This paper introduces TimeML, a specification language for event and temporal expressions in natural language text, developed within the AQUAINT program on Question Answering Systems. Unlike previous work, TimeML addresses three key phenomena: (1) anchoring event predicates to temporally denoting expressions, (2) ordering event expressions both intrasententially and in discourse, and (3) allowing for delayed interpretation of underspecified temporal expressions. The language includes four major data structures: EVENT, TIMEX3, SIGNAL, and LINK, which handle various aspects of event and temporal markup. TimeML extends TIMEX2 attributes, introduces temporal functions, identifies signals determining temporal expression interpretation, and captures all classes of event expressions. The LINK tag is a major innovation, encoding various relations between temporal elements and establishing event orderings. The paper also discusses event causation in TimeML, distinguishing between different types of causal relations. Finally, it outlines future developments, including integration with the DAML language and the creation of a gold standard corpus, TIMEBANK, for testing and evaluation.This paper introduces TimeML, a specification language for event and temporal expressions in natural language text, developed within the AQUAINT program on Question Answering Systems. Unlike previous work, TimeML addresses three key phenomena: (1) anchoring event predicates to temporally denoting expressions, (2) ordering event expressions both intrasententially and in discourse, and (3) allowing for delayed interpretation of underspecified temporal expressions. The language includes four major data structures: EVENT, TIMEX3, SIGNAL, and LINK, which handle various aspects of event and temporal markup. TimeML extends TIMEX2 attributes, introduces temporal functions, identifies signals determining temporal expression interpretation, and captures all classes of event expressions. The LINK tag is a major innovation, encoding various relations between temporal elements and establishing event orderings. The paper also discusses event causation in TimeML, distinguishing between different types of causal relations. Finally, it outlines future developments, including integration with the DAML language and the creation of a gold standard corpus, TIMEBANK, for testing and evaluation.