Mapping gender, agency, and objectification in Arabic poetry: an ontology-guided computational study

Abstract Gender, agency, and objectification are important dimensions of how participants and relationships are represented in poetry, yet their systematic study in Arabic verse remains difficult because meaning is often conveyed through metaphor, ellipsis, compressed syntax, and context-dependent relations. This study investigates how gendered agency and objectification are represented across a large corpus of Arabic poetry and examines whether computational analysis can capture these literary-semantic relations while preserving textual evidence and interpretive uncertainty. A neuro-symbolic computational framework was developed to represent participants, events, agency, gender representation, and objectification as explicit, evidence-linked semantic relations. Large language model extraction was combined with formal constraint validation and expert adjudication. Evaluated against an expert-annotated sample of 4,500 verses, automated semantic extraction achieved an F1-score of 0.884 on the hold-out test set. Following evaluation and targeted adjudication within the annotated sample, the resulting constraint-guided pipeline was applied across 31,209 single-verse poems spanning twelve historical periods and eight poetic genres. Corpus-level analysis identified distinct distributions in gendered agency: male-coded participants were classified as active in 61.8\% of agency-related instances, compared with 48.2\% for female-coded participants, while the latter showed higher proportions of reactive and passive involvement. Objectification was distributed across multiple forms, with symbolic objectification emerging as the most frequent category (38.4\%), exceeding direct physical objectification (21.6\%). These results make recurrent patterns of gendered literary representation systematically inspectable at corpus scale while retaining links to textual evidence, ambiguity, and provenance. Ultimately, the study demonstrates how evidence-grounded computational methods can support large-scale investigation of literary representation while retaining interpretive boundaries.