A Text Typological Assessment of Automatic Translation

The extensive use of artificial intelligence in translation projects has motivated this study to evaluate the quality of two popular systems, GPT Open AI and Google Translate. To assess the systems’ performance 15 sentences were chosen from a variety of text types based on Katharina Riess’s text typology informative, expressive, and vocative and were used as input data for translation into the Kurdish language. The output translations were then assessed using two evaluation metrics: BLEU and TER. The findings revealed that overall, GPT outperformed Google Translate, as it achieved a higher BLEU score reflecting better choices in equivalence and sentence structure and a lower TER score, indicating fewer necessary corrections in the translated text compared to the human (reference sentences) translation. In particular, GPT presents a better performance in the translation of expressive and vocative texts, where understanding emotions and persuasive language is more difficult.