Dr Guy Emerson is an Academic Fellow in the Department of Computer Science and Technology at the University of Cambridge, and Executive Director of Cambridge Language Sciences. He also teaches as a College Lecturer at Gonville & Caius College. Guy was in conversation with Shrankhla Pandey, a PhD student at the Department of Computer Science and Technology.
Could you share the key questions or challenges your work seeks to address, and why these are important for the future of language science?
My research is broadly focused on semantics—understanding meaning in natural language. I’m interested in how people use and learn language. Traditionally, formal semantics has been rooted in logic, treating meaning in terms of truth: understanding a sentence means knowing whether it’s true or false in any given context. This is called “truth-conditional semantics”, and my work has explored this through machine learning, using large datasets. A key issue is the challenge of modelling language in high-dimensional spaces, which is common in modern machine learning but not traditionally considered in linguistic theory.
On the positive side, it turns out that, yes, we can train truth-conditional models on large real-world datasets, and such models provide a better fit for human behavioural data. But on the negative side, it also turns out that a lot of the things that a linguist would like to do with such a model are intractable—they are far too computationally expensive to be even remotely plausible as a cognitive model. Getting past this requires revisiting fundamental questions, both in terms of how to develop linguistic theory, and also in terms of how to develop computational models. This kind of challenging interdisciplinary work is why I appreciate being part of Cambridge Language Sciences.
What shaped your journey into language science? What influences have been pivotal in guiding you?
My academic journey has been winding, but that’s also given me a broad foundation. I studied mathematics as an undergraduate and completed a master’s in computer science, focusing on language processing. I’ve always been fascinated by languages; in school, I learned French, German, and Mandarin. After completing my PhD, I also returned to learn my mother’s native language, Hokkien. Learning languages gave me some sense of both the general principles of learning language and also the challenges of expressing yourself in different languages each with their own rules. My initial PhD proposal was actually on morphosyntactic parsing of Mandarin, but I ultimately focused on distributional semantics of English! My maths background has helped with some specific models, like energy-based models, but it’s also been great training for not being intimidated by scary-looking maths in papers. More recently, complexity theory is playing a big role in my thinking, and that has also coincidentally been helped by being asked to teach the topic in undergrad supervisions. I’ve been fortunate to have had many mentors over the years, but above all Professor Ann Copestake, who supervised both my master’s and PhD projects and has had a big impact on how I approach research.
What does your day-to-day look like?
My day-to-day work is varied. Academic life involves juggling many different things! During term time, I have teaching responsibilities, which can take over most of my time. In the last couple of years, I’ve mostly taught theoretical computer science, including courses like Algorithms, Complexity Theory, and Computation Theory. From October, I’m also starting a new position and will be lecturing the Computational Linguistics module in the Linguistics Tripos. Outside of term, I might have more time for research. And then, beyond research and teaching, I also have both College duties and Department duties, like sitting on committees and interviewing prospective students. At Cambridge Language Sciences, I support research activities and help plan and run events. Just last month, we hosted Professor Emily M. Bender, co-author of The AI Con (https://www.languagesciences.cam.ac.uk/events/ai-con-conversation-professor-emily-m-bender). Then I have external duties like reviewing, which all academics are expected to help with. Increasingly, I also engage in policy discussions, such as contributing evidence to parliamentary inquiries on AI (committees.parliament.uk/writtenevidence/124276/html/) and speaking with government officials, which is not something I would have expected a few years ago.
Do you use large language models (LLMs) in your day-to-day?
Despite my research involving machine learning and language, I don’t use LLMs in my day-to-day work. I’ve experimented with them but haven’t found a use case that doesn’t leave me frustrated. I’m still interested to hear how people are using them, and genuinely useful applications are usually far away from the hype.
When discussing LLMs, there’s a challenge that comes from how chatbots are designed to use the first person, which tricks people into thinking that there’s a mind behind the machine. That’s a psychological bias—throughout human history, only people have produced language—so it’s easy to anthropomorphize these systems and fuel the AI hype.
What broader societal or technological changes do you think your research could help shape?
My current research is quite theoretical. In the long term, it could contribute towards devising new methods for teaching and learning—whether in a classroom setting or in language rehabilitation in healthcare. By providing a deeper understanding of the learning process in terms of cognitive mechanisms, my research could help in developing curricula, or in adaptive learning to provide targeted support. I’m aware that’s quite speculative, of course!
It's plausible I might have more immediate impact from policy work. There, I would say it’s crucial to recognize that in any AI application, the bigger questions go beyond the technical. For example, when looking at machine translation in the NHS or in the court system, it’s not just about what systems exist and how well they perform. There’s a bigger question: how do we support people who speak different languages? This goes beyond technology and into the social context of use.
What is your ambition for the field of language sciences as we look toward 2050?
My personal ambition for language science in 2050 is for the field to understand the importance of high dimensionality. I know that sounds dry and technical, but I really think it’s important! A computational perspective changes how you think about theory building.
But more broadly, with the rise of LLMs and public interest in them, there is an opportunity for everyone in the language sciences to be bolder, while we have the spotlight.