Natural Language Processing research projects
Document Machine Translation as Deep Structured Prediction
(Google Faculty Award, 2018-2019)
Project lead: Dr. Reza Haffari
Deep learning with neural networks is revolutionising many domains of Artificial Intelligence, including machine translation (MT). Despite their flexibility, most of neural MT models translate sentences independently, and hence ignore the document context. Discourse phenomenon, such as pronominal anaphora, lexical coherence and consistency are important aspects of a document which are neglected in sentence-based translation. We aim to develop effective neural architectures for translating documents, taking into account the global documental context in both the source and target languages.
Google AI for Social Impact: The Use of AI to Code Ambulance Data for Suicide Prevention
(Google Grant, 2019-2022)
Project lead: Prof. Wray Buntine
Ambulances are typically the first point of contact with someone who is suicidal. Ambulance clinical records are an important and rich data source, containing details of the nature and background to the attendance, the location of the event, and the clinical outcome. For more than 20 years and in partnership with Ambulance Victoria, Turning Point has been providing a Victorian alcohol, illicit and pharmaceutical drug surveillance system using coded paramedic clinical data. This world-first surveillance system has recently been expanded to include reporting of national data in partnership with jurisdictional ambulance services, and there is an opportunity to apply the same methodology to suicidal behaviour in these datasets. However, the time and resources needed to code additional suicide-related attendances is prohibitive without significant ongoing funding. We propose using AI to allow us to also code national ambulance data that will establish a cost-effective model that could be adopted globally.