Abstract :
Natural language generation (NLG) is the process of generating text or speech from some originally non linguistic source of data. It has been used, for instance, to allow expert systems to explain their reasoning and to produce radio weather reports from numerical data. A range of technologies are available for the implementation of NLG, from simple “mail merge” products to more general approaches that reason about the content to be included and the grammatical forms to be selected. In general, an appropriate technology has to be selected for a given task according to how stereotyped the desired form of speech/text is. NLG in general is hard because it involves making choices between different linguistic possibilities and we only have partial knowledge of how to justify such choices. Its main disadvantage at present is the need to have an appropriate source of data as its input. On the other hand, it promises a number of advantages compared with human authoring of texts
