A metrical scansion system for fixed-metre Spanish poetry

In this article an automatic scansion model for fixed-metre Spanish poetry is presented. It is a hybrid model that combines hand-made rules with probabilistic information. Through the set of rules, the model is able to extract the syllabic structure of each word, to classify them as stressed or unstressed and to resolve metrical phenomena such as synaloephas or diaereses. The article is mainly focused on the metrical ambiguities produced by synaloephas: verse lines from which it is possible to derive two or more metrical patterns. This metrical ambiguity is resolved through probabilities, assuming a relation between high probabilities and metricality. The system has been evaluated through more than 1,000 lines extracted from a corpus of Golden-Age Spanish sonnets. An accuracy of 95% has been achieved, resulting in not only considerable progress if we compare it to previous proposals, but also in an adequate way of performing the task when compared to human performance.