• Macquarie Home »
• Faculty of Science »
• Department of Computing

Using semantic categories to improve syntactic parsing

This project involves extracting semantic categories in an unsupervised or semi-supervised manner from a very large corpus, and using them as features in a syntactic parser. The idea is that this would improve the parser's ability to model the properties of low frequency words not seen in the parser's training data.

Parsing in parallel using GPUs

Because of their extensive parallelism, the Graphics Processing Units (GPUs) of modern computers are computationally more powerful than CPUs. This project involves writing GPU code to parse one or more sentences in parallel on GPUs, with the goal of obtaining a significant improvement in parsing speed.

Using text data-mining to predict stock-market movements

This project involves using text data mining from e.g., newspaper stories, and company reports to predict stock price movements of the stocks that the reports describe. This will involve named entity recognition and machine learning.

Computational models of human word learning

This project involves studying how human children learn to identify words in running speech. It builds on existing work we've already done in this area to study the utility of factors such as prosody, intonation and external reference in word learning. It would involve collaborating with psycholinguists and linguists here at MQ doing experimental work on human language acquisition.

Using more complex features in dependency parsing

This project involves using sampling techniques to exploit more complex longer-range features that cannot be easily incorporated into standard dynamic-programming approaches to parsing. The intended outcome would be a significantly more accurate model of dependency parsing that uses linguistically realistic features.