Natural Language Generation (NLG) systems generate texts in English and other human languages from non-linguistic input data. Usually there are a large number of possible texts that can communicate the input data, and NLG systems must choose one of these. We argue that style can be used by NLG systems to choose between possible texts, and explore how this can be done by (1) explicit stylistic parameters, (2) imitating a genre style, and (3) imitating an individual's style.
展开▼
