Building the Oranian-English Parallel Corpus: Methodology and Compilation Process

Résumé

The scarcity of linguistic resources poses a major challenge for automated translation and processing of dialects. These resources are crucial for natural language processing experts conducting research on dialect recognition, processing, and machine translation. This paper describes the compilation of a dataset for an Algerian low-resource language as it emphasizes the importance of developing resources for Algerian dialects. It examines existing relevant corpora and details the creation process and unique features of the pioneering Oranian-English Parallel Corpus (OEPC). OEPC is the first parallel corpus built from scratch that pairs an Algerian dialect with its English counterparts. The paper outlines the criteria and steps involved in compiling a monolingual corpus for the Oranian dialect (ORN), including data sources and formats. ORN comprises 8500 sentences, which were then translated into English to form OEPC. This valuable linguistic resource is a product of the ERAD project, an initiative aimed at providing NLP professionals with diverse Algerian mono-, multi-, and cross-dialectal corpora. The paper also explains the data compilation and augmentation techniques used to expand the project's outputs.