TimeML is a rich specification language for event and temporal expressions in natural language text, developed within the AQUAINT program on Question Answering Systems. Unlike previous work, TimeML systematically anchors event predicates to temporally denoting expressions, orders event expressions relative to one another, and allows for delayed interpretation of partially determined temporal expressions. It addresses four key problems in event-temporal identification: (a) time stamping of events, (b) ordering events, (c) reasoning with underspecified temporal expressions, and (d) reasoning about event persistence. TimeML includes data structures such as EVENT, TIMEX3, SIGNAL, and LINK, which help represent events, temporal expressions, and their relationships. The EVENT tag captures event types like occurrence, state, and aspectual, while TIMEX3 marks explicit temporal expressions like dates, durations, and times. SIGNAL indicates how temporal objects are related, and LINK establishes relationships between events and times. TimeML also supports causation by distinguishing between event-caused events, entity-caused events, and discourse marker events. The paper concludes that TimeML provides a robust framework for temporal and event annotation, with future developments including integration with DAML-TIME and the creation of a gold standard corpus, TIMEBANK.TimeML is a rich specification language for event and temporal expressions in natural language text, developed within the AQUAINT program on Question Answering Systems. Unlike previous work, TimeML systematically anchors event predicates to temporally denoting expressions, orders event expressions relative to one another, and allows for delayed interpretation of partially determined temporal expressions. It addresses four key problems in event-temporal identification: (a) time stamping of events, (b) ordering events, (c) reasoning with underspecified temporal expressions, and (d) reasoning about event persistence. TimeML includes data structures such as EVENT, TIMEX3, SIGNAL, and LINK, which help represent events, temporal expressions, and their relationships. The EVENT tag captures event types like occurrence, state, and aspectual, while TIMEX3 marks explicit temporal expressions like dates, durations, and times. SIGNAL indicates how temporal objects are related, and LINK establishes relationships between events and times. TimeML also supports causation by distinguishing between event-caused events, entity-caused events, and discourse marker events. The paper concludes that TimeML provides a robust framework for temporal and event annotation, with future developments including integration with DAML-TIME and the creation of a gold standard corpus, TIMEBANK.