Much of what we learn about the world comes not from our own experience but from others, often with language. Using generic statements (or generics, e.g., "Dogs have fur", "Trucks are cool", "Elephants live in Africa and Asia"), one can flexibly communicate abstract generalizations to another, and this kind of language is ubiquitous in everyday conversation, formal teaching, and child-directed speech. The normal tools for formalizing meaning, however, don't easily apply to generics, and so there are no formal models for how an agent should update their beliefs from a generic statement. I explore a hypothesis that the informational content of a generic sentence is equivalent to presenting the listener with a single positive observation (an exemplar of the category with the property) and that inferences which are stronger than those implied by just a single observation result from pragmatic (social) reasoning. I present evidence for the simple pragmatic account (in comparison to a number of alternatives) based on learning from (i) generics about novel categories (e.g., "Feps eat berries"; "Dorbs get cancer") and (ii) generics about conjunctive properties (e.g., "Elephants live in Africa and Asia"). Finally, I'll show how fine-grained syntactic expectations guide the incremental understanding of conjunctive generics, suggesting that information seamlessly passes through levels of linguistic analysis (syntax, semantics, and pragmatics) to guide language understanding. The empirical validation for this computational account of generic language opens the door to understanding more precisely the benefits of learning from language.