One of the most important aspects of natural-language processing for many practical applications is establishing a relationship between language and facts about the world -- a task referred to as "semantics". The work in semantics that has been motivated by work in information extraction has been quite different, and much more closely tied to the needs of specific applications, than similar work in linguistics or computational linguistics generally. In this tutorial I will examine what is involved in developing a semantic theory for information extraction, using recent work in the ACE ("Automated Content Extraction") program as an example. The ACE program has placed a great deal of emphasis on the development of a semantic theory of information extraction that can be extended to broad domains of application, and covers a wide range of linguistic and pragmatic phenomena, and therefore provides a good foundation for discussion and understanding of the relvant issues. Topics that will be covered include the ontology of entities, relations, and events, as well as coreference resolution, metonymy resolution, and distinguishing generic from specific references.