Support Vector Machine for Error Analysis in Machine Assisted English Chinese Technical Translation: A Comparative Study with RF and BPNN

Abstract

With the rapid advancement of globalization, technical translation has become crucial for effective cross cultural communication and technology dissemination. Machine assisted translation (MAT) enhances translation efficiency and quality but often suffers from tra nslation errors that affect output accuracy. This study introduces a support vector machine (SVM) approach to systematically analyze errors in English Chinese technical translation and compares its performance with Random Forest (RF) and Back Propagatio n N eural Network (BPNN). Using 5,000 sentence pairs from domains including mechanical engineering, electronic technology, and computer science, we extract grammatical features via dependency parsing, lexical features using TF IDF, and semantic features thr oug h Word2Vec embeddings. The task is treated as a multi class classification problem, targeting lexical, grammatical, semantic, and spelling errors. Experimental results demonstrate that SVM outperforms RF and BPNN in both classification accuracy and gene ral ization ability. SVM achieves 87.6% accuracy, compared to 79.5% for BPNN and 73.2% for RF. The SVM also exhibits superior performance in 10 fold cross validation with lower mean square error (MSE) and higher R² scores. The radial basis function (RBF) ke rne l yielded optimal results among tested kernel functions. This research provides valuable insights for optimizing MAT systems and suggests that future enhancements may be achieved through deeper learning models and expanded datasets.