ResearchPod Summary
Literary historians have long debated whether regional schools of poetry existed during the Tang dynasty. This study investigates whether a poet's geographic origin—specifically their administrative circuit—leaves a quantifiable linguistic fingerprint in their work, and whether this signal persists or changes across different historical eras.
The authors constructed a corpus of 357 poets from the Complete Tang Poems, linking each author to their administrative circuit of origin via the China Biographical Database (CBDB). They framed the problem as a multi-class classification task, comparing simple character n-gram TF-IDF models against a fine-tuned classical-Chinese transformer (GuwenBERT). To ensure a fair comparison, they implemented a hierarchical frozen-encoder representation for the transformer, allowing it to process the entire corpus of a poet's work rather than just isolated fragments.
The researchers found that a poet's region (North vs. South) can be predicted with 0.69 accuracy, significantly outperforming the baseline. The regional signal is primarily driven by imagery and lexical choices rather than tonal patterns. Notably, the strength of this signal is not constant: it is weakest during the High Tang, suggesting a period of court-driven linguistic homogenization, and strongest in the Late Tang, reflecting regional divergence as central authority waned. Furthermore, the model's errors in the Early Tang—where southern poets were consistently misclassified as northern—provide quantitative evidence of the prestige and influence of the northern court idiom during that period.
This research demonstrates that interpretable machine learning can serve as a powerful hypothesis-generating tool for literary history. By identifying not just where a model succeeds, but also where it makes systematic, historically meaningful errors, the authors show that computational methods can offer new, empirical perspectives on long-standing debates regarding cultural integration and regional identity in premodern China.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.