Professor Mehrnoosh Sadrzadeh, University College London

Logical models of natural language analyse linguistic structure such as grammar and discourse. It has been argued that these structures are based on the mechanisms underlying human language processes. Statistical models take account of co-occurrence frequencies and have done really well in text prediction, leading to large language models such as GPT and BERT. I work on unifying the two: the logical and the statistical, in order to develop better models of language processing for both humans and machines. Some of the models my colleagues and I developed use tensors, a natural inhabitant of quantum mechanics, thus some of the constructions of our models can be learnt even by quantum computers.