A lattice-based model identifies meters in Ottoman poetry

A new model using a hidden Markov model over a lattice of candidate paths identifies the meters and scansion of Ottoman Divan poetry.

The system analyzes poems written in aruz, a quantitative meter where syllable weights determine the rhythm. This is particularly useful for modern Turkish spellings that lack the specific diacritics found in scholarly transliterations of the original Arabic script.

The model performs two primary tasks by identifying the specific meter used in a poem and aligning those weights with the poem's syllabification. This process reveals both the technical flaws and the creative freedoms a poet took during the composition. The researchers trained the system using the printed meter of a corpus of 52,000 lines divided by poet, rather than relying on individual scansion labels.

The tool outperformed both a rule-based scanner and a neural tagger on a held-out test set. Researchers released the code as open-source and made the tool available on the web. Future applications involve refining the accuracy of automated analysis for historical literary texts.