Anders Søgaard: The Uphill Battles of Doing NLP without Psychology and Theoretical Machine Learning

Big parts of NLP are psychologically naïve: standard approaches to punctuation, word representation, attention, to name a few examples. Our interpretations of our results are often a bit naïve, also, from a machine learning perspective. This talk is an argument for more psychology and more machine learning in NLP. Understanding the psychology of language is key to robust NLP that works; not only today, but also tomorrow. Understanding when and why our models work, or why they don’t, is impossible without deep knowledge of machine learning. But as Martin Luther King  said: Let us not wallow in the valley of despair. I have a dream – a dream that one day we will be able to bridge the gap between theory and practice, establishing formal results about generalization and learnability, on the pillars of cognitive psychology.

Slides (pdf)

Anders Søgaard is a full professor in Computer Science at the University of Copenhagen. He has been awarded an ERC Starting Grant and a Google Focused Research Award, as well as several best paper awards.