Tom Brand: ‘It’s the combination of analysis and language that makes linguistics so enjoyable’
Alumni image: Nehir Aksel
An elective course on a Peruvian language led Tom Brand to linguistics. He now puts his knowledge to use at TNO, where he works on reliable AI applications.
The Peruvian language Quechua set Tom Brand on the path to studying linguistics. ‘In the final year of my dual Bachelor’s degree in Mathematics and Physics, a classmate of mine took an extracurricular course on it,’ he explains. ‘His mother was from Peru and he wanted to connect more with his roots. I don’t have any particular connection to Peru myself, but I thought it would be interesting to learn about the grammar of a language so different from Western languages.’
That introduction developed into years of study. ‘I enjoyed the combination of analysis and language in that elective so much that I went on to do the whole bachelor’s degree and then the master’s,’ he recalls. ‘I also did a Master’s in Physics, but I found that I eventually grew tired of it. At secondary school, as a real nerd, I used to watch lots of YouTube videos about physics in my spare time. When I started studying physics, that urge disappeared, whereas the linguistics degree had the opposite effect; it only served to fuel my interest in language and linguistics outside the course itself.’
Training with synthetic data
A career related to his degree in linguistics was therefore an obvious choice. ‘At first, I looked into whether I could do a PhD,’ says Brand. ‘I’d written my master’s thesis on how well ChatGPT can detect, correct and explain grammatical errors. That left me wanting more, but the field of AI suddenly became so huge that there were sometimes hundreds of applicants for a single position. One of my thesis supervisors, Stephan Raaijmakers, then pointed me towards this role at TNO.’
TNO is the largest independent research institute in the Netherlands. It acts as a bridge between academia and industry, converting scientific knowledge into practical innovations. Brand’s work there focuses on hybrid AI. ‘We combine AI with data-driven or knowledge-driven methods, so that we can use it reliably and prevent it from making things up on its own. In one of my projects, for example, we’re creating synthetic data for customer service teams. At the moment, customer service staff have to fill in another list of metadata after every call. It’s a real chore for them. Many telecoms companies would like to use AI for this, but those lists contain all sorts of personal data. You don’t want to train AI on that, which is why we’re creating artificial data.’
Brand works on a different project every few months. ‘For example, I also worked on a project to reduce bias in the labour market. We often want to shift employers’ focus from qualifications to hard and soft skills. This required a list of those skills, but if you describe them in too complex a way, part of the labour market still won’t find it useful. I therefore worked on rule-based AI that supports experts in their writing, so that their texts don’t become unnecessarily complex.
Programmer and student assistant
A large part of Brand’s work involves programming. Brand: ‘I took as many programming modules as possible in both linguistics and physics. I really enjoyed the fact that with programming you get immediate feedback: something either works or it doesn’t. I can also highly recommend it to students who have even a passing interest in it, because you gain skills that are very useful in the job market. Becoming a student assistant is also a good idea. I’ve helped out with maths courses a few times. You really learn well how to convey complex information. That’s particularly useful in the technical field.’