KAIROS aimed at the automatic composition and generation of weather forecast (henceforth WF) drafts for two languages, English and Greek, on the basis of numerical meteorological data. The project's results provide Natural Language Processing researchers with the necessary Natural Language Generation (henceforth NLG) know-how and infrastructure. On the other hand, meteorologists are able to post-edit automatically -generated WFs, thus making productive use of their time for meteorological data and weather phenomena analysis.

The final products of the project consist of:

• a prototype for the automatic generation of WF drafts
• a set of linguistic resources, directed to the meteorology and climatology sublanguages for both English and Greek
• a methodology for the automatic composition of different types of WFs, based on the state-of-the-art architecture by Reiter [Reiter 1997, 2000]

The project's innovation lies both in its pioneering role in the greek NLP community, as well as in its system architecture. The following system features should be noted:

• a modular architecture design allows immediate use and integration of existing linguistic resources, as well as better control and management during the development process of system subunits
• the same NLG mechanism may be used for the processing of different types of weather data (METAR and TAF meteorological telegrams), as well as processing in different languages (English and Greek)
• use of XML structures in all processing stages of the system
• expert users have direct access to linguistic resources. By modification, for example, of entries in the terminological lexica, a more "natural" text style can be produced by the system
• the use of clause and phrase templates allows flexible and easy integration of the linguistic structures that are present in the WF corpus used for the study of the meteorological and climatological sublanguages