Abstract
My talk will start by giving an overview of kinds of generalization historically and currently studied in the field of natural language processing (NLP). I will discuss underlying motivations and working assumptions of NLP, including what we want to build, how we evaluate what we build, and what kinds of generalization is seen as reasonable to expect from our systems (and how all of this has changed over time). Then, I will shift the discussion towards generalization in *natural* language processing, i.e., how language itself demands and is driven by generalization from its users. Finally, I will propose some implications on this in the development of language technologies.