---
title: "TTLab at AraSentEval — paper by Abusaleh et al. (2026)"
description: "The paper describes TTLab's entry to the AraSentEval 2026 shared task. 21 references. Among them: 0."
authors: ["Ali Abusaleh", "Bhuvanesh Verma", "Alexander Mehler"]
year: 2026
venue: "The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks"
source: "https://aclanthology.org/2026.osact-1.35/"
doi: "10.63317/4wj6s3ys5osk"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/abusaleh-2026-ttlab-arasenteval-sarf-sentiment-analysis/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/abusaleh-2026-ttlab-arasenteval-sarf-sentiment-analysis/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/46/) / TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based Fusion for Multi-Dialectal Arabic

# TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based Fusion for Multi-Dialectal Arabic

**Authors:** [Ali Abusaleh](https://md.research.elamur.ai/free/en/authors/ali-abusaleh/), [Bhuvanesh Verma](https://md.research.elamur.ai/free/en/authors/bhuvanesh-verma/) and [Alexander Mehler](https://md.research.elamur.ai/free/en/authors/alexander-mehler/) · **Year:** 2026 · **Venue:** [The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks](https://md.research.elamur.ai/free/en/venues/osact/) ([Workshop on Open-Source Arabic Corpora and Processing Tools with Shared Tasks 2026](https://md.research.elamur.ai/free/en/venues/osact/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.osact-1.35/) · [PDF](https://aclanthology.org/2026.osact-1.35.pdf) · [DOI 10.63317/4wj6s3ys5osk](https://doi.org/10.63317/4wj6s3ys5osk) · [OpenAlex](https://openalex.org/W7166524314)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper describes TTLab's entry to the AraSentEval 2026 shared task. The task is three-way sentiment classification (positive, negative, neutral) of hotel-review sentences written in four Arabic dialects. The authors argue that rich morphology and dialectal lexical variation make the task hard. Their system, SARF, encodes every sentence three times with one shared MARBERTv2 encoder: as the raw surface text, as stems and as roots. The stem and root encodings are combined with the surface encoding through cross-attention and averaging. A hybrid head, adapted from He and Abisado (2024), then sends the result through a CNN branch for local n-gram cues and a BiLSTM branch for longer-range context. The model was first trained on SemEval-2016 Arabic hotel reviews and then fine-tuned on the task data (MDS-3). It reached a macro-F1 of about 0.926 on the blind test set and placed 2nd of 15 teams. A human re-annotation study found substantial label noise in the neutral class. The authors conclude that annotation consistency, rather than model design, may be what limits scores on this task.

## Key points
- Data: MDS-3 has 3,000 sentences, of which 1,731 are in the training split. It covers Saudi, Jordanian, Moroccan (Darija) and Egyptian. The texts are hotel reviews translated into each dialect and checked by native speakers.
- Setup and tools:
  - The SemEval-2016 ABSA Arabic hotel reviews were used for 10 epochs of preliminary training, followed by 7 epochs of fine-tuning at a lower learning rate.
  - Stems and roots were extracted with the rule-based tools PyArabic, Tashaphyne and Qalsadi.
  - The head uses a 128-dimensional LSTM and a CNN with 200 filters (kernel sizes 3, 4 and 5).
- Development set macro-F1:
  - The plain BERT-CNN-BiLSTM model scored 0.9425.
  - Adding cross-morphological attention gave 0.9388.
  - An ensemble of MARBERTv2, AraBERT and QARIB gave 0.9316.
- Test set: macro-F1 0.9262 (the abstract reports 0.9263), balanced accuracy 0.8941, rank 2 of 15.
- Label audit:
  - Agreement between the authors' annotators and the original labels was κ = 0.536 in a relaxed setting. In a strict setting (68 samples where more than one annotator agreed) it fell to κ = 0.296.
  - Of 60 originally neutral items, 23 were relabelled positive and 11 negative.
  - On disputed items, the model reproduced the original (contested) label 46.9% of the time.
  - Accuracy on Darija in this analysis was 44.4%, compared with roughly 65–67% for Jordanian and Saudi.

## Interesting ideas
- The surface sentence acts as the attention query, while stem and root views supply keys and values, so normalised forms add evidence without replacing the original wording (the paper reports that rooting alone discards sentiment cues).
- A sample of the gold labels is re-annotated, and agreement is reported twice: once for all re-annotated items and once only for items several annotators agreed on, which separates real gold errors from individual annotator disagreement.
- On items where the gold label is disputed, the authors count how often the model copies the contested label versus how often it predicts the corrected one, which measures how much annotation bias the model has memorised.
- Errors are broken down by dialect and by error direction, which exposes a "safe bias" of collapsing polar items into neutral that hits one dialect (Darija) hardest.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tokens): lemma and morphological-feature views could be attached to surface tokens as auxiliary channels in the tagger, keeping the surface form primary instead of hard-normalising it.
- **I2** (learn): silver trees built from witness parsers could be audited the same way, reporting agreement with gold under loose and multi-witness-consensus filters before deciding which data to trust.
- **I3** (learn): this could serve as a negative-control style metric showing whether the parser imitates known treebank annotation errors rather than generalising.
- **I4** (tree): evaluating the parser separately for each UD treebank or dialect, and checking which labels absorb the errors, could reveal systematic fallback labels such as overusing a generic dependency relation.

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

📚 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): [😊 Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · 972 works · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Ali Abusaleh, Bhuvanesh Verma and Alexander Mehler. 2026. TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based Fusion for Multi-Dialectal Arabic. The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{abusaleh2026ttlab,
  title = {{TTLab at AraSentEval: SARF( صرف) Sentiment Analysis via Root-based Fusion for Multi-Dialectal Arabic}},
  author = {Ali Abusaleh, Bhuvanesh Verma and Alexander Mehler},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks},
  year = {2026},
  doi = {10.63317/4wj6s3ys5osk},
  url = {https://doi.org/10.63317/4wj6s3ys5osk}
}
```

- [← previous ttda704 at SemEval-2026 Task 6: Structured Chain-of-Thought Prompting for Political Evasion Detection](https://md.research.elamur.ai/free/en/papers/tan-2026-ttda704-semeval-2026-task-6/)
- [next → TTLab at Daleel 2026: STAR-Ar, Sequence Tagging for Argument Recognition in Arabic](https://md.research.elamur.ai/free/en/papers/verma-2026-ttlab-daleel-2026-star-ar/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/46/)

## References

21 references, 10 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, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2021. [ARBERT & MARBERT: Deep bidirectional transformers for Arabic](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/). In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 7088–7105, Online. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/2021.acl-long.551)
2. Hind Alamro, Manal Alshehri, Basma Alharbi, Zuhair Khayyat, Manal Kalkatawi, Inji Ibrahim Jaber, and Xiangliang Zhang. 2021. [Overview of the arabic sentiment analysis 2021 competition at kaust](https://md.research.elamur.ai/free/en/papers/alamro-2021-overview-arabic-sentiment-analysis-2021/). [↗](https://doi.org/10.48550/arxiv.2109.14456)
3. Maram Alharbi, Saad Ezzini, Hansi Hettiarachchi, Tharindu Ranasinghe, and Ruslan Mitkov. 2025a. Evaluating large language models on arabic dialect sentiment analysis. In Proceedings of the 15th International Conference on Recent Advances in Natural Language Processing (RANLP). [find](https://scholar.google.com/scholar?q=Maram+Alharbi%2C+Saad+Ezzini%2C+Hansi+Hettiarachchi%2C+Tharindu+Ranasinghe%2C+and+Ruslan+Mitkov%2E+2025a%2E+Evaluating+large+language+models+on+arabic+dialect+sentiment+analysis%2E+In+Proceedings+of+the+15th+Intern)
4. Maram I Alharbi, Salmane Chafik, Saad Ezzini, Ruslan Mitkov, Tharindu Ranasinghe, and Hansi Hettiarachchi. 2025b. [Ahasis: Shared task on sentiment analysis for arabic dialects](https://md.research.elamur.ai/free/en/papers/alharbi-2025-ahasis-shared-task-sentiment-analysis/). In Proceedings of the Shared Task on Sentiment Analysis for Arabic Dialects, pages 1–6.
5. Maram I Alharbi, Saad Ezzini, Hansi Hettiarachchi, Tharindu Ranasinghe, and Ruslan Mitkov. 2025c. [Ador: Dataset for arabic dialects in hotel reviews: A human benchmark for sentiment analysis](https://md.research.elamur.ai/free/en/papers/alharbi-2025-ador-dataset-arabic-dialects-hotel/). In Proceedings of the First Workshop on Advancing NLP for Low-Resource Languages, pages 187–191.
6. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. [AraBERT: Transformer-based model for Arabic language understanding](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/). In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection, pages 9–15, Marseille, France. European Language Resource Association. [↗](https://doi.org/10.48550/arxiv.2003.00104)
7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. [BERT: Pre-training of deep bidirectional transformers for language understanding](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/). In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, Minneapolis, Minnesota. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/n19-1423)
8. Obeida ElJundi, Wissam Antoun, Nour El Droubi, Hazem Hajj, Wassim El-Hajj, and Khaled Shaban. 2019. [hULMonA: The universal language model in Arabic](https://md.research.elamur.ai/free/en/papers/eljundi-2019-hulmona-universal-language-model-arabic/). In Proceedings of the Fourth Arabic Natural Language Processing Workshop, pages 68–77, Florence, Italy. Association for Computational Linguistics.
9. Saad Ezzini, Shadi Abudalfa, Maram Alharbi, Salmane Chafik, Hamzah Luqman, Mo El-Haj, Paul Rayson, Samaher Alghamdi, and Reem Alotaibi. 2026. AraSentEval: A shared task on sentiment analysis and swapping in arabic. In Proceedings of the 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7), co-located with the Language Resources and Evaluation Conference (LREC 2026), Palma, Mallorca, Spain. [find](https://scholar.google.com/scholar?q=Saad+Ezzini%2C+Shadi+Abudalfa%2C+Maram+Alharbi%2C+Salmane+Chafik%2C+Hamzah+Luqman%2C+Mo+El%2DHaj%2C+Paul+Rayson%2C+Samaher+Alghamdi%2C+and+Reem+Alotaibi%2E+2026%2E+AraSentEval%3A+A+shared+task+on+sentiment+analysis+and+swapp)
10. Imène Guellil and Faiçal Azouaou. 2017. Asda: Analyseur syntaxique du dialecte alg {\’e} rien dans un but d’analyse s {\’e} mantique. ArXiv preprint, abs/1707.08998. [↗](https://arxiv.org/abs/1707.08998)
11. Aixiang He and Mideth Abisado. 2024. Text sentiment analysis of douban film short comments based on bert-cnn-bilstm-att model. Ieee Access, 12:45229–45237. [↗](https://doi.org/10.1109/access.2024.3381515)
12. Pratik Joshi, Sebastin Santy, Amar Budhiraja, Kalika Bali, and Monojit Choudhury. 2020. [The state and fate of linguistic diversity and inclusion in the NLP world](https://md.research.elamur.ai/free/en/papers/joshi-2020-state-fate-linguistic-diversity-inclusion/). In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 6282–6293, Online. Association for Computational Linguistics.
13. Bing Liu. 2012. Sentiment analysis and opinion mining, volume 5 of Synthesis lectures on human language technologies. Morgan & Claypool Publishers. [find](https://scholar.google.com/scholar?q=Bing+Liu%2E+2012%2E+Sentiment+analysis+and+opinion+mining%2C+volume+5+of+Synthesis+lectures+on+human+language+technologies%2E+Morgan+%26+Claypool+Publishers%2E)
14. Maria Pontiki, Dimitrios Galanis, Harris Papageorgiou, Ion Androutsopoulos, Suresh Manandhar, Mohammad Al-Smadi, Mahmoud Al-Ayyoub, Yanyan Zhao, Bing Qin, Orphée De Clercq, et al. 2016. [Semeval-2016 task 5: Aspect based sentiment analysis](https://md.research.elamur.ai/free/en/papers/pontiki-2016-semeval-2016-task-5-aspect/). In Proceedings of the 10th international workshop on semantic evaluation (SemEval-2016), pages 19–30. [↗](https://doi.org/10.60692/1cmj7-8b839)
15. Stephan Procházka. 2021. Arabic Dialectology, Cambridge Handbooks in Language and Linguistics, page 214–243. Cambridge University Press. [find](https://scholar.google.com/scholar?q=Stephan+Proch%C3%A1zka%2E+2021%2E+Arabic+Dialectology%2C+Cambridge+Handbooks+in+Language+and+Linguistics%2C+page+214%E2%80%93243%2E+Cambridge+University+Press%2E)
16. Said A Salloum, Mostafa Al-Emran, Azza Abdel Monem, and Khaled Shaalan. 2017. Using text mining techniques for extracting information from research articles. In Intelligent natural language processing: Trends and Applications, pages 373–397. Springer. [↗](https://doi.org/10.1007/978-3-319-67056-0_18)
17. Zhiqiang Shi and Ruchit Agrawal. 2025. [A comprehensive survey of contemporary Arabic sentiment analysis: Methods, challenges, and future directions](https://md.research.elamur.ai/free/en/papers/shi-2025-comprehensive-survey-contemporary-arabic-sentiment/). In Findings of the Association for Computational Linguistics: NAACL 2025, pages 3760–3772, Albuquerque, New Mexico. Association for Computational Linguistics.
18. Taha Zerrouki. 2010. pyarabic, an arabic language library for python. [find](https://scholar.google.com/scholar?q=Taha+Zerrouki%2E+2010%2E+pyarabic%2C+an+arabic+language+library+for+python%2E)
19. Taha Zerrouki. 2012a. qalsadi, arabic mophological analyzer library for python. [find](https://scholar.google.com/scholar?q=Taha+Zerrouki%2E+2012a%2E+qalsadi%2C+arabic+mophological+analyzer+library+for+python%2E)
20. Taha Zerrouki. 2012b. Tashaphyne, arabic light stemmer. [find](https://scholar.google.com/scholar?q=Taha+Zerrouki%2E+2012b%2E+Tashaphyne%2C+arabic+light+stemmer%2E)
21. Taha Zerrouki. 2020. Towards an open platform for arabic language processing. [↗](https://doi.org/10.13140/rg.2.2.29882.82881)

*Updated 1 November 2026*