Masader: Metadata Sourcing for Arabic Text and Speech Data Resources

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6340–6351 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6340 Masader: Metadata Sourcing for Arabic Text and Speech Data Resources Zaid Alyafeai1, Maraim Masoud2, Mustafa Ghaleb3, and Maged S. Al-shaibani1 1 ICS, King Fahd University of Petroleum and Minerals (KFUPM), Saudi Arabia, 2 Independent Researcher 3 Interdisciplinary Research Center for Intelligent Secure Systems (IRC-ISS), KFUPM g201080740@kfupm.edu.sa Abstract 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. Keywords: Metadata, LR Infrastructures and Architectures, Less-Resourced 1. Introduction The emergence of deep learning and its applications in many fields had a great impact on the development of various natural language processing (NLP) and speech techniques that were adapted to many languages. Many might correlate that to the availability of data especially with the existence of social media and the manufactur- ing of hardware devices that fostered research in the field, namely GPUs. Typically, we are referring to the era of deep learning which started roughly after 2010. Following that, many public Arabic NLP and speech datasets have been published in conjunction with the recent advances in deep learning (Zaghouani, 2017). Currently, there is no online centralized catalogue for Arabic NLP and speech datasets with annotated at- tributes. It is unclear how many online datasets there are as well as the metadata describing the datasets’ characteristics, such as diversity, demographic distri- bution, ethical considerations, quality, and so on. This study attempts to identify the publicly available Ara- bic NLP datasets and to provide a catalogue of Ara- bic datasets to researchers. The catalogue will increase the discoverability and provide some key metadata that will help researchers identify the most suitable dataset for their research questions. We highlight our contributions as the following: • We create the largest catalogue with 25 attributes for 200 Arabic NLP and speech datasets. • We design a metadata schema for annotating the datasets. • We analyse the current status of the Arabic NLP and speech datasets, discover issues and recom- mend solutions. The paper is structured as follows. Section 2 looks into previous work in the literature. Section 3 summarizes our approach to develop the catalogue. It discusses the research methodology, …

Content truncated to 500 words