A local sequence alignment approach to recognizing fixed poetic forms across languages.

Area of Science:

• Computational Linguistics
• Literary Studies
• Digital Humanities

Background:

• Fixed poetic forms are crucial in European literary traditions.
• Large-scale empirical research on cross-lingual distribution and evolution of these forms is limited.

Purpose of the Study:

• To develop a language-independent, unsupervised method for identifying recurrent rhyme-based poetic forms.
• To enable large-scale empirical analysis of poetic form distribution and evolution across European literary traditions.

Main Methods:

• Utilized local sequence alignment (Smith-Waterman algorithm) and a compact eight-symbol alphabet to encode rhyme schemes.
• Applied dimensionality reduction (UMAP) and density-based clustering (HDBSCAN) to a dataset of 187,719 poems from six European languages (PoeTree collection).
• Evaluated method recall against existing annotations and tested robustness through supervised classification experiments.

Main Results:

• Identified 61 distinct clusters of rhyme-based forms, many corresponding to known fixed poetic forms.
• Demonstrated strong recall in evaluations against existing annotations.
• Showcased robust learnability of form categories in the induced vector space for cross-lingual classification.

Conclusions:

• The developed method provides a scalable and language-independent approach to identifying and analyzing poetic forms.
• The findings facilitate new avenues for literary research, including cross-tradition influence, topical affinities, and geographic associations of poetic forms.