Speaker: Phil Blunsom, lecturer at University of Oxford, Oxford, U.K
Vector space models of word meaning have achieved a great deal of success across a range of natural language processing applications. Recently researchers have begun to focus on how to compose distributed representations learnt at the word level into practical representations of larger units such as sentences and documents. In this talk I will survey recent research from the Oxford Computational Linguistics Group applying advances in Deep Learning to tasks requiring compositional models of sentence meaning. I will present results showing that generic tools such as neural language models and convolution networks can be combined to build models of question answering, dialogue processing, and machine translation.