish

Abstract

The paper presents the generation process and analysis of Chat GPT-produced terminology, definitions, translation and ontology creation of content derived from a restricted domain of electrotechnology on the basis of German manufacturing instruction (Fertigungsvorschrift) for the assembly and functioning of thermal switches. The first part of the study presents a brief historical sketch of corpus linguistics towards the development of LLMs and their applications. Research steps include the generation of terminology and a basic restricted domain ontology in German and their equivalents in English and Polish as identified in general web resources, followed by the same tasks based on the analysed tool instructions. The tests were first performed on earlier ChatGPTPro, followed by its recent version ChatGPT4, also with regard to their English and Polish equivalents, for different tasks, i.a., relevant thesauri building. An Assistant called SLA (Special Language Assistant) was created for the purpose of analysing prompts, recognising context and intent, and processing data using language models.The tests have been carried out in terms of 4 prompts. First, specialist terms typical of the product and its parts were identified in German and the translation of these terms and of other relevant specialist phrases into English and Polish was performed. Finally, the ontological categorisation and its visualisation were generated. Results indicate areas of fair correspondences with manual intervention needed for the term definition refinement and ontology specification. The presentation emphasizes in the conclusions the effects of AI tasks as a contribution to lexicography, translation and foreign language education. The study can also serve as a reference for NLP researchers to improve the functioning of LLM tools.