Our Presentations at EMNLP 2026

We were pleased to present our recent work at EMNLP 2026 and its co-located events in Budapest, Hungary. Across the main conference, Findings, ArabicNLP, and workshops, our contributions explored multilingual and multimodal reasoning, speech language models, culturally grounded evaluation, content safety, and resilience against misinformation.

EMNLP Main and Findings

• MemeLens: Multilingual Multitask VLMs for Memes introduces a unified multilingual vision-language model for understanding memes across multiple tasks, including propaganda and hate detection.
• MENASpeechBank: A Reference Voice Bank with Persona-Conditioned Multi-Turn Conversations for SpeechLLMs provides a large speech and conversation resource covering English, Modern Standard Arabic, and regional Arabic varieties.
• Multi-turn Conversational AI from Text to Multimodal Interaction surveys data, models, and evaluation for conversational systems spanning text, speech, vision, video, and tool use.

ArabicNLP 2026

• AHA-Memes: A Fine-Grained Multimodal Benchmark for Hate Detection in Arabic Memes presents a benchmark for identifying hatefulness, attack types, and targets in Arabic memes.
• ImageEval 2026: Culturally Grounded Arabic Multimodal Evaluation evaluates spoken visual question answering, hallucination detection, and culturally accurate image generation in Arabic and English.
• ArGuard Shared Task: Harmful Content Detection in Arabic Memes and LLM Prompts introduces shared evaluation tracks for harmful Arabic memes and unsafe prompts directed at LLMs.

Co-located Workshops

We also presented CritiSense at the Workshop on NLP for Positive Impact, MENASpeechBank at SALMA, Building Multimodal QA for Everyday Knowledge at ORACLE, and our multi-turn conversational AI survey at IMPACT-SPEECH. These sessions provided valuable opportunities to discuss practical, culturally inclusive, and responsible multilingual AI with the broader research community.

Thank you to our collaborators, the organizers, and everyone who joined the presentations and discussions. See the full paper summaries, authors, datasets, models, and application links.