In this project we will investigate various aspects of knowledge engineering, including how to create new knowledge, how to curate knowledge and how to integrate knowledge from multiple domains. Knowledge creation is required both for "bootstrapping" knowledge from existing (semi-) structured sources, and for keeping knowledge up to date. Integration is required in order to make cross-domain inferences, e.g., combining food and health knowledge to suggest to someone with high blood pressure who wants to eat noodles that they should chose Pho in preference to Ramen (both Pho and Ramen are noodle dishes, but the latter is salty, and high blood pressure indicates avoidance of salty food). However, knowledge resources are often incomplete and/or incorrect, problems that are exacerbated when knowledge is bootstrapped and/or integrated, so curation is also an important issue.
Given the large and ever growing size of knowledge resources it is essential that knowledge engineering tasks are largely automated, but this is extremely challenging. We propose to address this challenge by starting from our existing work on combining statistical and reasoning techniques that use existing knowledge resources as oracles in order to automate the training of classifiers to perform tasks such as semantic annotation and canonicalisation. The goal will be to develop and combine a range of such techniques in order to provide a (semi-) automated knowledge engineering system.