A Bilingual Bimodal Benchmark for Arabic-English NLP across Grammatical Correction, Essay Scoring, Morphological Tagging, and Speech Recognition

Abstract

Building comprehensive datasets that support a variety of NLP tasks and cover a diversity of languages and domains is vital for NLP evaluation purposes. In this paper, we present ZAEBUC*, a dataset that builds upon and enriches prior corpora with new annotations and benchmarking experiments. ZAEBUC* serves as a benchmark for a range of NLP tasks, including grammatical error correction, automated essay scoring, automatic speech recognition, and morphological tagging, which includes tokenization, part-of-speech tagging, and lemmatization. The dataset covers Arabic and English in both written and spoken forms, offering a bilingual and bimodal resource. Furthermore, the corpus brings together a collection of resources gathered from a similar population, enabling cross-linguistic and cross-modal comparisons. We provide benchmarking results, demonstrating the performance of NLP models, including LLMs, across various tasks, languages, and modalities.

Resources

Details