This course is an introduction to a range of language technologies. Recordings of the lectures and practical sessions have automatic transcriptions provided by YouTube or Zoom.
Contents
- Introduction to Languages
- Introduction to Language Data
- Natural Language Processing
- Automatic Speech Recognition
- Speech Synthesis
- Indigenous AI
- Sign Languages and Communicative Gesture
- Language Models and Language Generation
- Neural Machine Translation
- Conversational AI
## Introduction to Languages
A general introduction to the world's languages, including an overview of essential language and linguistics terminology that is required to know for other modules in the course.
Lecture
- Recording playlist
- Lecture slides
Practical session
## Introduction to Language Data
Aspects of data preparation for machine learning and language tech development and use. Data fundamentals such as international standards (ISO codes and Unicodes) are included.
Lecture
Practical session
## Natural Language Processing
An introduction to processing language texts, including Natural Language Processing and Understanding techniques such as tokenisation, chunking, sentiment analysis, named entity recognition and discourse analysis.
Lecture
Practical session
## Automatic Speech Recognition
The requirements for developing and using speech recognition technology, with a focus on Kaldi, ESPnet and wav2vec2 used in the Elpis ASR system. This module also includes considerations for developing user-friendly language technologies.
Lecture
Practical session
## Speech Synthesis
An overview of the major techniques used in developing and using speech synthesis and text to speech (TTS) systems, with background information relating to sound formats and compression.
Lecture
Practical session
## Indigenous AI
What is AI, and why does it need Indigenous knowledge? This module explores the value and impact of different perspectives on AI and ML technology development, with examples of object recognition in a language context.
Lecture
Practical session
## Sign Languages and Communicative Gesture
This module covers the latest developments in sign and gesture recognition and production. Further considerations for ethically co-designing appropriate technologies are also covered. The lecture and practical session recordings have Auslan interpreters.
Lecture
Practical session
## Language Models and Language Generation
Language models are the probabilities of a particular sequence of words occuring in a sentence. Language models underpin many language technologies. This module introduces the concept and techniques of building language models, and the application of these in the generation of language.
Lecture
Practical session
## Neural Machine Translation
This module describes approaches to machine translation, requirements for data preparation to build and use machine translation systems, and detail of machine translation techniques.
Lecture
Part A - Overview
Part B - Technical stuff
Prac
Further reading. This is not required reading, just for those who are instered in more detail about topics covered in the lecture, or want to know more about spoecific parts of NMT.
- Really good talk on RNNs (for general sequence processing)
- The original transformers paper, Attention Is All You Need
- Blog post about Understanding LSTM Networks
- BERT is an example of an "improvement" on the transformer for a specific task
- An overview of Attention and Augmented Recurrent Neural Networks
- More technical explanation of attentional RNNs
- BLUE score
- Back-translation: Improving Neural Machine Translation Models with Monolingual Data and Phrase-Based & Neural Unsupervised Machine Translation
## Conversational AI
In this module, the history of chat tech is described along with ethical implications of conversational language tech. In the tutorial we will build chatbots using Amazon Lex.
Lecture
Prac