Papers by Majd Hawasly
Exploring Alignment in Shared Cross-lingual Spaces (2024.acl-long)
Copied to clipboard
| Challenge: | a new study examines the degree of alignment between languages in multilingual embeddings . cross-lingual embeds are designed to encode linguistic concepts that bridge equivalent semantic meaning . a comprehensive approach is needed to address these questions. |
| Approach: | They employ clustering to uncover latent concepts within multilingual models . they introduce two metrics to quantify alignment and overlap of these concepts . |
| Outcome: | The proposed model can capture linguistic nuances across languages, but is not language-agnostic? the proposed model is able to capture nuances in multiple languages, the authors say. |
Beyond the Leaderboard: Understanding Performance Disparities in Large Language Models via Model Diffing (2025.emnlp-main)
Copied to clipboard
| Challenge: | a recent study shows that benchmarking fails to explain why models outperform others . open-weight large language models have transformed the AI landscape . |
| Approach: | They use model diffing to analyze capability differences between Gemma-2-9b-it and SimPO-enhanced variants. |
| Outcome: | The proposed model diffing approach can provide fine-grained insights beyond leaderboard metrics . it can also help to identify model performance gaps, the authors say . |
Fanar-Sadiq: A Multi-Agent Architecture for Grounded Islamic QA (2026.acl-industry)
Copied to clipboard
Ummar Abbas, Mourad Ouzzani, Mohamed Y. Eltabakh, Omar Sinan, Gagan Bhatia, Hamdy Mubarak, Majd Hawasly, Mohammed Qusay Hashim, Kareem Mohamed Darwish, Firoj Alam
| Challenge: | Large language models (LLMs) can answer religious knowledge queries fluently, but they often hallucinate and misattribute sources. |
| Approach: | They propose a bilingual Arabic-English Islamic QA system that uses a multi-agent, tool-augmented architecture to route Islamic queries to specialized modules. |
| Outcome: | The proposed system is based on a multi-agent, tool-augmented architecture and has received over 1.9M accesses in less than a year. |
LAraBench: Benchmarking Arabic AI with Large Language Models (2024.eacl-long)
Copied to clipboard
Ahmed Abdelali, Hamdy Mubarak, Shammur Chowdhury, Maram Hasanain, Basel Mousi, Sabri Boughorbel, Samir Abdaljalil, Yassine El Kheir, Daniel Izham, Fahim Dalvi, Majd Hawasly, Nizi Nazar, Youssef Elshahawy, Ahmed Ali, Nadir Durrani, Natasa Milic-Frayling, Firoj Alam
| Challenge: | Recent advances in Large Language Models (LLMs) have significantly influenced the landscape of language and speech research. |
| Approach: | They used GPT-3.5-turbo, GPT-4, BLOOMZ, Jais-13b-chat, Whisper, and USM to tackle 33 distinct tasks across 61 datasets. |
| Outcome: | The proposed model outperforms SOTA models in zero-shot learning, with a few exceptions. |
AraSafe: Benchmarking Safety in Arabic LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | AraSafe is the first large-scale native Arabic safety benchmark for large language models (LLMs) it addresses the pressing need for culturally and linguistically representative evaluation resources. |
| Approach: | They propose to use Arabic prompts to annotate harmful and non-harmful prompts into nine fine-grained safety categories to support classifiers for harmful content. |
| Outcome: | The proposed benchmarks address the need for culturally and linguistically representative evaluation resources. |
Nahw: A Comprehensive Benchmark of Arabic Grammar Understanding, Error Detection, Correction, and Explanation (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing corpora address individual linguistic aspects like spelling or diacritization, but rarely provide explanations of grammatical errors. Existing datasets and benchmarks that capture Arabic's grammatological complexity are scarce. |
| Approach: | They propose a benchmark for Arabic grammar that covers error detection, correction, and explanation. |
| Outcome: | The proposed model performs better on GPT-4o than on the best performing model (ALLaM-7B) despite fine tuning with synthetic data, the model perform better on Arabic grammar tasks. |
Self-Consistency from Only Two Samples: CoT–PoT Ensembling for Efficient LLM Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Self-consistency (SC) is a popular technique for improving the reasoning accuracy of large language models but it comes at a high computational cost due to extensive sampling. |
| Approach: | They propose a hybrid ensembling approach that leverages the complementary strengths of Chain-of-Thought and Program-of -Thus . they propose encapsulating two different modes of reasoning to create a single output and a final answer is selected as the most frequently occurring one among these outputs. |
| Outcome: | The proposed approach reduces the number of samples required for SC by 9.3x . the majority of tasks can be addressed with only two samples, which has not been possible with prior methods. |
Scaling up Discovery of Latent Concepts in Deep NLP Models (2024.eacl-long)
Copied to clipboard
| Challenge: | Existing studies on pre-trained language models focus on clustering latent spaces within pre-training models . authors propose metrics for assessing the quality of discovered latent concepts . |
| Approach: | They propose metrics to assess the quality of discovered latent concepts . they propose to scale latent concept discovery to larger datasets and models . |
| Outcome: | The proposed clustering algorithms improve performance while maintaining quality of the obtained concepts. |