The Machine Translation (MT) of Proverbs in the ENG-PL Language Pair

Abstract

The translation of proverbs is an act of cross-cultural transmission that plays a vital role in the preservation and dissemination of shared knowledge within a community. While traditional dictionaries are invaluable in the translation of sayings, there is a growing trend to replace them with machine translation (MT) programmes. Although some studies have investigated the quality of machine translation of proverbs, research in this area is relatively limited. The study evaluates the quality of machine translation of 18 proverbs in Google Translate, DeepL, PONS, and the two most recent large language model tools, ChatGPT-3.5 and Google Bard. The research is conducted with two hypotheses: first, human translator supervision is essential in the case of non-literal translations such as proverbs; second, the role of the translator is changing, and in its current form, it is more often reduced to an attentive post-editor. The paper aims to achieve two objectives: first, the evaluation of the quality of machine-translated proverbs in the EN-PL pair, and second, the analysis and classification of errors occurring in machine translation of proverbs in the EN-PL pair. The research revealed that AI tools presented the highest accuracy for proverb translations, with scores of 61% (ChatGPT) and 44% (Google Bard) in Polish-to-English translation and 38% (ChatGPT) and 50% (Google Bard) in English-to-Polish translation. Furthermore, the study revealed errors typical of translation tools based on large language models such as ChatGPT and Google Bard.

Keywords

• proverbs
• translation
• machine translation
• machine translation errors
• machine translation tools
• chatgpt
• google bard