| 208 | 0 | 98 |
| 下载次数 | 被引频次 | 阅读次数 |
本研究运用DistilBERT深度神经网络模型，结合传统自然语言处理的N元匹配、潜在语义分析和GloVe模型，将4类(13个)语义变量纳入记叙文、说明文、议论文3种文体，构建传统量化与深度学习相结合的语义质量评价模型。结果表明，语义变量与译文质量之间均存在显著相关性，且贡献程度各异。综合所有变量的回归模型拟合度最高，相关系数超过0.80，且模型稳定性良好。研究发现，传统量化方法的预测效果较好，DistilBERT模型展现出独特优势，能够结合传统量化方法，进行多维语义评价。未来研究可以探索传统量化方法与深度学习技术的融合路径，实现更高效的翻译语义自动评价。
Abstract：This study intends to construct a semantic quality evaluation model by employing the DistilBERT deep neural network in combination with traditional natural language processing techniques,including N-gram matching,Latent Semantic Analysis (LSA),and the GloVe model.The evaluation model incorporates four categories (13 semantic variables) across three genres (narrative,expository,and argumentative).The results reveal significant correlations between the semantic variables and translation quality,with varying degrees of contribution texts.The comprehensive regression model incorporating all variables demonstrated the highest goodness-of-fit,achieving a correlation coefficient exceeding0.80 along with good model stability.The findings reveal that traditional methods yield better predictive results,while the DistilBERT model demonstrates unique advantages in complementing traditional measurement approaches to enable multidimensional semantic evaluation.Future studies could explore integration pathways between conventional methods and deep learning to realize more efficient automated evaluation of translation semantics.
[1]Areshey,A.&H.Mathkour.Exploring transformer models for sentiment classification:A comparison of BERT,Ro BERTa,ALBERT,Distil BERT,and XLNet[J].Expert Systems,2024(11):e13701.
[2]Deerwester,S.,S.T.Dumais,G.W.Furnas,et al.Indexing by latent semantic analysis[J].Journal of the American Society for Information Science,1990(6):391-407.
[3]Devlin J.,M.W.Chang,K.Lee,et al.Bert:Pre-training of deep bidirectional transformers for language understanding[C].Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies,2019:4171-4186.
[4]Newmark,P.A Textbook of Translation[M].New York:Prentice Hall,1988.
[5]Nida,E.A.Translating Meaning[M].California:English Language Institute,1982.
[6]Papineni,K.,S.Roukos,T.Ward,et al.Bleu:A method for automatic evaluation of machine translation[C].Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics,2022:311-318.
[7]P e nnington,J.,R.Socher&C.D.Manning.Glove:Global vectors for word representation[C].Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing,2014:1532-1543.
[8]Sanh,V.,L.Debut,J.Chaumond,etal.Distil BERT,a distilled version of BERT:Smaller,faster,cheaper and lighter[J].ar Xiv,2019.DOI:10.48550/ar Xiv.1910.01108.
[9]Sanh,V.,L.Debut,J.Chaumond,et al.The Distil BERT model:A promising approach to improve machine reading comprehension models[J].International Journal on Recent and Innovation Trends in Computing and Communication,2023(11):293-309.
[10]Strubell,E.,A.Ganesh&A.Ma Callum.Energy and Policy Considerations for Modern Deep Learning Research[C].Proceedings of the 2020AAAI Conference on Artificial Intelligence,2020:13693-13696.
[11]Wei,S.,D.Yu&C.Lv.A distilled BERT with hidden state and soft label learning for sentiment classification[C].Proceedings of CISAI 2020,2020:012076.
[12]车万翔,窦志成,冯岩松等.大模型时代的自然语言处理：挑战、机遇与发展[J].中国科学：信息科学,2023(9):1645-1687.
[13]陈德光,马金林,马自萍等.自然语言处理预训练技术综述[J].计算机科学与探索,2021(8):1359-1389.
[14]冯志伟,李颖.自然语言处理中的预训练范式[J].外语研究,2021(1):1-14;112.
[15]冯志伟,张灯柯.G P T与语言研究[J].外语电化教学,2023(2):3-11;105.
[16]冯志伟,张灯柯.计算语言学中语言知识生产范式的变迁[J].当代修辞学,2024a(2):23-44.
[17]冯志伟,张灯柯.人工智能中的大语言模型[J].外国语文,2024b(3):1-29.
[18]韩雨潇.生成式人工智能作品版权归属问题研究[J].扬州大学学报（人文社会科学版）,2023(4):52-64.
[19]何武.从英语小说阅读到英语写作--一项潜伏语义分析理论驱动的英语写作教学实验报告[J].外语学刊,2013(5):124-129.
[20]黄施洋,奚雪峰,崔志明.大模型时代下的汉语自然语言处理研究与探索[J].计算机工程与应用,2025(1):80-97.
[21]梁茂成.学习者书面语语篇连贯性的研究[J].现代外语,2006(3):284-292;330.
[22]马畅,田永红,郑晓莉等.基于知识蒸馏的神经机器翻译综述[J].计算机科学与探索,2024(7):1725-1747.
[23]苏祺,杨佳野.语言智能的演进及其在新文科中的应用探析[J].中国外语,2023(3):4-11.
[24]王金铨.汉译英自动评分系统中的语义相似度比较研究[J].外语测试与教学,2011(4):8-17.
[25]王金铨,何泊稼.基于神经网络模型的翻译语义质量量化评价[J].中国外语,2024(1):92-101.
[26]王金铨,梁茂成,俞洪亮.基于N-gram和向量空间模型的语句相似度研究[J].现代外语,2007(4):405-413;438.
[27]王金铨,牛永一.计算机辅助翻译评价系统中的翻译质量评估[J].上海翻译,2023(6):52-57.
[28]王金铨,文秋芳.中国学生大规模汉译英测试机助评分模型的研究与构建[J].现代外语,2009(4):415-420;438.
[29]王旭阳,耿留青,张鑫.结合Distil BERT与标签关联性的多标签文本分类[J].计算机工程与应用,2024(23):168-175.
[30]许钧,穆雷.翻译学概论[M].南京：译林出版社,2009.
[31]张乐,冷基栋,吕学强等.MWEC:一种基于多语义词向量的中文新词发现方法[J].数据分析与知识发现,2022(1):113-121.
[32]张莹,艾则孜,吴顺祥.利用N-gram和语义分析的维吾尔语文本相似性检测方法[J].计算机应用研究,2019(9):2722-2725;2729.
基本信息:
DOI：10.13564/j.cnki.issn.1672-9382.2025.05.007
中图分类号:H315.9
引用信息:
[1]王金铨,吕文静.翻译语义自动评价——基于传统量化与深度学习的模型构建[J].中国外语,2025,22(05):74-82.DOI:10.13564/j.cnki.issn.1672-9382.2025.05.007.
基金信息:
国家社科基金重点项目“基于大语言模型的中国英语学习者翻译能力发展及量化评价研究”（编号：24AYY019）的阶段性成果