Two of the lab's three EMNLP 2026 submissions were accepted to Findings of EMNLP 2026: QatarGuard: A Culturally-Aware Safety Benchmark for Arabic and Multilingual LLMs and ArabicDialectSafety: A Dialect-Aware Benchmark for Arabic Content Safety Classification. The papers advance culturally grounded and dialect-aware evaluation of LLM safety in Arabic, examining how safety behaviour varies across Qatari, broader Arabic, and dialectal contexts. EMNLP 2026 will be held in Budapest on 24–29 October 2026.

Three MARSAD papers were accepted to the Main Track of BESC 2026 in Bangkok: From Script to Feed: Gender Representation in Hollywood Films and Arabic Social Media Discourse; AraCosmetic: Building and Analysing an Arabic Facebook Corpus of Cosmetic Discourse; and AraHeritage: Exploring Historical and Cultural Heritage Discourse on Arabic Facebook. The studies examine gender representation, cosmetic discourse, and cultural-heritage discourse through large-scale Arabic social-media analysis. BESC 2026 will be held on 26–28 October, with proceedings published in Springer LNCS.

The new preprint When Do VLMs Help Arabic Manuscript OCR? examines when vision-language models can improve OCR for challenging Arabic documents, including historical manuscripts, aged printed texts, and handwritten material. The study finds that OCR-assisted VLMs can substantially improve recognition when the initial OCR output remains recoverable, while performance varies across scripts and document types, with particularly important challenges for Maghribi manuscripts. The work brings together researchers from UDST, QCRI, and Northwestern University in Qatar and is available as a preprint on arXiv.

The lab presented nine papers across seven ACL 2026 workshops in San Diego, covering Arabic NLP, LLM evaluation, annotation quality, AI safety, and privacy. Papers were presented at CHum 2026, LAW XX, KnowFM 2026, Privacy in NLP, GEM 2026, TrustNLP, Cross-Cultural NLP, and the Multilinguality in the Era of LLMs workshop. Topics spanned Arabic humor as a diagnostic probe for LLM reasoning, guideline-induced annotation errors, safety alignment, linguistic identity leakage in anonymized text, dialect-aware safety evaluation, and rethinking evaluation metrics in natural language generation.

The lab presented Discourse of Depression and Trauma on Arabic Social Media: A Matched-Pairs Approach to Gendered Narratives of Mental Health in the Computational Methods Division at ICA 2026 in Cape Town. Using a matched-pairs computational approach, the study examines gendered differences in how men and women discuss depression and trauma on Arabic social media while accounting for contextual factors that complicate comparisons across online populations.

The paper Building Arabic NLP from the Ground Up: Twenty Years of Lessons, Failures, and Open Problems has been selected for a Spotlight Talk at the Big Picture Workshop at ACL 2026 in San Diego. The talk will reflect on two decades of work in Arabic NLP, including key lessons learned, recurring challenges, research failures, and the open problems that continue to shape the field.

The lab's paper Beyond Annotator Disagreement: Guideline-Induced Errors in Arabic Hate Speech Annotation was accepted to the 20th Linguistic Annotation Workshop (LAW XX), co-located with ACL 2026 in San Diego. The work argues that many annotation errors in Arabic hate speech datasets originate not from annotator disagreement but from weaknesses in the annotation guidelines themselves, identifying cultural misclassification, dialectal ambiguity, and annotation projection from English-centric moderation frameworks as three error-producing mechanisms, and proposing a taxonomy of guideline-induced errors alongside a practical diagnostic framework for dataset builders.

The lab co-organized ArGuard 2026: Harmful Content Detection in Arabic Memes and LLM Prompts, a shared task hosted at ArabicNLP 2026 and co-located with EMNLP 2026 in Budapest. The task addresses Arabic-specific challenges such as dialectal variation, sarcasm, cultural references, and code-switching through two competitive tracks: multimodal hateful meme detection (Track A) and harmful prompt detection for large language models (Track B).

The lab enters the 2026 conference season with 21 accepted papers at LREC 2026, co-organization of the StanceNakba 2026 shared task, the Program Chair role at PoliticalNLP 2026, and an accepted Web Conference 2026 paper on longitudinal global climate discourse on Facebook.

The lab took part in the WISE Research & Policy Dialogue Trust Me, I'm an Algorithm? AI, Disinformation, and Higher Education, held at the ThinkBay Auditorium in Education City. Convened with Hamad Bin Khalifa University and Northwestern University in Qatar, the session presented findings from a twelve-month study on AI, disinformation, and higher education in Qatar through a panel of lead researchers and an interactive workshop facilitated by Siren Associates.

Recent 2026 outputs include the Findings of EACL paper on multi-task learning for Arabic women's discourse, the AraStress dataset paper at AbjadNLP 2026, and the IUI 2026 Companion paper Eterna on AI-powered journaling and digital legacy creation. The lab also mentored two Social Good teams on bias, fairness, and culturally responsible Arabic NLP at the Second Arabic NLP School 2026 in Rabat.

MARSAD was featured at Web Summit Qatar 2026 through the talk Inside MARSAD: AI for Understanding the Arab Digital Sphere and the session Beyond the Blank Page, highlighting Arabic AI, public communication, and AI-assisted writing. The lab also delivered a panel presentation on AI and radio at World Radio Day 2026 at the Ministry of Information Theatre in Muscat, Oman.

The lab released the MARSAD preprint on real-time social media analysis and co-authored the WISE research and policy report Fortifying Education in the Age of Disinformation, extending its impact across AI, public discourse, and education policy. The lab also participated in the ME Council Workshop on Hybrid Threats in the Information Space, co-hosted by the Middle East Council on Global Affairs and the Embassy of Poland in Qatar.

The lab co-organized major ArabicNLP 2025 shared-task initiatives at EMNLP 2025, including ImageEval, MAHED, and QIAS, advancing Arabic image captioning, multimodal hope and hate detection, and Islamic inheritance reasoning benchmarks. The lab also contributed to the WISE 12 global education summit panel on culturally aware AI, and introduced MARSAD as a teaching and research tool at an NU-Q AIM Lab lecture attended by more than 50 researchers and students.

The lab contributed multiple RANLP 2025 papers on emotion detection, hope speech, hate speech, and dialect sentiment, and co-organized the CLEF 2025 CheckThat! Lab Task 1 on subjectivity in news articles. The HopeEmo bilingual corpus was also presented at CMC-Corpora 2025 in Bayreuth. Additionally, the lab moderated a high-level AI ethics panel on language, culture, and bias in large language models at Hamad Bin Khalifa University.

The lab contributed expert commentary on AI hallucinations and mitigation strategies to The Peninsula Qatar, extending public engagement around responsible and trustworthy AI development.

Two ICWSM 2025 papers highlight the lab's work on Arabic social media analysis: ThatiAR for subjectivity detection in Arabic news sentences and a dataset-driven study of digital polarization on hijab discourse.

The lab was featured in an NU-Q Views article on the growing number of Arabic large language models in the GCC, with expert commentary on digital sovereignty, cultural grounding, and the future of Arabic AI development.

The lab presented MARSAD and Arabic AI at Web Summit Qatar 2025 through three separate talks spanning Arabic NLP, AI-mediated communication, and the NU-Q AI² R&D Initiative. The lab also contributed to policy engagement through the HBKU Workshop on the Index of Social Cohesion in Doha (January 2025), bringing social-media analytics into a national policy discussion attended by around 40–50 participants from academia, government, and international organisations.

The lab contributed the session Enhancing Online Safety and Wellbeing through AI: The Power of NLP and LLMs at the 25th International Web Information Systems Engineering Conference in Qatar.

The lab was active in major 2024 shared-task initiatives at ACL 2024 in Bangkok, including ArAIEval on propagandistic techniques detection, FIGNEWS on news media narratives, and ArabicNLU 2024, attracting over 100 registered teams across the tasks.