Machine Translation

3 Credits · Level: Graduate, Undergraduate

After more than 60 years since Machine Translation (MT) research started at Georgetown, this area of Natural Language Processing (NLP) research is more active than ever. In this course we explore the data-driven approaches to translate human language with computers that supplanted rule-based approaches in the past quarter century. First, we lay foundations for the course with statistical NLP relevant to MT and corpus preparation. Next, we start exploring statistical MT (SMT) – from word-based models to phrase-based models to tree-based models. We will then cover domain-adaptation, incremental learning and how to integrate linguistic information. We will learn how to evaluate system output with automatic and human evaluation methods. Recently, deep learning-based approaches have proven to produce superior translation quality compared to SMT. We will investigate the current state-of-the-art in neural MT (NMT) and contrast its strength and weaknesses with SMT. Machine translation does not exist in a vacuum; it is now used to provide draft translations for human translators and is embedded in other NLP systems. With better quality, raw MT is increasingly used in in written and spoken human communication. We study the adaptation of MT for the most common applications