Between Century and Poet: Graph-Based Lexical Semantic Change in Persian Poetry

A team of computational linguists and literary scholars has published a groundbreaking study on arXiv that analyzes how word meanings evolve in Persian poetry across centuries, using a novel combination of word embeddings and graph theory. The research, announced on April 26, 2024, as the preprint arXiv:2604.06674v1, was conducted to move beyond traditional methods of measuring semantic change by examining the relational and contextual networks of words within the poetic tradition. The study's core innovation lies in treating the lexicon not as a set of isolated vectors but as a dynamic, interconnected system shaped by literary history. The research challenges the conventional approach of modeling semantic change purely as the displacement of a word's vector in an embedding space over time. Instead, the authors propose a graph-based framework that analyzes a word's 'neighborhood'—the constellation of other words it commonly appears near, the rhetorical frames it inhabits, and the specific poetic voices that employ it. By applying this method to a corpus spanning multiple centuries and major Persian poets, the study captures how a word's semantic force shifts not just in definition but in its entire network of associations and connotations within the literary ecosystem. This methodology allows for a more nuanced understanding of lexical historicity, revealing that words in Persian poetry maintain a persistent identity through tradition while simultaneously adapting their expressive power. The findings suggest that meaning is co-constructed by historical continuity and relational change, offering a powerful new tool for digital humanities. The study bridges computational analysis with deep literary inquiry, providing a model that could be applied to other poetic and linguistic traditions to trace the subtle, network-driven evolution of language in art.