Natural Language has been and will remain as the most preferred way to store and transfer knowledge. More than 80% of electronic data in modern societies are generated and stored in textual format. How to process unstructured text to extract useful insights and support actionable decision making and discover the hidden treasure of collective intelligence is of enormous value. In this unit, we start with traditional text processing techniques using Regular Expressions and discuss the needs of text processing and normalisation. We then introduce fundamental pipelines of natural language processing (NLP), including part-of-speech tagging and various ways of sentence parsing, with the aim of introducing traditional text feature collection techniques for higher-level tasks such as sentiment or document classification. Building on the understanding of the pros and cons of feature-based NLP pipeline approaches, the unit moves onto the modern approach of deep learning for NLP, focusing on word vector representation, neural language models, and recurrent neural networks for NLP. The unit situates the techniques around major NLP tasks, including information extraction, sentiment detection, dialogue systems and machine translation.

Students are able to (1) apply pre-processing techniques for textual data preparation; (2) build pipelines for core NLP tasks; (3) critically analyse different language models; (4) explain how vector representations of words can be obtained; (5) evaluate performance of NLP solutions, both traditional and neural; and (6) undertake core components of major NLP tasks.