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.
Outcome:
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.
