Lessons from deploying NLG technology for marine weather forecast text generation

It is concluded that the success of a technology owes as much to its functional superiority as to its suitability to the various stakeholders such as developers and users.

Abstract

SUMTIME-MOUSAM is a Natural Language Generation (NLG) system that produces textual weather forecasts for offshore oilrigs from Numerical Weather Prediction (NWP) data. It has been used for the past year by Weathernews (UK) Ltd for producing 150 draft forecasts per day, which are then post-edited by forecasters before being released to end-users. In this paper, we describe how the system works, how it is used at Weathernews and finally some lessons we learnt from building, installing and maintaining SUMTIME-MOUSAM. One important lesson has been that using NLG technology improves maintainability although the biggest maintenance work actually involved changing data formats at the I/O interfaces. We also found our system being used by forecasters in unexpected ways for understanding and editing data. We conclude that the success of a technology owes as much to its functional superiority as to its suitability to the various stakeholders such as developers and users.