Dr. Denis Paperno
Dr. Denis Paperno
Assistant Professor
My research focusses on meaning and modern AI models. How do we make sure that AI captures non-trivial aspects of meaning in natural languages such as English, French, or Dutch? How do we properly test whether AI can reason with language?
It is sometimes argued that AI tends to show only parrot-like behavior, giving answers that are close to the data it is trained on. How can we verify if AI generalizes well with language, approximating understanding rare or novel words, novel coherent text, and subtle meaning contrasts? What kind of pretraining is necessary for AI to better reflect human reasoning, language understanding, and learning in those areas? Over the years, I’ve been developing evaluation datasets and procedures involving diverse language phenomena.