Large language models represent word meanings as vectors in a multidimensional space. Building on this property, the present study offers a geometric perspective on parallelism in classical Chinese poetry, complementing traditional symbolic interpretations. In regulated verse, parallelism is a crafted pairing of two lines “side by side,” where words in the same positions mirror each other in grammar and meaning (often as matched or contrasted images). To automatically detect parallelism in poetic verse, the authors trained a BERT-based classifier on a dataset of over 140,000 regulated pentasyllabic poems (lüshi 律詩), achieving performance on par with generative models such as GPT-4.1 and DeepSeek R1. Unlike general-purpose models, the custom classifier yields unique insights into how poetic meaning is encoded geometrically: parallel lines exhibit alignment in the model’s attention patterns, with the "key" vectors of corresponding characters pointing in the same direction, while this alignment disappears in non-parallel couplets. This result extends Roman Jakobson's "poetic function” of language (equivalence projected from selection onto combination) by showing that poetic equivalence can be realized not only through sound and sequence but also through spatial relations in semantic vector space. The mechanism for this spatial equivalence can be understood through Peter Gärdenfors's theory of cognitive semantics, which posits that humans make sense of the world by organizing experience into distinct conceptual regions. From this perspective, parallelism functions as a bridging mechanism that temporarily unites such disparate regions of meaning, gesturing towards a deeper, geometric order that underlies language itself.