Elamur Research⌕
☰
home / papers / 2026 / Quantum Compositional NLP for Arabic: Grammar, Morphology, and Word Sense in Circuit Topology

Quantum Compositional NLP for Arabic: Grammar, Morphology, and Word Sense in Circuit Topology

Summary

The paper is the first to apply pregroup-grammar quantum compositional NLP (the DisCoCat line of work) to Arabic. Arabic was chosen because two of its properties punish models that ignore structure: it freely alternates between SVO, VSO and verbless Nominal sentences, and it builds words from roots and patterns. Arabic-specific pregroup rules turn each sentence into an IQP circuit, and the grammar decides how the word gates are wired together. All runs are simulated through lambeq, with CAMeL Tools and Stanza handling preprocessing. Three controlled tasks are tested against AraVec and AraBERT baselines: word order, morphological tense and verb sense. The headline result is an ablation. With grammar wiring alone and no trainable entangling gates, accuracy on matched word-order pairs sits at exactly chance. Adding a single parameterised entangling layer lifts it to roughly 65%, so the gain can be attributed to entanglement. The author also releases a vocabulary-controlled Arabic WSD set and describes an optimiser label-flipping effect. Accuracy stays far below fine-tuned AraBERT. The contribution is an interpretable, causally isolated structural signal rather than state-of-the-art performance.

Key points

  • On the matched-pair word-order task, bag-of-words models score 12.8%, and the structure-only QFM L0 circuit scores exactly 50.0%, a theoretically guaranteed value with zero variance. QFM L1, with one entangling layer, reaches 64.9% ± 3.2% (10 seeds × 15 folds; CI [62.8, 66.3]).
  • The new WSD dataset has 200 Arabic sentences over four verbs. It is built from matched pairs and draws subjects and objects from shared pools, so that word choice alone cannot reveal the sense.
  • In symmetric binary tasks, SPSA training often converges to inverted labels. Ancilla-qubit encoding cut this inversion rate from 99% to 23% on the hardest task.
  • Fine-tuned AraBERT achieves near-perfect accuracy on every task. The paper explicitly does not claim a quantum advantage.
  • The released resources are a corpus of 1,140 sentences across seven dataset keys, code, and result JSONs (GitHub disco_in_arabia, MIT licence). Kiraz's multi-tape analysis of Semitic morphology is linked to tensor products, and the author presents this as an interpretive step rather than a theorem.

Interesting ideas

  • I1. Build test pairs that contain exactly the same words but differ in structure, so any order-blind model can do no better than chance and only genuine structural knowledge scores above it. — For Mova (learn): Generate minimal pairs with identical lemmas and swapped agent/patient as negative controls, and require a bag-of-lemmas baseline to fail on them before trusting a tree-based gain.
  • I2. Use a parameter-free, structure-only variant whose score is known in advance as the ablation floor, so any improvement over it is cleanly attributable to the single component that was added. — For Mova (learn): For each new module (e.g. a repair rule or frame feature), report a deterministic "wiring only" baseline with a predictable score, making attribution exact rather than seed-dependent.
  • I3. Control vocabulary in sense-disambiguation data by drawing all arguments from shared subject and object pools, so a specific noun cannot give away the verb sense. — For Mova (frames): Build VerbNet sense and role tests where every sense of a verb draws arguments from the same pool, forcing the frame selector to rely on the tree rather than lexical co-occurrence.
  • I4. Keep grammatical wiring and lexical parameters separate, so a new variety can reuse the structure and retrain only the lexicon on a few examples. — For Mova (tree): When adapting gparse or the perceptron parser to another English variety or UD dialect, freeze the structural components and re-fit only word-level features on a small treebank.
  • I5. Watch symmetric binary tasks for training that converges to flipped labels, and break the symmetry with an extra anchoring input, as the ancilla qubit does here. — For Mova (learn): Track label-inversion rates in symmetric classifiers such as active/passive or speaker/addressee, and add an asymmetric anchor feature when flips appear.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Word sense disambiguation

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2607.14100 · arXiv 2607.14100

@misc{mohammed2026quantum,
  title = {{Quantum Compositional NLP for Arabic: Grammar, Morphology, and Word Sense in Circuit Topology}},
  author = {Mohammed, Wajahath},
  year = {2026},
  doi = {10.48550/arxiv.2607.14100},
  eprint = {2607.14100},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2607.14100}
}

References

25 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abbaszade, M., & Zomorodi, M. (2023). Toward quantum machine translation of syntactically distinct languages. arXiv:2307.16576. ↗
  2. Andreas, J., Beguš, G., Bronstein, M.M., Diamant, R., Delaney, D., Gero, S., Goldwasser, S., Gruber, D.F., de Haas, S., Malkin, P., Pavlov, N., Payne, R., Petri, G., Rus, D., Sharma, P., Tchernov, D., Tønnesen, P., Torralba, A., Vogt, D., & Wood, R.J. (2022). Toward understanding the communication in sperm whales. iScience, 25(6), 104393. https://doi.org/10.1016/j.isci.2022.104393 ↗
  3. Antoun, W., Baly, F., & Hajj, H. (2020). AraBERT: Transformer-based model for Arabic language understanding. Proceedings of the LREC 2020 Workshop on Arabic NLP, Marseille. find our card
  4. Coecke, B., Sadrzadeh, M., & Clark, S. (2010). Mathematical foundations for a compositional distributional model of meaning. Linguistic Analysis, 36, 345–384. arXiv:1003.4394. ↗
  5. Coecke, B., de Felice, G., Meichanetzidis, K., & Toumi, A. (2020). Foundations for near-term quantum natural language processing. arXiv:2012.03755. ↗
  6. Djemmal, R., & Belhadef, H. (2025). AraBERT-QC: a novel quantum-based classification architecture to classify short Arabic sentences. The Journal of Supercomputing. https://doi.org/10.1007/s11227-025-07966-5 ↗
  7. Duneau, T., Bruhn, S., Matos, G., Laakkonen, T., Saiti, K., Pearson, A., Meichanetzidis, K., & Coecke, B. (2024). Scalable and interpretable quantum natural language processing: An implementation on trapped ions. NeurIPS 2024 Workshop on Compositional Learning (Oral). arXiv:2409.08777. ↗
  8. Gero, S., Whitehead, H., & Rendell, L. (2016). Individual, unit and vocal clan level identity cues in sperm whale codas. Royal Society Open Science, 3(1), 150372. https://doi.org/10.1098/rsos.150372 ↗
  9. Habash, N. (2010). Introduction to Arabic natural language processing. Morgan & Claypool Publishers. find
  10. Holes, C. (2004). Modern Arabic: Structures, functions, and varieties. Georgetown University Press. find
  11. Kartsaklis, D., Fan, I., Yeung, R., Pearson, A., Lorenz, R., Toumi, A., Meichanetzidis, K., & Coecke, B. (2021). lambeq: An efficient high-level Python library for quantum NLP. Proceedings of EMNLP 2021: System Demonstrations, 195–202. arXiv:2110.04236. ↗
  12. Kiraz, G.A. (2001). Computational nonlinear morphology: With emphasis on Semitic languages. Cambridge University Press. find
  13. Lorenz, R., Pearson, A., Meichanetzidis, K., Kartsaklis, D., & Coecke, B. (2023). QNLP in practice: Running compositional models of meaning on a quantum computer. Journal of Artificial Intelligence Research, 76, 1305–1342. https://doi.org/ 10.1613/jair.1.14329 ↗
  14. Meichanetzidis, K., Gogioso, S., de Felice, G., Chiappori, N., Toumi, A., & Coecke, B. (2020). Quantum natural language processing on near-term quantum computers. Proceedings of QPL 2020, EPTCS 340, 221–241. https://doi.org/10.4204/EPTCS. 340.11 ↗
  15. Nausheen, F., Ahmed, K., Khan, M.I., & Riaz, F. (2025). Quantum natural language processing: A comprehensive review of models, methods, and applications. arXiv:2504.09909. ↗
  16. Obeid, O., Zalmout, N., Khalifa, S., Taji, D., Oudah, M., Eryani, F., Inoue, G., Nassar, A., Dakkak, S., Al-Khalifa, H., & Habash, N. (2020). CAMeL Tools: An open source Python toolkit for Arabic natural language processing. Proceedings of LREC 2020, 7702–7709. find our card
  17. Pasha, A., Al-Badrashiny, M., Kholy, A.E., Eskander, R., Diab, M., Habash, N., Pooleery, M., Rambow, O., & Roth, R.M. (2014). MADAMIRA: A fast, comprehensive tool for morphological analysis and disambiguation of Arabic. Proceedings of LREC 2014, 1094–1101. find our card
  18. Qi, P., Zhang, Y., Zhang, Y., Bolton, J., & Manning, C.D. (2020). Stanza: A Python natural language processing toolkit for many human languages. Proceedings of ACL 2020: System Demonstrations, 101–108. find our card
  19. Sadrzadeh, M. (2007). Pregroup analysis of Persian sentences. In C. Casadio & J. Lambek (Eds.), Computational algebraic approaches to natural language, 121–144. Polimetrica, Milan. find
  20. Sharma, P., Gero, S., Payne, R., Gruber, D., Rus, D., Torralba, A., & Andreas, J. (2024). Contextual and combinatorial structure in sperm whale vocalisations. Nature Communications, 15, 3617. https://doi.org/10.1038/s41467-024-47221-8 ↗
  21. Soliman, A.B., Eissa, K., & El-Beltagy, S.R. (2017). AraVec: A set of Arabic word embedding models for use in Arabic NLP. Procedia Computer Science, 117, 256–265. find
  22. Srivastava, N., Babu H, A., Mishra, A., & Tripathi, A. (2023). Enabling quantum natural language processing for Hindi language. arXiv:2312.01221. ↗
  23. Warstadt, A., Parrish, A., Liu, H., Mohananey, A., Peng, W., Wang, S.F., & Bowman, S.R. (2020). BLiMP: The Benchmark of Linguistic Minimal Pairs for English. Transactions of the Association for Computational Linguistics, 8, 229–249. find our card
  24. Waseem, M.H., Liu, J., Wang-Maścianica, V., & Coecke, B. (2022). Language-independence of DisCoCirc’s text circuits: English and Urdu. EPTCS, 366, 50–60. https://doi.org/10.4204/EPTCS.366.7 ↗
  25. Zaghouani, W., Diab, M., Mansouri, A., Pradhan, S., & Palmer, M. (2010). The revised Arabic PropBank. Proceedings of the Fourth Linguistic Annotation Workshop (LAW IV @ ACL 2010), 222–226. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mohammed-2026-quantum-compositional-nlp-arabic-grammar/