David Marr has famously put forward that any information processing system should be analysed from three distinct, complementary levels of analysis. In this talk we will look at Combinatory Categorial Grammar (CCG) on all these levels. On the most abstract, computational level, I will show that CCG has the right computational power to explain observed linguistic universals regarding the order of core NP and VP elements across world’s languages. On the algorithmic level, I will show that the new CCG parsing algorithms can be very incremental without making too many speculations about the future like other parsers do. This is done by solving the ``right-adjunction problem”. Finally, on the implementational level I will show the results that provide strong indication that the algorithmic processing of a CCG parser correlates with the human brain activity during language processing.
Miloš Stanojević is a Senior Research Scientist in DeepMind. Prior to that he did a PostDoc at University of Edinburgh with Mark Steedman where he worked on Combinatory Categorial Grammars (CCG), and collaborated with Ed Stabler on Minimalist Grammars. He has received a PhD degree from University of Amsterdam for the work on machine translation. His main research interests lie in incremental sentence processing, structured inference, formal grammars, and generative linguistics.