Artificial intelligence and neural machine translation

This chapter examines the linguistic, social, and educational implications of integrating deep learning into Machine Translation (MT) through neural network technology. Neural Machine Translation (NMT) is widely regarded as a valuable tool for society and institutions. However, it raises important questions about human involvement in machine learning, particularly with regard to the supervision of NMT systems and the evaluation of translation quality. The recent proliferation of Large Language Models (LLMs), such as GPT-4, which rely heavily on English-centric training data, further complicates these issues by potentially reinforcing language homogenisation and socio-cultural bias. This chapter explores how the prevalence of such training data in deep learning and the lack of human supervision in NMT training could affect linguistic diversity and perpetuate bias. It first outlines current deep learning algorithms in Natural Language Processing (NLP) and the importance of human intervention to mitigate errors and bias. The chapter then addresses the need for more inclusive language corpora to ensure representation of low-resource languages. Finally, it highlights the critical role of end-user awareness in the evaluation of NMT applications, especially in contexts such as language learning and academic communication, where the impact of AI on human cognition is significant.