This module covers advanced level topics in natural language processing, with a focus on deep learning-based approaches. These include text classification, synthetic parsing, part of speech tagging, named-entity recognition, coreference resolution, and machine translation.

The module will focus on deep learning-based approaches to NLP. These include text classification, synthetic parsing, part of speech tagging, named-entity recognition, coreference resolution, and machine translation. 1. Us of Distributed Word Representations (word embeddings) through the exploration of tools and methods such as: word2vec, The Skip-Gram Model, The Continuous Bag of Words (CBOW) Model, Global Vectors for Word Representation (GloVe). 2. Application of Dependency Parsing such as: Syntactic Structure (consistency and dependency), Dependency Grammar and Treebanks, Transition-based dependency parsing, Neural dependency parsing. 3. Types and structutrse of Recurrent Neural Networks (RNNs) including: Language Modelling with RNNs, application of RNNs (POS and NER tagging), Long Short-Term Memory RNNs (LSTMs), Gated recurrent units (GRUs), Bidirectional and multi-layer RNNs. 4. Sequence-to-sequence learning (Seq2Seq) methods and approaches such as: statistical machine translation, neural machine translation, Sequence-to-sequence with attention, attention variants etc. 5. Contextual Word Representations and Pretraining for downstream processing sich as: ELMo, ULMfit, Transformer Architectures, GPTs, BERT etc 6. Algorithmic Bias and Disinformation: Social and Ethical Considerations in NLP Systems.

On successful completion of this module, students will be able to: 1. Use pretrained word embeddings such as word2vec and GloVe in downstream NLP tasks such as text classification. 2. Analyze the syntactic structure of sentences using state-of-the-art dependency parsers. 3. Use RNNs and their variants such as LSTMs and Gated Recurrent Units (GRUs) for part of speech tagging and named entity recognition. 4. Use Seq2Seq models for translation tasks. 5. Make use of large pretrained transformer-based models such as BERT and its variants (ALBERT, RoBERTa, DistilBERT, SciBERT) in downstream tasks.

On successful completion of this module, students will be able to: 1. Recognize and be mindful of ethical issues in modern NLP systems such as algorithmic bias (gender, race) and disinformation (fake news).

On successful completion of this module, students will be able to: N/A

The module will be delivered using fully online using on-line lectures, labs and tutorials.