Compositionality in vector space models of meaning

Hosted by the Wellcome Neural Dynamics PhD Programme

We interact with computers every day, and often using something like human language. There is therefore a huge amount of research going into how to represent human language computationally, which has been very successful. However, the most successful large language models are opaque and moreover can make simple errors. On the other hand, theoretical modelling of language – formal semantics – gives a clear account of how to compose words. I will give an overview of the model I work with that combines word embeddings with formal semantics. I will describe how this model can deal with difficult phenomena such as metaphor, and extensions into the visual domain. I will also talk about some new research directions that look at how to implement these ideas in a spiking neural network.

If joining online: https://bristol-ac-uk.zoom.us/j/94138286231?pwd=MlRURE1SWjR6OTZCR1Fnak9QbGxhUT09Meeting ID: 941 3828 6231, Passcode: 277162

Contact information

Contact Luke Burguete with any enquiries.