A novel, rule-based, automatic framework for identifying rhetorical elements in classical Arabic poetry is described. Since rule-based approaches have well-known limitations, it is proposed as an interim solution until a sufficient quantity of annotated text has been amassed with which to train a machine-learning algorithm. The manual process of identifying rhetorical features in classical Arabic poetry is both time-consuming and requires high-level expertise in Arabic literature. Hence, an automatic recognition system will solve this challenge. Automatic identification is, however, challenging, mainly because there is no existing annotated corpus with which to train a machine-learning-based classifier. The framework proposed here combines natural language-processing techniques with a rule-based reasoning approach, and will continually improve as more examples become available. It is intended as an initial step toward building the essential annotated corpus. Its focus is 20 rhetorical elements, all of importance according to classical Arabic rhetoricians, and it achieves the extremely encouraging result of an overall F-measure of 0.902.
Alhadi et al. (Fri,) studied this question.