1　Introduction

This section introduces the application background of large language models in linguistic and applied linguistic research, noting that corpus studies have entered the "intelligent post-editing" era with significantly lowered technical barriers; analyzes the multiple challenges facing parallel corpus construction and the deficiencies in existing research, particularly the scarcity of cases where AI empowers the entire workflow for small-to-medium scale multilingual parallel corpora of classical texts; and finally clarifies that this paper takes the Chinese-Russian parallel corpus of *The Analects* as a case study to systematically demonstrate the specific applications of AI technology throughout the entire workflow, including OCR recognition, cross-lingual automatic alignment, and retrieval system development, providing corpus researchers with a practical reference for "AI thinking."

2　Literature Review

This section reviews the traditional paradigms of parallel corpus construction and their challenges, including limited OCR recognition accuracy, poor performance of alignment tools in handling complex translation relationships, and high technical barriers for retrieval systems, and points out the special difficulties faced by multilingual corpus construction for classical texts, such as low-resource language processing. It then reviews the current state of large language model applications in corpus research, covering four main directions: corpus generation and data augmentation, automatic annotation and information extraction, domain-specific large model construction, and classical text processing. Finally, it identifies gaps in existing research, including the absence of full-process case studies, lack of empirical cross-lingual research on classical texts, insufficient "AI thinking" among researchers, and gaps in AI-assisted retrieval system development, laying the foundation for this paper's demonstration of full-process AI applications using the Chinese-Russian parallel corpus of *The Analects* as an example.

3　Corpus Construction Practice

This section systematically elaborates on the construction practice of the Chinese-Russian parallel corpus of *The Analects*, covering four core stages: corpus sources, text digitization, cross-lingual alignment, and retrieval system development. The corpus source is the 2001 Russian-published collection of *The Analects* translations, which includes translations by five major translators including Vasiliev and Popov, supplemented with translations by other translators such as Alexeev for reference. The construction process is divided into three stages: corpus digitization and correction, cross-lingual multi-translation alignment, and retrieval system development and deployment.

In the text digitization stage, Baidu's open-source PaddleOCR tool was used to replace traditional commercial software, leveraging its deep learning architecture, multilingual support, and free open-source advantages to complete the recognition of Russian scanned documents. To address typical post-OCR issues such as word adhesion, case errors, similar letter confusion, line break fragmentation, and special character misrecognition, the study designed specialized prompts to call the DeepSeek R1 large model for batch correction, followed by manual review by Russian language professionals, forming an "AI processing + human verification" model that reduced manual proofreading time by 70%–80%.

The cross-lingual alignment stage faced challenges including "one-to-five" multi-translation structures, significant differences in translation strategies, and omission phenomena. The study called DeepSeek R1 through Python scripts to output alignment results in JSON format, with prompt design specifying translator names, anticipating quantity inconsistencies, strictly prohibiting text modification, and standardizing output structure. After manual verification, all alignment results were found to be error-free, verifying the feasibility for non-Russian researchers to complete cross-lingual alignment with the assistance of large models.

Retrieval system development broke through the technical barriers and functional limitations of traditional platforms. The MiniMax large model Agent mode was adopted to directly generate Web system code through natural language description, followed by fine-tuning with Claude 4.5. The final system implemented six functions: Chinese-Russian dual-condition retrieval, part-of-speech filtering, Lemma retrieval, chapter specification, translation selection, and per-translation frequency statistics, with the entire development cycle not exceeding two working days, demonstrating the empowering value of AI code generation technology for customized corpus tool development.

4　Discussion

This section analyzes the empowering effectiveness of AI technology throughout the full workflow of Chinese-Russian parallel corpus construction for *The Analects*, noting that AI significantly improved efficiency in text digitization and correction, cross-lingual alignment, and retrieval system development, lowered technical barriers, and shifted researchers' core competitiveness from "technical execution ability" to "research design and academic judgment." The section summarizes three practical experiences: precise demand description requires mastering industry terminology to improve prompt quality; cultivating "AI thinking" to actively consider task optimization paths; and adhering to the "AI processing + human review" collaborative paradigm to ensure quality. Finally, it acknowledges limitations including regular text structure, lack of control experiments, and output randomness, and prospects the development of multimodal models and stronger multilingual models.

5　Conclusion

Taking the construction of the Chinese-Russian parallel corpus of *The Analects* as a case study, this paper systematically demonstrates the innovative applications of AI technology throughout the full workflow of corpus construction, including PaddleOCR text recognition, DeepSeek-assisted correction, large model-driven cross-lingual automatic alignment, and customized retrieval system development. The core argument emphasizes that corpus researchers should establish "AI thinking," integrating AI tools into all stages of research to expand research boundaries, while noting that academic literacy must serve as the prerequisite and human review as the safeguard, achieving an organic combination of AI efficiency and human judgment.