Department of Computer Science, Tulane University,
New Orleans, USA
Bonnie J Dorr
Institute for Human and Machine Cognition (IHMC),
Department of Cybersecurity, St. Bonaventure University,
St. Bonaventure, USA
Department of Telecommunications and Systems Engineering, Universitat Autonoma de Barcelona,
Cerdanyola del Valles, Spain
School of Engineering, Computer, and Mathematical Sciences, Auckland University of Technology,
Auckland, New Zealand
Department of Computer Engineering, Vali-e-Asr University of Rafsanjan,
Department of Linguistics and Translation Studies, Vali-e-Asr University of Rafsanjan,
The biggest issue with low-resource languages is the extreme difficulty of obtaining enough resources. Machine Translation (MT) has proven successful for several language pairs. However, each language comes with its own challenges. Low-resource languages have largely been left out of the MT revolution. In low-resource languages there are often very few written texts and of those that exist, they do not have a parallel text in another language. MT has made significant progress in recent years with a shift to statistical and neural models and rapid development of new architectures such as the transformer. However, current models trained on little parallel data tend to produce poor quality translations and without the parallel texts, statistical or neural MT will give subpar results. This challenge is exacerbated in the context of social media, where we need to enable communication for languages with no corresponding parallel corpora or unofficial languages. We are pleased to invite the academic community to respond to this issue on low-resource MT.
Research topic should be relevant to low-resource MT, including, but not limited to: Unsupervised statistical or neural MT for low-resource language pairs. Semi-supervised statistical or neural MT for low-resource language pairs. Pretraining methods leveraging monolingual data. Multilingual statistical or neural MT for low-resource languages.