基于机器学习的汉译英计算机辅助评估系统构建-王小曼/朱玉彬

人工智能时代口译技术应用研究
王华树 | 国内首部聚焦口译技术应用和教学的著作
新书推荐


口笔译教育与评价国际论坛 二号公告
在厦门大学百年校庆之际，邀您齐聚厦门、共襄盛举
论坛推荐

基于机器学习的汉译英计算机辅助评估系统构建-王小曼/朱玉彬

2019年第二届翻译教育国际研讨会
机助译员训练 (CATT): 人机之间
观看大会全程直播盛况

The Building of a Computer-assisted Assessing System for Chinese-to-English Translation on the Basis of Machine Learning

王小曼(Xiaoman WANG)
安徽三联学院(Anhui Sanlian University)
朱玉彬(Yubin ZHU)
安徽大学(Anhui University)

本文以机器翻译质量评估工具BLEU的评测结果为特征因子，结合监督式机器学习模型，探讨机器学习是否可以用于评估学生汉译英的译文质量。我们在采用预测分级的方式建立机器学习模型的前提下，利用Python 内置语言处理包NLTK中的N元组BLEU评测方法以及平滑优化后的BLEU评测方法比照学生译文和参考译文，得出评测结果并将其作为监督式机器学习模型的特征因子，再利用R语言中的Caret套件中的K-近邻算法和E1071套件中的支持向量机建立机器学习模型。随后结合混淆矩阵的精确度，95%置信区间, P值检定和各评价区间的预测精确度、敏感性和特异性评定预测结果的信度和效度，对评估结果进行评分佐证。结果显示，借助监督式机器学习模型进行评估预测的测试数据利用最佳模型可达87.18% 的预测精准度, Kappa系数达0.8084。由此可见，BLEU和监督式机器学习模型相结合的方式，能够为教师评估学生译文质量提供更为有效的帮助，从而提高翻译评估的效率。

This study uses the BLEU assessment results as the factors and the supervised machine learning model to explore whether machine learning can be applied to the evaluation of students’ Chinese-to-English translation. On the basis of a machine learning model by a predictive grading, we evaluated student translations and reference translation by means of Python NLTK package and BLEU optimized by the smoothing method, and used the evaluated results as the factors of the supervised machine learning model. Then support vector machine (SVM) and K-nearest neighbor algorithm (KNN) in this machine learning model are built by means of Caret and E1071 suite in R language. And the reliability and validity of the predictions are measured by the accuracy of Confusion Matrix, 95% confidence interval, P-value and accuracy, sensitivity and specificity of each evaluation interval. The results show that the predictive test data can be predicted by the supervised machine learning model in the best model with a prediction accuracy of 87.18%, and its Kappa coefficient is 0.8084. This demonstrates that the combination of BLEU and supervised machine learning model can effectively assist teachers in assessing students’ translation, thus improving the efficiency of translation assessment.

《2019翻译教育国际论坛》专题栏目

相关推荐
★★★★★ 5/5