Abstract
Abstract As digital technology innovation emerges as a critical driver of green economic transformation, understanding its environmental implications becomes paramount. Leveraging matched data from the China Industrial Enterprise Database, the Corporate Pollution Emission Database, and the China Patent Database spanning 1998 to 2014, this study innovatively employs the Sentence-BERT (SBERT) language model to conduct semantic analysis of patent texts, achieving precise identification of digital technology patents. Empirical results indicate that digital technology innovation significantly reduces corporate pollution emissions. Specifically, a 10% increase in digital patent applications and grants is associated with a 1.01% and 1.32% reduction in pollution emissions, respectively. These findings remain robust after a series of tests, including alternative variable specifications, adjusted clustered standard errors, Propensity Score Matching combined with Difference-in-Differences (PSM + DID), and endogeneity checks. Mechanism analysis reveals that digital technology innovation mitigates emissions through three channels: optimizing energy consumption structures, upgrading pollution control technologies, and improving production efficiency. Heterogeneity analysis demonstrates that the emission-reducing effects are more pronounced in large-scale enterprises, state-owned enterprises, technology-intensive industries, and regions with weaker environmental regulation stringency. This study not only advances methodological precision by employing the SBERT language model to identify digital technology patents, overcoming the limitations of traditional IPC-based approaches, but also elucidates the micro-mechanisms through which digital technology innovation resolves the “Solow Paradox” to achieve a win-win outcome for economic and environmental performance.