Meteocogent: a Knowledge-based Tool for Generating Weather Forecast Texts

The production of textual forecasts by professional meteorologists is a time-constrained activity requiring the interaction of knowledge about meteorology, knowledge about language and knowledge about the conventions used to communicate the right amount of weather information to meet the needs of diverse user communities. In situations where forecasts are written manually, the time spent composing text is usually time lost for analyzing the ever-increasing amount of guidance information. The production of written draft forecasts from forecast data has long appeared to be an excellent application for knowledge-based techniques, given the high volume of text produced and the circumscribed and recurrent nature of the task. Today, however, the issue is not whether to build a rule-based system, but rather how much knowledge representation is required, how text production should be integrated into the user’s authoring process, and how flexible and user-accessible to make the knowledge components of such a system. These issues are closely related to the lifecycle costs and benefits of the software, but more importantly, to user acceptance.