The focus of language model evaluation has transitioned towards reasoning and knowledge-intensive tasks, driven by advancements in pretraining large models. While state-of-the-art models are partially trained on large Arabic texts, evaluating their performance in Arabic remains challenging due to the limited availability of relevant datasets. To bridge this gap, we present ArabicMMLU, the first multi-task language understanding benchmark for the Arabic language, sourced from school exams across diverse educational levels in different countries spanning North Africa, the Levant, and the Gulf regions. Our data comprises 40 tasks and 14,575 multiple-choice questions in Modern Standard Arabic (MSA) and is carefully constructed by collaborating with native speakers in the region. Our comprehensive evaluations of 35 models reveal substantial room for improvement, particularly among the best open-source models. Notably, BLOOMZ, mT0, LLama2, and Falcon struggle to achieve a score of 50%, while even the top-performing Arabic-centric model only achieves a score of 62.3%.

Instruction tuning has emerged as a prominent methodology for teaching Large Language Models (LLMs) to follow instructions. However, current instruction datasets predominantly cater to English or are derived from English-dominated LLMs, leading to inherent biases toward Western culture. This bias negatively impacts non-English languages such as Arabic and the unique culture of the Arab region. This paper addresses this limitation by introducing CIDAR, the first open Arabic instruction-tuning dataset culturally aligned by native Arabic speakers. CIDAR contains 10,000 instruction and output pairs that represent the Arab region. We discuss the cultural relevance of CIDAR via the analysis and comparison to a few models fine-tuned on other datasets. Our experiments indicate that models fine-tuned on CIDAR achieve better cultural alignment compared to those fine-tuned on 30x more data.

Datasets are foundational to many breakthroughs in modern artificial intelligence. Many recent achievements in the space of natural language processing (NLP) can be attributed to the fine-tuning of pre-trained models on a diverse set of tasks that enables a large language model (LLM) to respond to instructions. Instruction fine-tuning (IFT) requires specifically constructed and annotated datasets. However, existing datasets are almost all in the English language. In this work, our primary goal is to bridge the language gap by building a human-curated instruction-following dataset spanning 65 languages. We worked with fluent speakers of languages from around the world to collect natural instances of instructions and completions. Furthermore, we create the most extensive multilingual collection to date, comprising 513 million instances through templating and augmenting existing datasets across 114 languages. In total, we contribute three key resources: we develop and open-source the Aya Dataset, the Aya Collection, and the Aya Evaluation Suite. The Aya initiative also serves as a valuable case study in participatory research, involving collaborators from 119 countries. We see this as an important framework for future research collaborations that aim to bridge gaps in resources.

Recently, there has been a growing interest in analyzing user-generated text to understand opinions expressed on social media. In NLP, this task is known as stance detection, where the goal is to predict whether the writer is in favor, against, or has no opinion on a given topic. Stance detection is crucial for applications such as sentiment analysis, opinion mining, and social media monitoring, as it helps in capturing the nuanced perspectives of users on various subjects. As part of the ArabicNLP 2024 program, we organized the first shared task on Arabic Stance Detection, StanceEval 2024. This initiative aimed to foster advancements in stance detection for the Arabic language, a relatively underrepresented area in Arabic NLP research. This overview paper provides a detailed description of the shared task, covering the dataset, the methodologies used by various teams, and a summary of the results from all participants. We received 28 unique team registrations, and during the testing phase, 16 teams submitted valid entries. The highest classification F-score obtained was 84.38.

Multitask prompted finetuning (MTF) has been shown to help large language models generalize to new tasks in a zero-shot setting, but so far explorations of MTF have focused on English data and models. We apply MTF to the pretrained multilingual BLOOM and mT5 model families to produce finetuned variants called BLOOMZ and mT0. We find finetuning large multilingual language models on English tasks with English prompts allows for task genrealization to non-English languages that appear only in the pretraining corpus. Finetuning on multilingual tasks with English prompts further improves performance on English and non-English tasks leading to various state-of-the-art zero-shot results. We also investigate finetuning on multilingual tasks with prompts that have been machine-translated from English to match the language of each dataset. We find training on these machine-translated prompts leads to better performance on human-written prompts in the respective languages. Surprisingly, we find models are capable of zero-shot generalization to tasks in languages they have never intentionally seen. We conjecture that the models are learning higher-level capabilities that are both task- and language-agnostic. In addition, we introduce xP3, a composite of supervised datasets in 46 languages with English and machine-translated prompts. Our code, datasets and models are freely available at https://github.com/bigscience-workshop/xmtf.

Numerous languages exhibit shared characteristics, especially in morphological features. For instance, Arabic and Russian both belong to the fusional language category. The question arises: Do such common traits influence language comprehension across diverse linguistic backgrounds? This study explores the possibility of transferring comprehension skills across languages to Arabic in a zero-shot scenario. Specifically, we demonstrate that training language models on other languages can enhance comprehension of Arabic, as evidenced by our evaluations in three key tasks: natural language inference, question answering, and named entity recognition. Our experiments reveal that certain morphologically rich languages (MRLs), such as Russian, display similarities to Arabic when assessed in a zero-shot context, particularly in tasks like question answering and natural language inference. However, this similarity is less pronounced in tasks like named entity recognition.

PromptSource is a system for creating, sharing, and using natural language prompts. Prompts are functions that map an example from a dataset to a natural language input and target output. Using prompts to train and query language models is an emerging area in NLP that requires new tools that let users develop and refine these prompts collaboratively. PromptSource addresses the emergent challenges in this new setting with (1) a templating language for defining data-linked prompts, (2) an interface that lets users quickly iterate on prompt development by observing outputs of their prompts on many examples, and (3) a community-driven set of guidelines for contributing new prompts to a common pool. Over 2,000 prompts for roughly 170 datasets are already available in PromptSource. PromptSource is available at https://github.com/bigscience-workshop/promptsource.

The NLP pipeline has evolved dramatically in the last few years. The first step in the pipeline is to find suitable annotated datasets to evaluate the tasks we are trying to solve. Unfortunately, most of the published datasets lack metadata annotations that describe their attributes. Not to mention, the absence of a public catalogue that indexes all the publicly available datasets related to specific regions or languages. When we consider low-resource dialectical languages, for example, this issue becomes more prominent. In this paper, we create Masader, the largest public catalogue for Arabic NLP datasets, which consists of 200 datasets annotated with 25 attributes. Furthermore, we develop a metadata annotation strategy that could be extended to other languages. We also make remarks and highlight some issues about the current status of Arabic NLP datasets and suggest recommendations to address them.

Natural language modelling has gained a lot of interest recently. The current state-of-the-art results are achieved by first training a very large language model and then fine-tuning it on multiple tasks. However, there is little work on smaller more compact language models for resource-limited devices or applications. Not to mention, how to efficiently train such models for a low-resource language like Arabic. In this paper, we investigate how such models can be trained in a compact way for Arabic. We also show how distillation and quantization can be applied to create even smaller models. Our experiments show that our largest model which is 2x smaller than the baseline can achieve better results on multiple tasks with 2x less data for pretraining.

Automating natural language understanding is a lifelong quest addressed for decades. With the help of advances in machine learning and particularly, deep learning, we are able to produce state of the art models that can imitate human interactions with languages. Unfortunately, these advances are controlled by the availability of language resources. Arabic advances in this field , although it has a great potential, are still limited. This is apparent in both research and development. In this paper, we showcase some NLP models we trained for Arabic. We also present our methodology and pipeline to build such models from data collection, data preprocessing, tokenization and model deployment. These tools help in the advancement of the field and provide a systematic approach for extending NLP tools to many languages.