[Purpose/Significance] For the study of highly intertextual pre-Qin classics, an unsupervised intertextual automatic discovery process is established to better carry out content analysis and textual research of ancient books based on large language models, which improves work efficiency.[Method/Process] It comparatively analyzed different technical routes of existing language models, and introduced a contrastive learning framework to train intertextual automatic discovery models in an unsupervised manner.By constructing an idiom origin tracing tasks to evaluate a series of models, it selected the optimal model.[Result/Conclusion] The results of idiom origin tracing show that there are still a large number of factual errors in the current Chat LLMs.The ESimCSE-GujiRoBERTa model has achieved the best results in the idiom origin tracing task.This model shows excellent semantic discrimination ability in the intertextual recognition of citations in the classics of pre-Qin scholars.At the same time, judging from the results of the intertextual identification of the “Chun Qiu San Zhuan”, the automatic discovery of intertextuality can provide a useful perspective for the textual research and collation of ancient books.
[1] 程锡麟. 互文性理论概述[J]. 外国文学, 1996(1): 72-78. (CHENG X. Overview of intertextuality theory[J]. Foreign literature, 1996(1): 72-78.)
[2] KRISTEVA J. The Kristeva reader[M]. New York: Columbia University Press, 1986.
[3] 陈永国. 互文性[J]. 外国文学, 2003(1): 75-81. (CHEN Y. Intertextuality[J]. Foreign literature, 2003(1): 75-81.)
[4] 黄念然. 当代西方文论中的互文性理论[J]. 外国文学研究, 1999(1): 15-21. (HUANG N R. Intertextuality theory in contemporary western literary theory[J]. Foreign literature studies, 1999(1): 15-21.)
[5] DUAN S, WANG J, YANG H, et al. Disentangling the cultural evolution of ancient China: a digital humanities perspective[J]. Humanities and social sciences communications, 2023, 10(1): 310.
[6] 黄一農. 從e考據看避諱學的新機遇: 以己卯本《石頭記》爲例[J]. 文史, 2019(2): 205-222. (HUANG Y N. New opportunities for taboo studies from the perspective of e-textual criticism: taking Ji Mao’s “The Story of the Stone” as an example [J]. Literature and history, 2019(2): 205-222.)
[7] 黄水清, 周好, 彭秋茹, 等. 引书的自动识别及文献计量学分析[J]. 情报学报, 2021, 40(12): 1325-1337. (HUANG S, ZHOU H, PENG Q, et al. Automatic recognition and bibliometric analysis of cited books[J]. Journal of the China Society for Scientific and Technical Information, 2021, 40(12): 1325-1337.)
[8] 柳士镇. 《世说新语》《晋书》异文语言比较研究[J]. 中州学刊, 1988(6): 99-102, 49. (LIU S Z. Comparative study on the different languages of “Shishuoxinyu” and “Jinshu” [J]. Zhongzhou academic journal, 1988(6): 99-102, 49.)
[9] 秦淑华. 《史记》与《战国策》的异文研究[J]. 汉字文化, 2002(4): 42-43. (QIN S H. Research on variant texts of “Shi Ji” and “Zhan Guo Ce” [J]. Sinogram culture, 2002(4): 42-43.)
[10] 王化平. 由《孔子家语》与《礼记》、《说苑》诸书的关系看其价值[J]. 古籍整理研究学刊, 2011(1): 1-7. (WANG H P. Judging its value from the relationship between “Confucius’ Family Language” and “Book of Rites” and “Shuo Yuan” [J]. Journal of ancient books collation and studies, 2011(1): 1-7.)
[11] 贾骄阳. 《世说新语》与《论语》的互文性关系探究[J]. 铜仁学院学报, 2017, 19(10): 19-204. (JIA J Y. On the intertextual relationship between a new account of tales of the world and the analects of Confucius[J]. Journal of Tongren University, 2017, 19(10): 19-24.)
[12] FORSTALL C, COFFEE N, BUCK T, et al. Modeling the scholars: detecting intertextuality through enhanced word-level n-gram matching[J]. Digital scholarship in the humanities, 2015, 30(4): 503-515.
[13] GANASCIA J G, GLAUDES P, DEL LUNGO A. Automatic detection of reuses and citations in literary texts[J]. Literary and linguistic computing, 2014, 29(3): 412-421.
[14] 李越. 《左传》《史记》同事异文自动发现及分析[D]. 南京: 南京师范大学, 2014. (LI Y. Automatic discovery and analysis of variant texts in “Zuo Zhuan” and “ Shi Ji”[D]. Nanjing: Nanjing Normal University, 2014.)
[15] DENG Z, YANG H, WANG J. A comparative study of “Shiji” and “Hanshu” from the perspective of digital humanities[C]// Proceedings of the 21st Chinese national conference on computational linguistics. Nanchang: Chinese Information Processing Society of China, 2022: 656-670.
[16] STURGEON D. Unsupervised identification of text reuse in early Chinese literature[J]. Digital scholarship in the humanities, 2018, 33(3): 670-684.
[17] 钮亮, 项玮. 基于BLAST文本重用的先秦诸子关系考察[J]. 文献与数据学报, 2023, 5(4): 55-64. (NIU L, XIANG W. An examination of the relationship between pre-Qin thinkers based on BLAST text reuse[J]. Journal of library and data, 2023, 5(4): 55-64.)
[18] 刘根辉, 李志芳. 《册府元龟》引《国语》考述——《册府元龟》引书系列研究之三[J]. 吉林师范大学学报(人文社会科学版), 2024, 52(2): 68-75. (LIU G H，LI Z F. The textual research on “Cefu Yuangui” quoting from Guoyu: the third study of the citation series in “Cefu Yuangui”[J]. Journal of Jilin Normal University(humanities & social science edition), 2024, 52(2): 68-75.)
[19] 梁媛, 王东波, 黄水清. 古籍同事异文的自动发掘研究[J]. 图书情报工作, 2021, 65(9): 97-104. (LIANG Y, WANG D B, HUANG S Q. Research on automatic mining of variants expressing the same event in the ancient books[J]. Library and information service, 2021, 65(9): 97-104.)
[20] 程宁. 数字人文视域下《文选》与唐诗的互文空间[J]. 清华大学学报(哲学社会科学版), 2023, 38(6): 75-88, 222. (CHENG N. The intertextual space between “Wenxuan” and the Tang poetry from the perspective of digital humanities[J]. Journal of Tsinghua University(philosophy and social sciences), 2023, 38(6): 75-88, 222.)
[21] LI W, SHAO Y, BI M. Data construction and matching method for the task of ancient classics reference detection[C]//Proceedings of the 21st Chinese national conference on computational linguistics. Nanchang: Chinese Information Processing Society of China, 2022: 600-610.
[22] REIMERS N, GUREVYCH I. Sentence-Bert: sentence embeddings using Siamese Bert-networks[J]. arXiv preprint, 2019, arXiv:1908.10084.
[23] GAO T, YAO X, CHEN D. SimCSE: simple contrastive learning of sentence embeddings[J]. arXiv preprint, 2021, arXiv:2104.08821.
[24] WU X, GAO C, ZANG L, et al. EsimCSE: enhanced sample building method for contrastive learning of unsupervised sentence embedding[J]. arXiv preprint, 2021, arXiv:2109.04380.
[25] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// Proceedings of the 31st international conference on neural information processing systems. California: Curran Associates, 2017: 6000-6010.
[26] DEVLIN J, CHANG M W, LEE K, et al. Bert: pre-training of deep bidirectional transformers for language understanding[J]. arXiv preprint, 2018, arXiv:1810.04805.
[27] RADFORD A, NARASIMHAN K, SALIMANS T, et al. Improving language understanding by generative pre-training[J]. arXiv Preprint. 2018. 1-12.
[28] RADFORD A, WU J, CHILD R, et al. Language models are unsupervised multitask learners. [J]. OpenAI blog, 2019, 1(8):9.
[29] BROWN T, MANN B, RYDER N, et al. Language models are few-shot learners [J]. arXiv preprint, 2020, arXiv:2005.14165.
[30] WANG D, LIU C, ZHAO Z, et al. GujiBERT and GujiGPT: construction of intelligent information processing foundation language models for ancient texts[J]. arXiv preprint, 2023, arXiv:2307.05354.
[31] BAI J, BAI S, CHU Y, et al. Qwen technical report[J]. arXiv preprint, 2023, arXiv:2309.16609.
