Skoči na glavni sadržaj

Izvorni znanstveni članak

Testing ChatGPT’s capabilities as an English-Croatian machine translation system in a real-world setting: eTranslation vs. ChatGPT at the European Central Bank

Krešimir Jozić ; Sveučilište u Zagrebu


Puni tekst: engleski pdf 437 Kb

str. 1-27

preuzimanja: 0

citiraj

Prilozi: H_11_Appendix_Kresimir_Jozic.pdf


Sažetak

The latest innovations in the field of natural language processing are large language models such as ChatGPT, which can perform various language tasks, including machine translation (MT). This study explores ChatGPT’s performance as an English-Croatian MT system in a realistic professional setting. An excerpt from a European Central Bank (ECB) document was translated with ChatGPT and eTranslation, the more conventional neural MT system used by ECB translators. A human evaluation, an automatic evaluation and an error analysis were performed in order to determine whether ChatGPT would be more useful than eTranslation for the ECB’s Croatian translation unit, and whether ChatGPT is better at dealing with context-related issues. The findings suggest that ChatGPT would presently not be as useful as eTranslation for English-Croatian MT at the ECB. Its performance in terms of context-related issues is still unclear. Future research could further explore this topic, as the use of ChatGPT for English-Croatian MT remains underexplored

Ključne riječi

neural machine translation, large language model, ChatGPT, eTranslation, English-Croatian translation, context-related issues

Hrčak ID:

323751

URI

https://hrcak.srce.hr/323751

Datum izdavanja:

16.12.2024.

Posjeta: 0 *