Triangle Seminars

Abstract:
Mathematical models of natural language can be organised into logical and
statistical. The former are based on grammatical structures of
phrases and sentences and the latter on distributions of words in corpora
of text. In joint work with Clark (Cambridge) and Coecke (Oxford), we developed a unifying
framework where the distributions of words are composed to form
distributions for phrases and sentences. This expanded the application
domains of the statistical models – e.g. automatic reasoning about
similarity – from words to phrases and sentences.

On the theoretical side, our model extends the word-based setting from vectors to tensors.
Tensors are main players in the mathematical models of quantum mechanics.
In this talk, I will review the theory and applications of our
model in simple terms and through examples. I will briefly explain how
`entanglement', a concept arising from tensors in quantum mechanics,
manifests itself and is used as a resource in the linguistic applications.
I will also explain how the reasoning toolkit used in this model is the same as that of Abramsky and Coecke in their
categorical Quantum Mechanics model.