Symmetry and Asymmetry of Translation Output Quality of MT Systems and AI Tools in Arabic-English Poetry Translation

The paper seeks to explore the symmetry and asymmetry of the translation output quality of two MT systems, namely Google Translate and DeepL and four AI tools, namely ChatGPT, Gemini, DeepSeek, and Copilot. It uses automatic evaluation, including TER, BLEU, and ChrF3 metrics along with human evaluation, including adequacy, fluency, cultural accuracy, and aesthetic quality criteria. The paper argues that there is a symmetry of the translation output quality of all the MT systems and AI tools as they all possess poor quality according to TER and BLEU metrics. On the other hand, both symmetry and asymmetry exist in the translation output quality of the aforementioned MT systems and AI tools according to both automatic evaluation ChrF3 metric and human evaluation criteria. The paper also claims that automatic evaluation metric scores are not compatible with human evaluation results with regard to the assessment of the translation output quality of MT systems and AI tools concerning Arabic-English poetry translation. This paper offers a baseline for exploring symmetry and asymmetry of the translation output quality of MT systems and AI tools within the context of Arabic-English poetry translation, which may have implications for other similar language-pair contexts.