DeepL et Google Translate face à l'ambiguïté phraséologique

Bacquelaine, Françoise

doi:10.46298/jdmdh.9118

Cited by 1 publication

(1 citation statement)

References 1 publication

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Comparing two free open-source NMT systems, Google Translate and DeepL, when translating Spanish phraseological units both show a similar performance which is weakened when encountering low-frequency expressions [28]. Other studies confirm such results in Portuguese-French, where phraseology, calque and nonsense were the most frequent errors [29].…”

Section: Relevant Workmentioning

confidence: 65%

Google Translate vs. ChatGPT: Can non-language professionals trust them for specialized translation?

Sanz-Valdivieso,

López-Arroyo

2023

Proceedings of the International Conference on Human-Informed Translation and Interpreting Technology 2023

View full text Add to dashboard Cite

Experts and professionals in specialized fields often need writing tools to communicate in English as a means to disseminate their knowledge or enter the international market. There are different tools to accomplish this and most of them are, lately, Machine Translation systems (MT) based on Neural Machine Translation (NMT), an approach using artificial neural networks to translate with outstanding fluency. Free and open systems such as Google Translate or, more recently, ChatGPT used as a translator, have popularized NMT to a multitude of users. However, there are experts and professionals who, due to their lack of command of English, often fail in their communication tasks by accepting NMT system's output as correct. This paper examines these systems' performance when translating terminology of the discourse in wine and olive oil tasting notes, specifically from Spanish into English. This domain may serve to represent lessstudied specialized languages where general language words and terms become closely intertwined. The aim is to determine whether these systems can translate terminology accurately within the domain, and, if so, whether the GPT-3.5 model outperforms Google Translate. Results will help identify or discard possible language solutions for users who need to obtain texts in specialized English with professional and internationalization purposes, but who do not have the linguistic or economic resources to ensure the quality of the English text. Results show that, although ChatGPT yields fewer terminological errors than Google Translate in terms of error severity and number of samples affected, professionals cannot rely solely on these tools just yet.

show abstract

Section: Relevant Workmentioning

confidence: 65%