This paper benchmarks multilingual and monolingual LLM performance across Arabic, English, and Indic languages, examining model compression effects like pruning and quantization. Multilingual models outperform language-specific counterparts, demonstrating cross-lingual transfer. Quantization maintains accuracy while promoting efficiency, but aggressive pruning compromises performance, particularly in larger models. Why it matters: The findings highlight strategies for scalable and fair multilingual NLP, addressing hallucination and generalization errors in low-resource languages.
Researchers introduce a benchmark to evaluate the factual recall and knowledge transferability of multilingual language models across 13 languages. The study reveals that language models often fail to transfer knowledge between languages, even when they possess the correct information in one language. The benchmark and evaluation framework are released to drive future research in multilingual knowledge transfer.
A new Bayesian matrix factorization approach is explored for performance prediction in multilingual NLP, aiming to reduce the experimental burden of evaluating various language combinations. The approach outperforms state-of-the-art methods in NLP benchmarks like machine translation and cross-lingual entity linking. It also avoids hyperparameter tuning and provides uncertainty estimates over predictions. Why it matters: Accurate performance prediction methods accelerate multilingual NLP research by reducing computational costs and improving experimental efficiency, especially valuable for Arabic NLP tasks.
This paper introduces GigaBERT, a customized bilingual BERT model pre-trained for Arabic NLP and English-to-Arabic zero-shot transfer learning. The study evaluates GigaBERT's performance on four information extraction tasks: named entity recognition, part-of-speech tagging, argument role labeling, and relation extraction. Results show that GigaBERT outperforms mBERT, XLM-RoBERTa, and AraBERT in both supervised and zero-shot transfer settings. Why it matters: GigaBERT advances Arabic NLP by providing a high-performing, publicly available model tailored for the complexities of the Arabic language and cross-lingual applications.
MBZUAI releases Bactrian-X, a multilingual parallel dataset of 3.4 million instruction-response pairs across 52 languages. They trained low-rank adaptation (LoRA) adapters using this dataset, creating lightweight, replaceable components for large language models. Experiments show the LoRA-based models outperform vanilla and existing instruction-tuned models in multilingual settings.
Giovanni Puccetti from ISTI-CNR presented research on linguistic probing of language models like BERT and RoBERTa. The research investigates the ability of these models to encode linguistic properties, linking this ability to outlier parameters. Preliminary work on fine-tuning LLMs in Italian and detecting synthetic news generation was also presented. Why it matters: Understanding the inner workings and linguistic capabilities of LLMs is crucial for improving their reliability and adapting them to diverse languages like Arabic.
Project LITMUS explores predicting cross-lingual transfer accuracy in multilingual language models, even without test data in target languages. The goal is to estimate model performance in low-resource languages and optimize training data for desired cross-lingual performance. This research aims to identify factors influencing cross-lingual transfer, contributing to linguistically fair MMLMs. Why it matters: Improving cross-lingual transfer is vital for creating more equitable and effective multilingual AI systems, especially for languages with limited resources.
Researchers introduce PALO, a polyglot large multimodal model with visual reasoning capabilities in 10 major languages including Arabic. A semi-automated translation approach was used to adapt the multimodal instruction dataset from English to the target languages. The models are trained across three scales (1.7B, 7B and 13B parameters) and a multilingual multimodal benchmark is proposed for evaluation.