Exploring natural language processing on quantum computers

The consortium, funded by the Royal Academy of Engineering for a Senior Research Fellowship at UCL, will build on a long-term exploration of quantum mechanics and linguistics by Quantinuum's chief scientist Professor Bob Coecke, Head of Artificial Intelligence Professor Stephen Clark, and Professor Mehrnoosh Sadrzadeh of UCL Computer Science.

The BBC hopes to find new ways to represent content in forms readable by computers, to support tasks such as content discovery and archival retrieval. This builds on the Corporation's previous work with Sadrzadeh on Enhancing Personalised Recommendations with the use of Multi Modal Information.

Ilyas Khan, Founder of Cambridge Quantum Computing and CEO of Quantinuum said: "Developing quantum computing so that the broadest and most diverse populations can benefit, means looking across the timing spectrum at applications that can be made productive in the short, medium and long term. As part of our long-term work, we anticipate that true language processing will become important with fault tolerant quantum processors, and our work with the BBC and UCL is a very significant step towards being prepared to take advantage of quantum computers when they become available at scale. Quantinuum is a leader in the fields in which it operates, and this leadership is built on deeply meaningful collaborations such as this."

In their 15-year collaboration, the researchers established a unified model of statistical and compositional meaning for natural language, in the seminal 2011 paper Mathematical Foundations of a Compositional Distributional Model of Meaning.  The foundational work was guided by Professor Coecke's categorical quantum mechanics formalism. Experimental evidence followed suite by Professor Sadrzadeh's work on Concrete Models and Experimental Evaluations for the Categorical Compositional Distributional Model of Meaning. The advance of these techniques beyond academic research, to a scaled industrial level, will take capabilities from mere sentence level to general text, using methods which were initiated in the papers The Mathematics of Text Structure and Evaluating Composition Models for Verb Phrase Elliptical Sentence Embeddings.

The broadcaster's archives reflect a century of global news and cultural life across the UK and beyond. It is one of the largest broadcast archives in the world, with over 15 million items, including audio, film, and text documents, as well as toys, games, merchandise, artefacts, and historic equipment.