Neural Machine Translation: A Survey of Methods used for Low Resource Languages

Machine translation (MT) is a maj or natural language processing subfield that is designed to automatically translate human spoken languages. Neural machine translation (NMT) has gained significant achievement in recent times and is becoming the emerging dominant approach in actual MT systems. The availability of significant quantities of parallel data is critical for the performance of Neural Machine Translation (NMT) models. A key challenge in machine translation for many language pairs is a dearth of parallel corpora. An important research concern in the field of machine translation is enhancing NMT accuracy for languages with limited resources. Extensive research is being done to enhance Neural Machine Translation accuracy by leveraging a limited parallel dataset. In this paper, we try to provide a broad and well-structured review of the methods used to enhance the translation accuracy of low-resource languages by exploiting monolingual data, which is typically easy to get and is available in a considerably higher amount, and further classification in two broad categories according to the adjustments they required to the NMT model's architecture is done. We believe that our survey will be useful to new students starting in this field.