MLMI 8 Neural Machine Translation and Dialogue Systems

Aims

This half-module provides an introduction to machine translation and task-oriented dialogue systems as problems that can be addressed by machine learning.   The presentation will employ sequence-to-sequence models to develop a uniform approach to these problems.

Objectives:

On completion of this model, students should have a working familiarity with:

• translation and dialogue as problems in natural language processing;
• data sets used in creating dialogue systems and machine translation systems;
• automatic and manual assessment of dialogue and translation quality;
• the statistical approach to task oriented dialogue systems and its component tasks;
• modelling approaches for neural machine translation;
• sequence-to-sequence models, such as the Transformer architecture and instances such as GPT2
• fine tuning and domain adaptation procedures;
• current research problems, including search and model correctness
• data biases and ethical concerns in translation and dialogue.