Deep NLP in the age of RNNs, straight from Oxford

What it does

This repository holds the lecture slides, videos, and reading lists for the University of Oxford’s Hilary Term 2017 Deep Natural Language Processing course. It covers the mathematical foundations and optimization algorithms behind neural NLP models, with an applied focus on recurrent neural networks for language modeling, text classification, speech transcription, translation, and question answering. The course also addresses practical implementation concerns on CPU and GPU hardware.

The interesting bit

The syllabus is essentially a curated time capsule of the RNN era, taught by a roster of Oxford and DeepMind researchers including Phil Blunsom, Chris Dyer, and Wang Ling. Rather than dumping papers, it structures the material into three narrative arcs: sequential language modeling, conditional transduction, and advanced hybrid applications.

Key highlights

• Full lecture videos and PDF slides for every session, publicly hosted.
• Extensive reading lists linking foundational word-embedding theory to then-contemporary work on LSTMs, dropout, and scalable softmax approximations.
• Four companion practicals—word2vec, text classification, RNNs, and an open project—hosted in separate repositories.
• Lecturer lineup mixing Oxford academics with DeepMind and NVIDIA engineers.
• Content organized around three themes: sequential modeling, conditional language models, and combined advanced systems.

Verdict

A solid reference if you want the conceptual lineage of modern NLP or a structured graduate-level survey of recurrent neural networks in language tasks. Skip it if you need copy-paste code or a survey of architectures outside the RNN-centric scope.