Morocco

Automatic detection of toxic and offensive content in Arabic social media is a challenging task due to rich morphology, dialectal variation, and noisy writing styles. While transformer-based language models have achieved strong performance, they often produce uncertain predictions in borderline cases. This paper presents a hybrid framework for Arabic toxicity detection that combines a pretrained Arabic-specific transformer model with a confidence-aware rule-based mechanism. The proposed approach activates automatically induced lexical rules only when the model prediction falls within a predefined gray zone of uncertainty, preserving neural dominance while improving robustness and interpretability. Experiments conducted on a manually annotated dataset of 35,000 Arabic posts demonstrate that the hybrid approach achieves consistent improvements over the baseline model, particularly in reducing false negatives for toxic content. The results indicate that selective rule activation is an effective strategy for enhancing reliability in real-world Arabic social media moderation systems.

EACL 2026

A Hybrid Confidence-Aware Framework for Arabic Toxicity Detection in Social Media

toxicity nlp classification arabic

workshop paper

#### *Message from the General Chair, Aline Villavicencio*
I’m delighted and honoured to welcome you to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), taking place in the beautiful city of Rabat, in Morocco, in March 24-29, 2026. EACL is the flagship European conference of the Association and EACL 2026 proudly continues our field’s tradition of excellence in scholarship, innovation, and inclusivity. I am deeply grateful to the many volunteers whose dedication, generosity, and tireless efforts have made this conference possible.
For the first time EACL is being hosted in the African continent. This is an important milestone for our community, and we are grateful to our Moroccan hosts for enabling this historic moment by bringing this edition of EACL to Rabat. We are also delighted that the Second Arabic NLP School is co-located with EACL. We hope attendees enjoy this wonderful opportunity to strengthen ties with the Computational Linguistics communities across the African continent. *[Read full message](https://drive.google.com/file/d/14NlmHvwM6fPJuMmOvVh7K0vtQbEyv3SZ/view?usp=sharing)*<br><br>

<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://underline.io/events/522/reception','_blank')">Go to Workshops and Tutorials Program</button></html>
<br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>Access credentials will be sent to your email from Underline -  subject line "Welcome to EACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.eacl.org/registration/) first.

**Online Registration Form**: https://acl.swoogo.com/eacl2026

Registration Required

Welcome to the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL). EACL 2026 will be held in Rabat, Morocco, from March 24–29, 2026. 

Arabic diacritics encode phonetic information essential for pronunciation, disambiguation, and downstream applications, yet most Arabic ASR systems generate undiacritized output. In this work, we study direct speech-to-diacritized-text recognition using a single-stage ASR pipeline that predicts diacritics jointly with Arabic letters, without text-based post-processing. We evaluate two Arabic-adapted ASR architectures—wav2vec 2.0 XLSR-53 and Whisper-base—under a unified experimental setup on the ClArTTS Classical Arabic dataset. Performance is assessed using surface and lexical WER/CER alongside diacritic error rate (DER) to disentangle base transcription accuracy from diacritic realization. Our results show that Arabic-adapted wav2vec 2.0 achieves substantially lower diacritic error rates than Whisper, indicating stronger exploitation of acoustic cues relevant to vowelization. We further analyze the effect of decoding strategy and provide a detailed breakdown of diacritic errors, highlighting challenges associated with short vowels and morphosyntactic markers. These findings underscore the importance of model architecture and Arabic-specific adaptation for accurate diacritized Arabic ASR.

Arabic-Adapted One-Step Speech-to-Diacritized ASR: Evaluation and Error Analysis

How do Arabic-speaking communities express and engage with psychological stress on social media? We introduce AraStress, the first large-scale Arabic corpus dedicated to psychological stress research, comprising 175,862 public social media posts from 2020 to 2024, covering pandemic and post-pandemic periods.It fills a significant gap in Arabic mental-health NLP resources focused on stress, enabling large-scale analysis of related expressions.Unlike prior work focusing primarily on Twitter and depression or suicidality, AraStress addresses the critical gap in stress-focused resources. Our lexicon-based analysis reveals that stress-related posts elicit predominantly affective engagement and exhibit a hybrid lexical framing that integrates religious and therapeutic language. AraStress provides a foundational resource for culturally grounded computational models of stress detection and digital wellbeing in Arabic-speaking communities.

From Posts to Pressure: An Arabic Dataset about Stress and Mental-Health Monitoring

Stereotype repositories are critical to assess generative AI model safety, but currently lack adequate global coverage. It is imperative to prioritize targeted expansion, strategically addressing existing deficits, over merely increasing data volume. This work introduces a multilingual stereotype resource covering four sub-Saharan African countries that are severely underrepresented in NLP resources: Ghana, Kenya, Nigeria, and South Africa. By utilizing socioculturally-situated, community-engaged methods, including telephonic surveys moderated in native languages, we establish a reproducible methodology that is sensitive to the region's complex linguistic diversity and traditional orality. By deliberately balancing the sample across diverse ethnic and demographic backgrounds, we ensure broad coverage, resulting in a dataset of 3,534 stereotypes in English and 3,206 stereotypes across 15 native languages.

SAFARI: A Community-Engaged Approach and Dataset of Stereotype Resources in the Sub-Saharan African Context

Many languages are predominantly spoken rather than written, and to bring the benefits of LLMs to speakers of these languages, it is essential that models cater to the voice modality. The typical approach is to cascade ASR, LLM and TTS models together, though this results in systems with high latency, making them unsuitable for natural, real-time interaction. We describe results on taking the encoder part of a Whisper-based model trained to recognise ten languages common in Uganda, and using the Ultravox architecture to project its output directly to the input embedding space of a text model based on Qwen 3 32B, also trained to have comprehension of those languages. The result is a speech LLM with high accuracy and very low latency. For most spoken prompts, we can begin streaming a text response within as low as 50 ms, and a speech audio response within around one second, making real-time spoken interaction with an LLM possible for the first time in these languages. The model is available open source.

Real-Time Spoken Instruction Following and Translation in Ugandan Languages

African languages remain severely underrepresented in large-scale speech resources, particularly for spontaneous, naturally occurring speech that reflects real-world linguistic use. We present African Voices, a large-scale, ethically governed speech dataset covering four Nigerian languages, [~2,500] hours of audio, and 2,865 speakers, with a focus on spontaneous and scripted speech across diverse sociolinguistic contexts. Unlike existing resources that primarily rely on read or scripted speech, African Voices captures natural variation in accent, dialect, register, and code-switching, accompanied by rich demographic and contextual metadata. We describe the data collection methodology, transcription and a principled governance framework designed to support responsible use of speech data in low-resource settings. We further provide baseline automatic speech recognition results across languages. African Voices enables research on robust and fair ASR and serves as a foundational resource for advancing NLP research in African languages.

African Voices Nigeria: 2500 hours of ethically sourced speech data for four Nigerian Languages

Creating speech datasets, models, and evaluation frameworks for low-resource languages remains challenging given the lack of a broad base of pertinent experience to draw from. This paper reports on the field collection of 612 hours of spontaneous speech in Bambara, a low-resource West African language; the semi-automated annotation of that dataset with transcriptions; the creation of several monolingual ultra-compact and small models using the dataset; and the automatic and human evaluation of their output. We offer practical suggestions for data collection protocols, annotation, and model design, as well as evidence for the importance of performing human evaluation. In addition to the main dataset, multiple evaluation datasets, models, and code are made publicly available.

Dealing with the Hard Facts of Low-Resource African NLP

We present SALT-31 benchmark dataset, for evaluation of machine translation (MT) models and covering 31 Ugandan languages. Unlike sentence-level evaluation sets, SALT-31 is constructed from short, scenario-driven mini-dialogues designed to preserve discourse context, pragmatics, and culturally grounded communication patterns common in everyday Ugandan settings. The dataset contains 100 English
sentences organized into 20 typical communication scenarios, each represented as a five sentence mini-sequence. It can therefore be used to evaluate both sentence-level and paragraph level machine translation, and includes nearly every language spoken in a country with high linguistic diversity.

SALT-31: A Machine Translation Benchmark Dataset for 31 Ugandan Languages

This paper introduces the first standardized benchmark for evaluating Automatic Speech Recognition (ASR) in the Bambara language, utilizing one hour of professionally recorded Malian constitutional text. Designed as a controlled reference set under near-optimal acoustic and linguistic conditions, the benchmark was used to evaluate 37 models, ranging from Bambara-trained systems to large-scale commercial models. Our findings reveal that current ASR performance remains significantly below deployment standards; the top-performing system in terms of Word Error Rate (WER) achieved 46.76\% and the best Character Error Rate (CER) of 13.00\% was set by another model, while several prominent multilingual models exceeded 100\% WER due to severe hallucinations. These results suggest that multilingual pre-training and model scaling alone are insufficient for underrepresented languages. Furthermore, because this dataset represents a best-case scenario of the most simplified and formal form of spoken Bambara, these figures likely establish an upper bound for performance in practical, real-world settings. We provide the benchmark and an accompanying public leaderboard to facilitate transparent evaluation and future research in Bambara speech technology.

Where Are We at with Automatic Speech Recognition for the Bambara Language?

Multimodal AI research has overwhelmingly focused on high-resource languages, hindering the democratization of advancements in the field. To address this, we present AfriCaption, a comprehensive framework for multilingual image captioning in 20 African languages and our contributions are threefold: (i) a curated dataset built on Flickr8k, featuring semantically aligned captions generated via a context-aware selection and translation process; (ii) a dynamic, context-preserving pipeline that ensures ongoing quality through model ensembling and adaptive substitution; and (iii) the AfriCaption model, a 0.5B parameter
vision-to-text architecture that integrates SigLIP and NLLB200 for caption generation across underrepresented languages. This unified framework ensures ongoing data quality and establishes the first scalable image-captioning resource for underrepresented African languages, laying the groundwork for truly inclusive multimodal AI.

AfriCaption: Establishing a New Paradigm for Image Captioning in African Languages

Existing AI bias evaluation benchmarks largely reflect Western perspectives, leaving African contexts underrepresented and enabling harmful stereotypes in applications across various domains. To address this gap, we introduce AfriStereo, the first open-source African stereotype dataset and evaluation framework grounded in local socio-cultural contexts. Through community engaged efforts across Senegal, Kenya, and Nigeria, we collected 1,163 stereotypes spanning gender, ethnicity, religion, age, and profession. Using few-shot prompting with human-in-the-loop validation, we augmented the dataset to over 5,000 stereotype–antistereotype pairs. Entries were validated through semantic clustering and manual annotation by culturally informed reviewers. Preliminary evaluation of language models reveals that nine of eleven models exhibit statistically significant bias, with Bias Preference Ratios (BPR) ranging from 0.63 to 0.78 (p ≤ 0.05), indicating systematic preferences for stereotypes over antistereotypes, particularly across age, profession, and gender dimensions. Domain-specific models appeared to show weaker bias in our setup, suggesting task-specific training may mitigate some associations. Looking ahead, AfriStereo opens pathways for future research on culturally grounded bias evaluation and mitigation, offering key methodologies for the AI community on building more equitable, context-aware, and globally inclusive NLP technologies.

Downloads

Next from EACL 2026

Arabic-Adapted One-Step Speech-to-Diacritized ASR: Evaluation and Error Analysis

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from EACL 2026

Arabic-Adapted One-Step Speech-to-Diacritized ASR: Evaluation and Error Analysis

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads