Building Publishing Content Corpora for the AI Age: Logical Premises, Current Challenges, and Institutional Pathways

Against the backdrop of a growing shortage of high-quality Chinese corpus, transforming published content into usable data assets has become critical to supporting the digital and intelligent transformation of the publishing industry, as well as the broader development of digital cultural industries. Using literature analysis, normative analysis, and case studies, this study maps current corpus development practices and diagnoses the systemic barriers impeding progress. Three primary models of corpus development have emerged in practice: independent construction, integration with large language models (LLMs), and cooperative construction. In the independent model, publishers leverage proprietary content resources to build vertical corpora. The LLM integration model focuses on connecting content with external AI capabilities, while the cooperative model involves combining editorial resources with the technical expertise of technology companies and universities. While these models reflect progress toward refined data governance, three core challenges persist: poorly defined licensing rights and value distribution, technical friction caused by fragmented formatting and annotation standards, and weak data-sharing incentives stemming from low trust and ambiguous revenue models. To address the challenges mentioned above, this paper proposes a series of integrated solutions. (1) Regarding the authorization and operation of corpus resources, the legal rights of publishing entities must be formally recognized. This involves affirming their authority to hold data resources, process and use content, and operate data products. The rights to hold and process data are grounded in the legal authorization of property rights within publishing contracts, while the right to operate and profit from data products depends on the substantive processing of these resources by the publishers. Furthermore, publishers should select operational models that align with their content advantages. Second, to resolve standard fragmentation, a collaborative alliance involving government, industry, and research institutions should be established. This body would lead the development of a standard-setting system that is guided by government leadership but driven by industry participation and multi-stakeholder coordination. Such an approach ensures that corpus standards are fundamental, practical, and capable of being widely adopted across the industry to facilitate data circulation. (3) The paper outlines three specific mechanisms to facilitate data circulation and reuse. First, establishing rules for the registration and confirmation of data asset rights. These rules would provide preliminary evidence for resolving ownership disputes and serve as essential credentials for balance-sheet recognition and market trading. Second, exploring data trust models for publishing content. This involves using informed consent and implied license rules as institutional tools for orderly sharing. Specifically, a dedicated data trust management body should be established to build "data pools", drawing on the operational experience of patent pools in the intellectual property field. Third, building a multi-dimensional incentive system. Economic incentives should follow the contribution principle and create a profit-sharing framework that covers all stakeholders in the data value chain. Technical incentives should focus on reducing participation costs and quantifying data value through innovation. Managerial incentives should include incorporating corpus construction into national financial support programs, providing research subsidies, and implementing tax preferences for participating institutions.