--- title: "ANLP-RG at NADI 2023 shared task: Machine Translation of" description: "This paper is the ANLP-RG team's entry to NADI 2023 Subtask 2. 11.14 BLEU. 11 references, cited by 3 papers here." authors: ["Wiem Derouich", "Sameh Kchaou", "Rahma Boujelbane"] year: 2023 venue: "Proceedings of ArabicNLP 2023" source: "https://aclanthology.org/2023.arabicnlp-1.75/" doi: "10.18653/v1/2023.arabicnlp-1.75" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/derouich-2023-anlp-rg-nadi-2023-shared/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/derouich-2023-anlp-rg-nadi-2023-shared/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/) / ANLP-RG at NADI 2023 shared task: Machine Translation of Arabic Dialects: A Comparative Study of Transformer Models # ANLP-RG at NADI 2023 shared task: Machine Translation of Arabic Dialects: A Comparative Study of Transformer Models **Authors:** Wiem Derouich, [Sameh Kchaou](https://md.research.elamur.ai/free/en/authors/sameh-kchaou/) and [Rahma Boujelbane](https://md.research.elamur.ai/free/en/authors/rahma-boujelbane/) · **Year:** 2023 · **Venue:** [Proceedings of ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2023](https://md.research.elamur.ai/free/en/venues/arabicnlp/2023/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2023.arabicnlp-1.75/) · [PDF](https://aclanthology.org/2023.arabicnlp-1.75.pdf) · [DOI 10.18653/v1/2023.arabicnlp-1.75](https://doi.org/10.18653/v1/2023.arabicnlp-1.75) · [OpenAlex](https://openalex.org/W4389518388) **Citations:** 3 citations (OpenAlex, 2026-10-14) · [3](https://md.research.elamur.ai/free/en/papers/derouich-2023-anlp-rg-nadi-2023-shared/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary This paper is the ANLP-RG team's entry to NADI 2023 Subtask 2. The task is sentence-level translation from four Arabic dialects (Egyptian, Emirati, Jordanian, Palestinian) into Modern Standard Arabic (MSA), using only the MADAR parallel corpus for training. MADAR has no Emirati–MSA section, so the team had to work around a dialect with no parallel data at all. They tried two families of approaches: - **Fine-tuning T5-style models.** Seven Arabic T5-style variants (AraT5-base, AraT5v2-base-1024, Sultan-ArabicT5, AraT5-MSA-Small/Base and others) were trained on several MADAR subsets. Some subsets added geographically neighbouring Gulf dialects (Qatari, Saudi) as a stand-in for Emirati. Others were augmented. - **Off-the-shelf tools.** They used TURJUMAN, plus back-translation through English with Google and Pons translators and dl-translate models. The best system was fine-tuned AraT5v2-base-1024, reaching 11.14 BLEU on dev and 10.02 on test. The tool-based routes scored close behind. Data augmentation helped little. The work shows how hard multi-dialect MT remains when a target variety has no parallel data. ## Key points - Data sizes: 111,096 training lines from MADAR, 400 dev sentences (100 per dialect) and 2,000 test sentences split evenly over the four dialects. - Corpus configurations tested include TS MADAR, Egy-Pal-Jor, Egy-Pal-Jor-Qat-Ksa, Aug-ALE, Aug-ALE-ALX-JER and Aug-PaysGolf. - Best model was AraT5v2-base-1024 (11.14 BLEU dev, 10.02 test). AraT5-tweet-base collapsed to a NADI BLEU of 0. - Tool-based results on dev: - Google, Pons and TURJUMAN scored about 10.1–11.0 BLEU overall. - On Emirati, scores ranged from roughly 7.7 to 13 BLEU depending on the tool. - dl-translate (nllb-200, m2m100) did clearly worse. - TURJUMAN was run with beam search (5 beams, no repeated 3-grams). Asking it for three outputs instead of one left BLEU essentially unchanged. ## Interesting ideas - When one target variety has no parallel data, train on its closest geographic neighbours as a proxy rather than skipping it, since related varieties share much of the signal. - Report scores per source dialect instead of only an overall number, because the dialect missing from training behaved very differently from tool to tool. - Sending text through a pivot language (dialect → English → MSA) serves as a cheap normalizer that can rival a dedicated fine-tuned model. - Requesting several decoded outputs and checking whether the metric moves is a quick stability probe for a generator. - The augmented examples inserted filler words and even changed a number (200 → 500), which likely explains why augmentation brought little gain. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (tree): when an English register or dialect lacks a UD treebank, borrow the nearest available treebanks as stand-in training data, and report the target-variety score separately to see if the proxy helped. - **I2** (learn): break parser and reader evaluations down by genre or source, so a weak slice hidden inside a decent average becomes visible. - **I3** (back): round-trip text through the meta-world and back to text, then compare the two meaning graphs, as a self-check of the generation and translation path. - **I4** (back): produce several surface realizations from one graph and confirm that re-parsing each gives the same frames, which flags fragile generation choices. - **I5** (learn): filter augmented silver data with a world-level consistency check (quantities, entities), so that altered meanings never enter training. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [University of Sfax](https://md.research.elamur.ai/free/en/universities/university-of-sfax/) — University, Sfax, Tunisia · 14 works · 231 citations 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works 🧭 Topics (2): [🔁 Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/) · 1,087 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works ## Cite Cite as: Wiem Derouich, Sameh Kchaou and Rahma Boujelbane. 2023. ANLP-RG at NADI 2023 shared task: Machine Translation of Arabic Dialects: A Comparative Study of Transformer Models. Proceedings of ArabicNLP 2023. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{derouich2023anlp, title = {{ANLP-RG at NADI 2023 shared task: Machine Translation of Arabic Dialects: A Comparative Study of Transformer Models}}, author = {Wiem Derouich, Sameh Kchaou and Rahma Boujelbane}, booktitle = {Proceedings of ArabicNLP 2023}, year = {2023}, doi = {10.18653/v1/2023.arabicnlp-1.75}, url = {https://doi.org/10.18653/v1/2023.arabicnlp-1.75} } ``` - [← previous Al-Jawaab at Qur'an QA 2023 Shared Task: Exploring Embeddings and GPT Models for Passage Retrieval and Reading Comprehension](https://md.research.elamur.ai/free/en/papers/zekiye-2023-al-jawaab-qur-qa-2023/) - [next → Automated De-Identification of Arabic Medical Records](https://md.research.elamur.ai/free/en/papers/kocaman-2023-automated-de-identification-arabic-medical/) - [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/12/) ## References 11 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Muhammad Abdul-Mageed, Chiyu Zhang, AbdelRahim Elmadany, Houda Bouamor, and Nizar Habash. 2021. [NADI 2021: The second nuanced Arabic dialect identification shared task](https://md.research.elamur.ai/free/en/papers/mageed-2021-nadi-2021-second-nuanced-arabic/). In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 244–259, Kyiv, Ukraine (Virtual). Association for Computational Linguistics. [↗](https://doi.org/10.48550/arxiv.2103.08466) 2. R. Al-Ibrahim and R. M. Duwairi. 2020. Neural machine translation from jordanian dialect to modern standard arabic. In 2020 11th International Conference on Information and Communication Systems (ICICS). [find](https://scholar.google.com/scholar?q=R%2E+Al%2DIbrahim+and+R%2E+M%2E+Duwairi%2E+2020%2E+Neural+machine+translation+from+jordanian+dialect+to+modern+standard+arabic%2E+In+2020+11th+International+Conference+on+Information+and+Communication+Systems+%28ICIC) 3. Sultan Alrowili and Vijay Shanker. 2022. [Generative approach for gender-rewriting task with ArabicT5](https://md.research.elamur.ai/free/en/papers/alrowili-2022-generative-approach-gender-rewriting-task/). In Proceedings of the The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 491– 495, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics. [↗](https://doi.org/10.48448/mpkx-ck82) 4. Laith Baniata, Seyoung Park, and Seong-Bae Park. 2018. A neural machine translation model for arabic dialects that utilizes multitask learning (mtl). In Computational Intelligence and Neuroscience. [↗](https://doi.org/10.1155/2018/7534712) 5. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, et al. 2018. The madar arabic dialect corpus and lexicon. In Proceedings of the eleventh international conference on language resources and evaluation (LREC 2018). [find](https://scholar.google.com/scholar?q=Houda+Bouamor%2C+Nizar+Habash%2C+Mohammad+Salameh%2C+Wajdi+Zaghouani%2C+Owen+Rambow%2C+Dana+Abdulrahim%2C+Ossama+Obeid%2C+Salam+Khalifa%2C+Fadhl+Eryani%2C+Alexander+Erdmann%2C+et+al%2E+2018%2E+The+madar+arabic+dialect+corpus) 6. Injy Hamed, Nizar Habash, Slim Abdennadher, and Ngoc Thang Vu. 2022. Investigating lexical replacements for arabic-english code-switched data augmentation. [find](https://scholar.google.com/scholar?q=Injy+Hamed%2C+Nizar+Habash%2C+Slim+Abdennadher%2C+and+Ngoc+Thang+Vu%2E+2022%2E+Investigating+lexical+replacements+for+arabic%2Denglish+code%2Dswitched+data+augmentation%2E) 7. Cong Duy Vu Hoang, Philipp Koehn, Gholamreza Haffari, and Trevor Cohn. 2018. [Iterative back-translation for neural machine translation](https://md.research.elamur.ai/free/en/papers/hoang-2018-iterative-back-translation-neural-machine/). In 2nd Workshop on Neural Machine Translation and Generation, pages 18–24. Association for Computational Linguistics. 8. Saméh Kchaou, Rahma Boujelbane, Emna Fsih, and Lamia Hadrich-Belguith. 2022. Standardisation of dialect comments in social networks in view of sentiment analysis : Case of Tunisian dialect. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 5436–5443, Marseille, France. European Language Resources Association. [find](https://scholar.google.com/scholar?q=Sam%C3%A9h+Kchaou%2C+Rahma+Boujelbane%2C+Emna+Fsih%2C+and+Lamia+Hadrich%2DBelguith%2E+2022%2E+Standardisation+of+dialect+comments+in+social+networks+in+view+of+sentiment+analysis+%3A+Case+of+Tunisian+dialect%2E+In+Proceed) 9. Saméh Kchaou, Rahma Boujelbane, and Lamia Hadrich. 2023. Hybrid pipeline for building arabic tunisian dialect-standard arabic neural machine translation model from scratch. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 22(3). [↗](https://doi.org/10.1145/3568674) 10. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2021. [Arat5: Text-to-text transformers for arabic language generation](https://md.research.elamur.ai/free/en/papers/nagoudi-2022-arat5-text-text-transformers-arabic/). arXiv preprint arXiv:2109.12068. [↗](https://arxiv.org/abs/2109.12068) 11. El Moatez Billah Nagoudi, AbdelRahim Elmadany, and Muhammad Abdul-Mageed. 2022. [Turjuman: A public toolkit for neural arabic machine translation](https://md.research.elamur.ai/free/en/papers/nagoudi-2022-turjuman-public-toolkit-neural-arabic/). In Proceedings of the 5th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT5), Marseille, France. European Language Resource Association. [↗](https://doi.org/10.48550/arxiv.2206.03933) *Updated 1 November 2026*