在英语学习中，很多中国学生习惯使用单一的表达方法和词汇。针对这类问题，paraphrase（句子改写）练习可以训练学生通过使用不同的句式、替换词汇、表达方式等来拓展表达方法，同时有意识的积累语言素材，提高英语表达能力。在线下教学中，教师可以准确评估学生的句子改写水平并给予有针对性的反馈和建议。在线上教学中，如果AI拥有了类似的能力，可以大幅提高批改效率，减轻教师负担，同时为学生提供更多的练习机会。然而，由于语言本身的复杂性，如何让AI在评估两个句子的语义是否相近的同时，又能评估这两个句子在句子结构、表达方式和词汇选择等方面的差异，是目前面临的一大挑战。在本课题中，我们希望聚焦在英语句子改写场景中，综合使用神经网络和其他深度学习方法从语义、语法、词汇、句型等不同维度来分析学生的改写练习，判断其和原句在语义和表达等方面的相似度，给出接近真人老师的评分和反馈。
In English learning, many Chinese students are used to using a single expression and vocabulary. To address this problem, paraphrase exercises can train students to expand their methods of expression through the use of different sentence patterns, substitution of vocabulary and expressions, and at the same time consciously accumulate language materials to improve their English expression skills. In offline teaching, teachers can accurately assess students' sentence rewriting level and give targeted feedback and suggestions. In online teaching, if the AI has a similar ability, it can significantly improve the efficiency of correction and reduce the burden of teachers, while providing more opportunities for students to practice. However, due to the complexity of language itself, how to allow AI to assess whether two sentences are semantically similar while at the same time evaluating the differences between these two sentences in terms of sentence structure, expression, and vocabulary choice is currently a major challenge. In this project, we hope to focus on the English sentence rewriting scenario, and comprehensively use neural networks and other deep learning methods to analyze students' rewriting exercises from different dimensions, such as semantics, grammar, vocabulary, and sentence patterns, to judge their similarity to the original sentence in terms of semantics and expression, and to give scores and feedbacks close to those of real teachers.
