pith. sign in

arxiv: 2607.00143 · v1 · pith:JAOML7XPnew · submitted 2026-06-30 · 💻 cs.CL · cs.AI

Hate Speech Detection in Turkish and Arabic Languages: A Comprehensive Study

Somaiyeh Dehghan , G\"ok\c{c}e Uludo\u{g}an , Mehmet Umut \c{S}en , Elif Erol , Arzucan \"Ozg\"ur , Berrin Yanikoglu This is my paper

Pith reviewed 2026-07-02 19:22 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords hate speech detectionTurkishArabicBERTdatasetmulti-task learningspan detectioncontent moderation
0
0 comments X

The pith

A new hate speech dataset for five Turkish topics and Arabic refugees comes with BERT models for category, intensity, target, and span tasks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a dataset drawn from Turkish social media posts on refugees, the Israel-Palestine conflict, anti-Greek sentiment, ethnic and religious communities, and LGBTI+ issues, together with Arabic examples focused on refugees. It pairs this data with fine-tuned BERT models that perform four linked tasks at once: assigning hate categories, scoring intensity, naming the target group, and marking the exact hateful spans in text. A sympathetic reader would see value in moving beyond simple yes-no detection toward finer-grained tools that could support more precise content decisions on platforms. The work treats the combination of topic-specific collection and multi-output modeling as the route to better coverage of hate speech in these languages.

Core claim

We introduce a comprehensive hate speech dataset covering five distinct topics in Turkish: refugees, the Israel-Palestine conflict, anti-Greek sentiment in Turkey, ethnic or religious communities (Alevis, Armenians, Arabs, Jews, and Kurds), and LGBTI+, alongside one topic in Arabic (refugees). In addition, we develop state-of-the-art BERT-based models to address multiple dimensions of hate speech analysis, including hate category classification, hate intensity prediction, target identification, and hate speech span detection, enabling a comprehensive understanding of hateful content in online discourse.

What carries the argument

BERT-based models fine-tuned on the new multi-topic dataset to output category labels, intensity scores, target groups, and character-level hateful spans in a single pass.

If this is right

  • Moderation systems can move from binary flags to outputs that also report how intense the hate is and which exact phrases trigger it.
  • Analysis becomes possible for hate patterns tied to specific Turkish topics such as ethnic communities or LGBTI+ discussions.
  • The same modeling approach can be applied to additional low-resource language settings where only topic-labeled text is available.
  • Links between online content and offline incidents can be studied with finer target and intensity labels attached to each post.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The dataset could support experiments testing whether intensity or span signals transfer across the Turkish topics better than category labels alone.
  • Deployment on live streams might expose cases where context outside the post changes whether the same text counts as hate.
  • Span detection outputs could be used to generate training examples for simpler keyword-based filters in the same languages.

Load-bearing premise

The posts gathered for the dataset reflect typical real-world hate speech in Turkish and Arabic on the listed topics, and the fine-tuned models will maintain reliable performance without major annotation errors or shifts in language use.

What would settle it

Fresh posts collected from the same platforms and topics produce substantially lower accuracy on any of the four tasks, or repeated annotation rounds show low agreement on labels and spans.

read the original abstract

Online hate speech has been linked to a global rise in violence against minorities, including incidents such as mass shootings, lynchings, and ethnic cleansing. Societies grappling with this issue, particularly when hate speech targets specific groups based on religion, race, ethnicity, culture, nationality, or migration status, face the challenge of balancing freedom of expression with the need for effective content moderation on widely used online platforms. In response to this challenge, we introduce a comprehensive hate speech dataset covering five distinct topics in Turkish: refugees, the Israel-Palestine conflict, anti-Greek sentiment in Turkey, ethnic or religious communities (Alevis, Armenians, Arabs, Jews, and Kurds), and LGBTI+, alongside one topic in Arabic (refugees). In addition, we develop state-of-the-art BERT-based models to address multiple dimensions of hate speech analysis, including hate category classification, hate intensity prediction, target identification, and hate speech span detection, enabling a comprehensive understanding of hateful content in online discourse.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript introduces a comprehensive hate speech dataset covering five topics in Turkish (refugees, Israel-Palestine conflict, anti-Greek sentiment, ethnic/religious communities including Alevis/Armenians/Arabs/Jews/Kurds, and LGBTI+) and one topic in Arabic (refugees). It additionally develops BERT-based models for multiple tasks: hate category classification, hate intensity prediction, target identification, and hate speech span detection.

Significance. If the dataset collection, annotation quality (including IAA), and model evaluations are rigorously documented and the models demonstrably outperform baselines, the work would provide valuable resources for hate speech detection in under-resourced languages, supporting content moderation research.

major comments (1)
  1. [Abstract] Abstract: The claim of developing 'state-of-the-art BERT-based models' for the listed tasks is unsupported by any reported metrics, baselines, data splits, or performance numbers, making it impossible to evaluate whether the central empirical claims hold.
minor comments (1)
  1. [Abstract] Abstract: No information is provided on data sources, collection methodology, annotation protocol, or inter-annotator agreement, which are standard requirements for dataset papers.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and the recommendation for major revision. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim of developing 'state-of-the-art BERT-based models' for the listed tasks is unsupported by any reported metrics, baselines, data splits, or performance numbers, making it impossible to evaluate whether the central empirical claims hold.

    Authors: We agree that the abstract's phrasing is problematic. The manuscript body contains the full experimental results, baselines, data splits, and metrics for the BERT models on the four tasks, but the abstract does not report any of these numbers and therefore cannot support the 'state-of-the-art' claim. In the revised version we will remove the phrase 'state-of-the-art' from the abstract and replace it with a neutral description of the models developed, ensuring the abstract makes no unsubstantiated performance claims. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

This is an empirical paper focused on dataset construction for hate speech detection across Turkish and Arabic topics plus standard BERT fine-tuning for classification, intensity prediction, target ID, and span detection. No derivation chain, equations, fitted parameters presented as predictions, or self-citation load-bearing steps exist in the provided text. Central claims rest on data collection protocols and model performance metrics that are externally verifiable rather than reducing to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is an empirical NLP study focused on dataset creation and model training; the abstract introduces no free parameters, mathematical axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5738 in / 1150 out tokens · 28687 ms · 2026-07-02T19:22:20.729702+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    A lexicon-based approach for hate speech detection.International Journal of Multimedia and Ubiquitous Engineering, 10:215–230, 2015

    Gitari N M, Zuping Z, Damien H, and Jun Long J. A lexicon-based approach for hate speech detection.International Journal of Multimedia and Ubiquitous Engineering, 10:215–230, 2015. 13 DEHGHAN et al. / A PREPRINT

    2015

  2. [2]

    Contextual-lexicon approach for abusive language detection

    Vargas F, G´ oes F R, Carvalho I, Benevenuto F, and Pardo T. Contextual-lexicon approach for abusive language detection. InProceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021), pages 1438–1447, Held Online, September 2021. INCOMA Ltd

    2021

  3. [3]

    A corpus of Turkish offensive language on social media

    C ¸ ¨ oltekin C ¸ . A corpus of Turkish offensive language on social media. InProceedings of the Twelfth Language Resources and Evaluation Conference, pages 6174–6184, Marseille, France, May 2020. European Language Resources Association

    2020

  4. [4]

    Automated detection of hate speech towards woman on twitter

    S ¸ahi H, Kılı¸ cY, and Saˇ glam R B. Automated detection of hate speech towards woman on twitter. In2018 3rd International Conference on Computer Science and Engineering (UBMK), pages 533–536, 2018

    2018

  5. [5]

    Automatic hate speech detection using machine learning: A comparative study.International Journal of Advanced Computer Science and Applications, 11, 2020

    Abro S, Shaikh S, Khand Z H, Ali Z, Khan S, and Mujtaba G. Automatic hate speech detection using machine learning: A comparative study.International Journal of Advanced Computer Science and Applications, 11, 2020

    2020

  6. [6]

    Identifying hate speech using neural networks and discourse analysis techniques

    H¨ us¨ unbeyi Z M, Akar D, and ¨Ozg¨ ur A. Identifying hate speech using neural networks and discourse analysis techniques. InProceedings of the First Workshop on Language Technology and Resources for a Fair, Inclusive, and Safe Society within the 13th Language Resources and Evaluation Conference, pages 32–41, Marseille, France, June

  7. [7]

    European Language Resources Association

  8. [8]

    A Turkish hate speech dataset and detection system

    Beyhan F, C ¸ arık B, Arın I, Terzio˘ glu A, Yanikoglu B, and Yeniterzi R. A Turkish hate speech dataset and detection system. InProceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), page 4177–4185, 2022.https://aclanthology.org/2022.lrec-1.443/

    2022

  9. [9]

    Large-scale hate speech detection with cross-domain transfer

    Toraman C, S ¸ahinu¸ c F, and Yilmaz E. Large-scale hate speech detection with cross-domain transfer. InProceedings of the Thirteenth Language Resources and Evaluation Conference, pages 2215–2225, Marseille, France, June 2022. European Language Resources Association

    2022

  10. [10]

    T¨ urk¸ ce hakaret ve nefret s¨ oylemi otomatik tespit modeli.Veri Bilimi, 6(1):61–73, 2023

    Kurt M S and Demirel E Y. T¨ urk¸ ce hakaret ve nefret s¨ oylemi otomatik tespit modeli.Veri Bilimi, 6(1):61–73, 2023

    2023

  11. [11]

    Multi-domain hate speech detection using dual contrastive learning and paralinguistic features

    Dehghan S and Yanikoglu B. Multi-domain hate speech detection using dual contrastive learning and paralinguistic features. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 11745–11755, Torino, Italia, May 2024. ELRA and ICCL

    2024

  12. [12]

    Hate and offensive speech detection on arabic social media.Online Social Networks and Media, 19:100096, 2020

    Alsafari S, Sadaoui S, and Mouhoub M. Hate and offensive speech detection on arabic social media.Online Social Networks and Media, 19:100096, 2020

    2020

  13. [13]

    Arabic hate speech detection system based on arabert

    Salomon P O, Kechaou Z, and Wali A. Arabic hate speech detection system based on arabert. In2022 IEEE 21st International Conference on Cognitive Informatics and Cognitive Computing (ICCI*CC), pages 208–213, Toronto, Canada, 2022

    2022

  14. [14]

    arhatedetector: detection of hate speech from standard and dialectal arabic tweets.Discover Internet of Things, 3, 2023

    Khezzar R, Moursi A, and Al Aghbari Z. arhatedetector: detection of hate speech from standard and dialectal arabic tweets.Discover Internet of Things, 3, 2023

    2023

  15. [15]

    A survey of machine learning models and datasets for the multi-label classification of textual hate speech in english.ACM Trans

    B¨ aumler J, Bl¨ ocher L, Frey L-J, Chen X, Bayer M, and Reuter C. A survey of machine learning models and datasets for the multi-label classification of textual hate speech in english.ACM Trans. Knowl. Discov. Data, June 2026. Just Accepted

    2026

  16. [16]

    All together against hate: ensemble based llms for multi class hate speech classification in the football context, 2026

    Santos G L, Santos V G d, and Kearns C. All together against hate: ensemble based llms for multi class hate speech classification in the football context, 2026

    2026

  17. [17]

    Exploring boundaries and intensities in offensive and hate speech: Unveiling the complex spectrum of social media discourse, 2024

    Ayele A A, Jalew E A, Ali A C, Yimam S M, and Biemann C. Exploring boundaries and intensities in offensive and hate speech: Unveiling the complex spectrum of social media discourse, 2024. Preprint athttps://doi.org/ 10.48550/arXiv.2404.12042

    work page doi:10.48550/arxiv.2404.12042 2024

  18. [18]

    Exploring intensities of hate speech on social media: A case study on explaining multilingual models with XAI

    Geleta R R, Eckelt K, Parada-Cabaleiro E, and Schedl M. Exploring intensities of hate speech on social media: A case study on explaining multilingual models with XAI. InProceedings of the 4th Conference on Language, Data and Knowledge, pages 532–537, Vienna, Austria, September 2023. NOVA CLUNL, Portugal

    2023

  19. [19]

    Deep learning for hate speech intensity analysis: Distilbert classification algorithm

    Riyadi S, Masyhur A M, and Andriyani A D. Deep learning for hate speech intensity analysis: Distilbert classification algorithm. In2024 IEEE Symposium on Industrial Electronics and Applications (ISIEA), pages 1–6, 2024. 14 DEHGHAN et al. / A PREPRINT

    2024

  20. [20]

    JL-hate: An annotated dataset for joint learning of hate speech and target detection

    B¨ uy¨ ukdemirci K, Kucukkaya I E,¨Olmez E, and Toraman C. JL-hate: An annotated dataset for joint learning of hate speech and target detection. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 9543–9553, Torino, Italia, may 2024. ELRA and ICCL

    2024

  21. [21]

    Detecting hate speech in Turkish print media: A corpus and a hybrid approach with target-oriented linguistic knowledge

    G Uludo˘ gan, Y¨ uksel A E, Tun¸ cer U, I¸ sık B, Korkmaz Y, Akar D, and¨Ozg¨ ur A. Detecting hate speech in Turkish print media: A corpus and a hybrid approach with target-oriented linguistic knowledge. InProceedings of the 7th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE 2024), pages 205–214, ...

    2024

  22. [22]

    ViHOS: Hate speech spans detection for Vietnamese

    Hoang P G, Luu C D, Tran K Q, Nguyen K V, and Nguyen N L-T. ViHOS: Hate speech spans detection for Vietnamese. InProceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 652–669, Dubrovnik, Croatia, May 2023. Association for Computational Linguistics

    2023

  23. [23]

    Explainability and hate speech: Structured explanations make social media moderators faster

    Calabrese A, Neves L, Shah N, Bos M, Ross B, Lapata M, and Barbieri F. Explainability and hate speech: Structured explanations make social media moderators faster. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar, editors, Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 398–408, Bangk...

    2024

  24. [24]

    Siu2023-nst - hate speech detection contest

    Arın I, I¸ sık Z, Kutal S, Dehghan S, ¨Ozg¨ ur A, and Yanikoglu B. Siu2023-nst - hate speech detection contest. In 2023 31st Signal Processing and Communications Applications Conference (SIU), pages 1–4, 2023

    2023

  25. [25]

    Dealing with Annotator Disagreement in Hate Speech Classification

    Dehghan S, Sen M U, and Yanikoglu B. Dealing with annotator disagreement in hate speech classification, 2025. Preprint athttps://doi.org/10.48550/arXiv.2502.08266

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2502.08266 2025

  26. [26]

    Overview of the hate speech detection in Turkish and Arabic tweets (HSD-2Lang) shared task at CASE 2024

    Uludo˘ gan G, Dehghan S, Arin I, Erol E, Yanikoglu B, and¨Ozg¨ ur A. Overview of the hate speech detection in Turkish and Arabic tweets (HSD-2Lang) shared task at CASE 2024. InProceedings of the 7th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE 2024), pages 229–233, St. Julians, Malta, March 2024...

    2024

  27. [27]

    Evaluating ChatGPT’s ability to detect hate speech in Turkish tweets

    Dehghan S and Yanikoglu B. Evaluating ChatGPT’s ability to detect hate speech in Turkish tweets. InProceedings of the 7th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE 2024), pages 54–59, St. Julians, Malta, March 2024. Association for Computational Linguistics

    2024

  28. [28]

    HATECAT-TR: A hate speech span detection and categorization dataset for Turkish

    S ¸eker H K, Uludo˘ gan G,¨Onal P, and ¨Ozg¨ ur A. HATECAT-TR: A hate speech span detection and categorization dataset for Turkish. In Christos Christodoulopoulos, Tanmoy Chakraborty, Carolyn Rose, and Violet Peng, editors, Findings of the Association for Computational Linguistics: EMNLP 2025, pages 25568–25579, Suzhou, China, November 2025. Association f...

    2025

  29. [29]

    BERT: Pre-training of deep bidirectional transformers for language understanding

    Devlin J, Chang M W, Lee K, and Toutanova K. BERT: Pre-training of deep bidirectional transformers for language understanding. InProceedings 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, June 201...

    2019

  30. [30]

    Supervised contrastive learning

    Khosla P, Teterwak P, Wang C, Sarna A, Tian Y, Isola P, Maschinot A, Liu C, and Krishnan D. Supervised contrastive learning. InProceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20, Red Hook, NY, USA, 2020. Curran Associates Inc

    2020

  31. [31]

    A simple framework for contrastive learning of visual representa- tions

    Chen T, Kornblith S, Norouzi M, and Hinton G. A simple framework for contrastive learning of visual representa- tions. InProceedings of the 37 th International Conference on Machine Learning, Vienna, Austria, 2020

    2020

  32. [32]

    Zero-shot hashtag segmentation for multilingual sentiment analysis, 2021

    Rodrigues R C, Inuzuka M A, Gomes J R S, Rocha A S, Calixto I, and Nascimento H A D. Zero-shot hashtag segmentation for multilingual sentiment analysis, 2021. Preprint athttps://doi.org/10.48550/arXiv.2112.03213

    work page doi:10.48550/arxiv.2112.03213 2021

  33. [33]

    Synthetic data generation with large language models for text classification: Potential and limitations

    Li Z, Zhu H, Lu Z, and Yin M. Synthetic data generation with large language models for text classification: Potential and limitations. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10443–10461, Singapore, December 2023. Association for Computational Linguistics

    2023

  34. [34]

    Generative ai for synthetic data generation: Methods, challenges and the future, 2024

    Guo X and Chen Y. Generative ai for synthetic data generation: Methods, challenges and the future, 2024. Preprint athttps://doi.org/10.48550/arXiv.2403.04190. 15

    work page doi:10.48550/arxiv.2403.04190 2024