Abstract
<title>Abstract</title> <p>This study explores the relationship between linguistic complexity, as measured by lexical and syntactic indices, and the quality of machine translation (MT) in philosophical texts from German to Chinese. Using BLEU (Bilingual Evaluation Understudy) scores as our quality metric, we evaluate the outputs from three popular machine translation systems: Neural Machine Translation (NMT) based on neural networks, DeepSeek which leverages Large Language Models (LLM), and DeepL driven by deep learning. This study constructs three regression models to predict translation quality based on lexical complexity (Model 1), syntactic complexity (Model 2), and a combination of both (Model 3). The results indicate that lexical richness and diversity, along with certain syntactic features, are pivotal in determining translation quality. The highest explanatory power is observed in Model 3, which integrates both lexical and syntactic complexity, underscoring the importance of a holistic approach to assessing translation quality. This research contributes to the field by providing insights into the factors that influence MT quality and by highlighting the potential of regression models in predicting translation outcomes based on linguistic complexity in low-resource domain.</p>