Towards Arabic Quantum Natural Language Processing: A Categorical Framework for Compositional Quantum Language Models

Quantum Natural Language Processing (QNLP) provides a promising intersection between quantum computing and compositional AI, exploring how the grammatical structure of language can be represented through tensor networks and quantum circuits. However, existing QNLP approaches have predominantly focused on English and word-level representations, leaving open the challenge of modeling morphologically rich languages where meaning is distributed across complex word forms and clitic structures. We introduce QArabicCat, a morphology-aware framework that extends categorical quantum language models to Arabic. By integrating Arabic morphological analysis with Combinatory Categorial Grammar (CCG) and DisCoCat, our approach incorporates morphological information into compositional quantum representations. Instead of representing Arabic clitics and inflectional markers as independent tokens, we encode them as grammatical features through clitic saturation and shared feature gates, preserving linguistic structure while reducing unnecessary circuit complexity. These representations are transformed into compact closed categorical diagrams and parameterized quantum circuits for downstream learning tasks. We evaluate QArabicCat on QuranNLP, a Classical Arabic benchmark for compositional language understanding, comparing multiple quantum representations under exact simulation and realistic IBM and IQM hardware noise models. Our results show that morphology-aware representations achieve competitive performance with word-level quantum models while producing more compact and robust circuits, reducing qubit and two-qubit gate requirements compared with token-based approaches. We do not claim quantum advantage; instead, we demonstrate that linguistic structure is a critical design principle for developing scalable and interpretable quantum language models. This work provides a foundation for extending Quantum NLP to more languages and highlights the role of language-aware representations in future quantum machine learning systems.