Key points are not available for this paper at this time.
I present an architecture for data-to-text systems, that is NLG systems which produce texts from non-linguistic input data; this essentially extends the architecture of Reiter and Dale (2000) to systems whose input is raw data instead of AI knowledge bases. This architecture is being used in the BabyTalk project, and is based on experiences in several projects at Aberdeen; it also seems to be compatible with many data-to-text systems developed elsewhere. It consists of four stages which are organised in a pipeline: Signal Analysis, Data Interpretation, Document Planning, and Microplanning and Realisation.
Ehud Reiter (Mon,) studied this question.