Scholarlink

Contemporary Arabic narrative generation systems often fail to capture nuanced cultural authenticity, primarily addressing dialectal variation as a lexical challenge rather than encoding deeper cultural schemas. The distinction between (Egyptian) and (Levantine) for “The story began in Cairo” exemplifies cultural resonance patterns that standardized approaches overlook. We present CULTURA, a neural-symbolic framework orchestrating three specialized agents: (1) CNN-based dialect classification (94.3% accuracy across six Arabic varieties); (2) OWL 2 DL ontology encoding cultural narrative schemas; (3) PPLM-enhanced generation with schema-guided cultural steering. Evaluation on 2,847 prompts across six dialects shows improvements over strong baselines: Dialect F1 = 84.2% vs. 68.9% (AraT5), CMPI = 0.82 vs. 0.49, and human cultural resonance 6.1 ± 0.3 vs. 4.3 ± 0.4 (7-point Likert), all by paired bootstrap, \(p<0.001\) .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CULTURA: A Multi-agent Neural–Symbolic System for Culturally-Aware Arabic Story Generation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mossab Ibrahim,
  • Pablo Gervás,
  • Gonzalo Méndez

摘要

Contemporary Arabic narrative generation systems often fail to capture nuanced cultural authenticity, primarily addressing dialectal variation as a lexical challenge rather than encoding deeper cultural schemas. The distinction between (Egyptian) and (Levantine) for “The story began in Cairo” exemplifies cultural resonance patterns that standardized approaches overlook. We present CULTURA, a neural-symbolic framework orchestrating three specialized agents: (1) CNN-based dialect classification (94.3% accuracy across six Arabic varieties); (2) OWL 2 DL ontology encoding cultural narrative schemas; (3) PPLM-enhanced generation with schema-guided cultural steering. Evaluation on 2,847 prompts across six dialects shows improvements over strong baselines: Dialect F1 = 84.2% vs. 68.9% (AraT5), CMPI = 0.82 vs. 0.49, and human cultural resonance 6.1 ± 0.3 vs. 4.3 ± 0.4 (7-point Likert), all by paired bootstrap, \(p<0.001\) .