İngilizce–Türkçe Sinirsel Makine Çevirisinde Katman Sayısı ve Dikkat Başlığı Sayısının Çeviri Kalitesine Etkisi

Öz

Morfolojik olarak zengin ve düşük kaynaklı bir dil olan Türkçe, sondan eklemeli yapısı, veri seyrekliği ve sözdizimsel esnekliği nedeniyle Sinirsel Makine Çevirisi alanında önemli zorluklar barındırmaktadır. İngilizce–Türkçe dil çifti, yapısal asimetrisi nedeniyle hizalama ve anlam temsili süreçlerinde ek güçlükler yaratmaktadır. Bu çalışma, model derinliği ve dikkat başlığı sayısının çeviri kalitesi üzerindeki etkisini inceleyerek, morfolojik olarak karmaşık diller için sinirsel mimarilerin nasıl optimize edilebileceğine dair kapsamlı bir değerlendirme sunmayı amaçlamaktadır. Workshop on Machine Translation’16 (WMT16) veri kümesi Bayt Çifti Kodlaması (BPE) yöntemiyle belirteçlenmiş (tokenize edilmiş) ve 12 farklı konfigürasyon ile (Katman Sayısı=1–6, Başlık Sayısı=4–8) değerlendirilmiştir. Deneysel sonuçlar, orta derinlikteki yapıların (özellikle Katman Sayısı =5, Başlık Sayısı =8) BLEU, ChrF ve TER metriklerinde en yüksek başarıyı gösterdiğini ortaya koymuştur. Katman ve başlık sayısının aşırı artırılması ise aşırı öğrenmeye ve bağlam genelleme zayıflığına yol açmıştır. Bulgular, Türkçe gibi morfolojik açıdan karmaşık dillerde mimari karmaşıklığın dikkatli biçimde dengelenmesi gerektiğini göstermektedir.

Anahtar Kelimeler

• Doğal Dil İşleme
• Sinirsel Makine Çevirisi
• Transformatör Mimarisi

The Effect of Number of Layers and Attention Heads on Translation Quality in English-Turkish Neural Machine Translation

Öz

Turkish, a morphologically rich and low-resource language, poses significant challenges in Neural Machine Translation (NMT) due to its agglutinative structure, data sparsity, and syntactic flexibility. The English–Turkish language pair, due to its structural asymmetry, creates additional challenges in alignment and semantic representation. This study aims to provide a comprehensive assessment of how neural architectures can be optimized for morphologically complex languages by examining the impact of model depth and the number of attention heads on translation quality. The WMT16 dataset was tokenized using Byte Pair Encoding (BPE), and twelve different architectural structures (Number of Layers =1–6, Number of Heads =4–8) were evaluated. Experimental results revealed that structures with medium depth (especially Number of Layers=5, Number of Heads =8) performed best in the BLEU, ChrF, and TER metrics. However, excessively increasing the number of layers and heads led to overlearning and poor context generalization. The findings suggest that architectural complexity must be carefully balanced in morphologically complex languages such as Turkish.

Anahtar Kelimeler

• Natural Language Processing
• Neural Machine Translation
• Transformer Architecture

Kaynakça

1. Bahdanau, D., Cho, K., Bengio, Y. Neural Machine Translation by Jointly Learning to Align and Translate, ArXiv, 2015. DOI: 10.48550/arXiv.1409.0473.
2. Sutskever, I., Vinyals, O., Le, Q. V. Sequence to Sequence Learning with Neural Networks, Proceedings of the 27th International Conference on Neural Information Processing Systems (NIPS 2014), Curran Associates, Inc., Red Hook, 2014, pp. 3104–3112. DOI: 10.48550/arXiv.1409.3215.
3. Vaswani, A. Attention is All You Need, Advances in Neural Information Processing Systems, ACM, New York, 2017.
4. Gong, L., Li, Y., Guo, J., Yu, Z., Gao, S. Enhancing Low-Resource Neural Machine Translation with Syntax-Graph Guided Self-Attention, Knowledge-Based Systems, 246(11), 2022, 108615. DOI: 10.1016/j.knosys.2022.108615.
5. Pan, Y., Li, X., Yang, Y., Dong, R. Multi-Source Neural Model for Machine Translation of Agglutinative Language, Future Internet, 12(6), 2020, 96. DOI: 10.3390/fi12060096.
6. Wan, F., Li, P. A Neural Machine Translation Method Based on Split Graph Convolutional Self-Attention Encoding, PeerJ Computer Science, 10(6), 2024, e1886. DOI: 10.7717/peerj-cs.1886.
7. Hujon, A. V., Singh, T. D., Amitab, K. Neural Machine Translation Systems for English to Khasi: A Case Study of an Austroasiatic Language, Expert Systems with Applications, 238(2), 2024, 121813. DOI: 10.1016/j.eswa.2023.121813.
8. Yazar, B. K., Kılıç, E. Improving Low-Resource Kazakh-English and Turkish-English Neural Machine Translation Using Transfer Learning and Part of Speech Tags, IEEE Access, 13, 2025, pp. 32341–32356. DOI: 10.1109/ACCESS.2025.3542491.

9. Papineni, K., Roukos, S., Ward, T., Zhu, W. J. BLEU: A Method for Automatic Evaluation of Machine Translation, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, 2002, pp. 311–318.
10. Banerjee, S., Lavie, A. METEOR: An Automatic Metric for MT Evaluation with Improved Correlation with Human Judgments, Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, 2005, pp. 65–72.
11. Snover, M., Dorr, B., Schwartz, R., Micciulla, L., Makhoul, J. A Study of Translation Edit Rate with Targeted Human Annotation, Proceedings of the 7th Conference of the Association for Machine Translation in the Americas: Technical Papers, 2006, pp. 223–231.
12. Popović, M. chrF: Character n-gram F-score for Automatic MT Evaluation, Proceedings of the Tenth Workshop on Statistical Machine Translation, 2015, pp. 392–395.
13. Rubino, R., Marie, B., Dabre, R., Fujita, A., Utiyama, M., Sumita, E. Extremely Low-Resource Neural Machine Translation for Asian Languages, Machine Translation, 34(4), 2020, pp. 347–382.
14. Araabi, A., Monz, C. Optimizing Transformer for Low-Resource Neural Machine Translation, Proceedings of the 28th International Conference on Computational Linguistics, Spain, 2020, pp. 3429–3435.
15. Phua, Y. T., Navaratnam, S., Kang, C. M., Che, W. S. Sequence-to-Sequence Neural Machine Translation for English-Malay, IAES International Journal of Artificial Intelligence, 11(2), 2022, pp. 658–665. DOI: 10.11591/ijai.v11.i2.pp658-665.
16. Puspitaningrum, D. A Study of English-Indonesian Neural Machine Translation with Attention (Seq2Seq, ConvSeq2Seq, RNN, and MHA): A Comparative Study of NMT on English-Indonesian, Proceedings of the 6th International Conference on Sustainable Information Engineering and Technology, ACM, New York, 2021, pp. 271–280.
17. Zoph, B., Yuret, D., May, J., Knight, K. Transfer Learning for Low-Resource Neural Machine Translation, ArXiv Preprint, 2016. DOI: 10.48550/arXiv.1604.02201.
18. Özdemir, Ö., Akın, E. S., Velioğlu, R., Dalyan, T. A Comparative Study of Neural Machine Translation Models for Turkish Language, Journal of Intelligent & Fuzzy Systems, 42(3), 2022, pp. 2103–2113.
19. Xu, T., Ozbek, O. I., Marks, S., Korrapati, S., Ahmadnia, B. Spanish-Turkish Low-Resource Machine Translation: Unsupervised Learning vs Round-Tripping, American Journal of Artificial Intelligence, 4(2), 2020, pp. 42–49. DOI: 10.11648/j.ajai.20200402.11.
20. Currey, A., Miceli-Barone, A. V., Heafield, K. Copied Monolingual Data Improves Low-Resource Neural Machine Translation, Proceedings of the Second Conference on Machine Translation, 2017, pp. 148–156.
21. Liu, X., Duh, K., Liu, L., Gao, J. Very Deep Transformers for Neural Machine Translation, ArXiv Preprint, 2020. DOI: 10.48550/arXiv.2008.07772
22. Babaali, B., Salem, M., Alharbe, N. R. Breaking language barriers with ChatGPT: enhancing low-resource machine translation between Algerian Arabic and MSA, International Journal of Information Technology, 2024. DOI: 10.1007/s41870-024-01926-7.
23. Liang, X., Khaw, Y. J., Liew, S. Y., Tan, T. P., Qin, D. Toward Low-Resource Languages Machine Translation: A Language-Specific Fine-Tuning with LoRA for Specialized Large Language Models, IEEE Access, 2025. DOI: 10.1109/ACCESS.2025.3549795.
24. Zhu, S., Pan, L., Jian, D., Xiong, D. Overcoming language barriers via machine translation with sparse Mixture-of-Experts fusion of large language models, Information Processing and Management, 2025. DOI: 10.1016/j.ipm.2025.104078.
25. Song, S., Chen, Y., Hu, X., Zhang, J. Domain-Aware Transformer for Multi-Domain Neural Machine Translation, Computers, Materials & Continua, 2026. DOI: 10.32604/cmc.2025.072392.
26. Tukeyev, U., Shormakova, A., Karibayeva, A., Rakhimova, D., Abduali, B., Amirova, D., Rakhmanberdi, N., Aliyev, R. An Integrated Approach to Adapting Open-Source AI Models for Machine Translation of Low-Resource Turkic Languages, Computers, 2026. DOI: 10.3390/computers15020073.
27. Satir, E., Bulut, H. Controlled beam search for neural machine translation using subword units leveraging phrase-based statistical machine translation outputs, Discover Computing, 2026. DOI: 10.1007/s10791-025-09633-y.
28. Gao, Y., Hou, F., Jahnke, H. Pre-ordering representations improve low-resource neural machine translation and application in the Māori language, Multimedia Tools and Applications, 2026. DOI: 10.1007/s11042-026-21153-5.
29. WMT16 Translation Task, Available at: https://www.statmt.org/wmt16/translation-task.html (accessed on 15.07.2025).
30. Tan, Z., Wang, S., Yang, Z., Chen, G., Huang, X., Sun, M., Liu, Y. Neural Machine Translation: A Review of Methods, Resources, and Tools, AI Open, 2020, pp. 5–21. DOI:10.1016/j.aiopen.2020.11.001
31. Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., Artzi, Y. Bertscore: Evaluating text generation with bert, arXiv preprint, 2019. arXiv:1904.09675.
32. Finkelstein, Mara, et al. TranslateGemma Technical Report. arXiv preprint arXiv:2601.09012, 2026.
33. Costa-Jussà, M. R., Cross, J., Çelebi, O., Elbayad, M., Heafield, K., Heffernan, K. et al. (NLLB Team). No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672, 2022.
34. Grattafiori, A., Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Vasic, P., et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024.